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/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/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::_dcbrt:
249 case vmIntrinsics::_dabs:
250 case vmIntrinsics::_fabs:
251 case vmIntrinsics::_iabs:
252 case vmIntrinsics::_labs:
253 case vmIntrinsics::_datan2:
254 case vmIntrinsics::_dsqrt:
255 case vmIntrinsics::_dsqrt_strict:
256 case vmIntrinsics::_dexp:
257 case vmIntrinsics::_dlog:
258 case vmIntrinsics::_dlog10:
259 case vmIntrinsics::_dpow:
260 case vmIntrinsics::_dcopySign:
261 case vmIntrinsics::_fcopySign:
262 case vmIntrinsics::_dsignum:
263 case vmIntrinsics::_roundF:
264 case vmIntrinsics::_roundD:
265 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
266
267 case vmIntrinsics::_notify:
268 case vmIntrinsics::_notifyAll:
269 return inline_notify(intrinsic_id());
270
271 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
272 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
273 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
274 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
275 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
276 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
277 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
278 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
279 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
280 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
281 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
282 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
283 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
284 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
285
286 case vmIntrinsics::_arraycopy: return inline_arraycopy();
287
288 case vmIntrinsics::_arraySort: return inline_array_sort();
289 case vmIntrinsics::_arrayPartition: return inline_array_partition();
290
291 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
292 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
293 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
294 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
295
296 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
297 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
298 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
299 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
303 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
304
305 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
306
307 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
308
309 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
310 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
311 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
312 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
313
314 case vmIntrinsics::_compressStringC:
315 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
316 case vmIntrinsics::_inflateStringC:
317 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
318
319 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
320 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
321 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
322 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
323 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
324 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
325 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
326 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
327 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
328
329 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
330 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
331 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
332 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
333 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
334 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
335 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
336 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
337 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
338
339 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
340 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
341 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
342 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
343 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
344 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
345 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
346 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
347 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
348
349 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
350 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
351 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
352 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
353 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
354 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
355 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
356 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
357 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
358
359 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
360 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
361 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
362 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
363
364 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
365 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
366 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
367 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
368
369 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
370 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
371 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
372 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
373 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
374 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
375 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
376 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
377 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
378
379 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
380 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
381 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
382 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
383 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
384 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
385 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
386 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
387 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
388
389 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
390 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
391 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
392 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
393 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
394 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
395 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
396 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
397 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
398
399 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
400 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
401 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
402 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
403 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
404 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
405 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
406 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
407 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
408
409 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
414
415 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
416 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
417 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
418 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
419 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
420 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
421 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
422 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
423 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
435
436 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
437 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
438 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
439 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
451
452 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
455 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
456
457 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
462
463 case vmIntrinsics::_loadFence:
464 case vmIntrinsics::_storeFence:
465 case vmIntrinsics::_storeStoreFence:
466 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
467
468 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
469
470 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
471 case vmIntrinsics::_currentThread: return inline_native_currentThread();
472 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
473
474 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
475 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
476
477 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
478 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
479
480 #if INCLUDE_JVMTI
481 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
482 "notifyJvmtiStart", true, false);
483 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
484 "notifyJvmtiEnd", false, true);
485 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
486 "notifyJvmtiMount", false, false);
487 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
488 "notifyJvmtiUnmount", false, false);
489 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
490 #endif
491
492 #ifdef JFR_HAVE_INTRINSICS
493 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
494 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
495 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
496 #endif
497 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
498 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
499 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
500 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
501 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
502 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
503 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
504 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
505 case vmIntrinsics::_getLength: return inline_native_getLength();
506 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
507 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
508 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
509 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
510 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
511 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
512 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
513
514 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
515 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
516
517 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
518
519 case vmIntrinsics::_isInstance:
520 case vmIntrinsics::_isHidden:
521 case vmIntrinsics::_getSuperclass:
522 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
523
524 case vmIntrinsics::_floatToRawIntBits:
525 case vmIntrinsics::_floatToIntBits:
526 case vmIntrinsics::_intBitsToFloat:
527 case vmIntrinsics::_doubleToRawLongBits:
528 case vmIntrinsics::_doubleToLongBits:
529 case vmIntrinsics::_longBitsToDouble:
530 case vmIntrinsics::_floatToFloat16:
531 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
532 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
533 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
534 case vmIntrinsics::_floatIsFinite:
535 case vmIntrinsics::_floatIsInfinite:
536 case vmIntrinsics::_doubleIsFinite:
537 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
538
539 case vmIntrinsics::_numberOfLeadingZeros_i:
540 case vmIntrinsics::_numberOfLeadingZeros_l:
541 case vmIntrinsics::_numberOfTrailingZeros_i:
542 case vmIntrinsics::_numberOfTrailingZeros_l:
543 case vmIntrinsics::_bitCount_i:
544 case vmIntrinsics::_bitCount_l:
545 case vmIntrinsics::_reverse_i:
546 case vmIntrinsics::_reverse_l:
547 case vmIntrinsics::_reverseBytes_i:
548 case vmIntrinsics::_reverseBytes_l:
549 case vmIntrinsics::_reverseBytes_s:
550 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
551
552 case vmIntrinsics::_compress_i:
553 case vmIntrinsics::_compress_l:
554 case vmIntrinsics::_expand_i:
555 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
556
557 case vmIntrinsics::_compareUnsigned_i:
558 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
559
560 case vmIntrinsics::_divideUnsigned_i:
561 case vmIntrinsics::_divideUnsigned_l:
562 case vmIntrinsics::_remainderUnsigned_i:
563 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
564
565 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
566
567 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
568 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
569 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
570 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
571 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
572
573 case vmIntrinsics::_Class_cast: return inline_Class_cast();
574
575 case vmIntrinsics::_aescrypt_encryptBlock:
576 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
577
578 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
579 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
580 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
581
582 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
583 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
584 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
585
586 case vmIntrinsics::_counterMode_AESCrypt:
587 return inline_counterMode_AESCrypt(intrinsic_id());
588
589 case vmIntrinsics::_galoisCounterMode_AESCrypt:
590 return inline_galoisCounterMode_AESCrypt();
591
592 case vmIntrinsics::_md5_implCompress:
593 case vmIntrinsics::_sha_implCompress:
594 case vmIntrinsics::_sha2_implCompress:
595 case vmIntrinsics::_sha5_implCompress:
596 case vmIntrinsics::_sha3_implCompress:
597 return inline_digestBase_implCompress(intrinsic_id());
598 case vmIntrinsics::_double_keccak:
599 return inline_double_keccak();
600
601 case vmIntrinsics::_digestBase_implCompressMB:
602 return inline_digestBase_implCompressMB(predicate);
603
604 case vmIntrinsics::_multiplyToLen:
605 return inline_multiplyToLen();
606
607 case vmIntrinsics::_squareToLen:
608 return inline_squareToLen();
609
610 case vmIntrinsics::_mulAdd:
611 return inline_mulAdd();
612
613 case vmIntrinsics::_montgomeryMultiply:
614 return inline_montgomeryMultiply();
615 case vmIntrinsics::_montgomerySquare:
616 return inline_montgomerySquare();
617
618 case vmIntrinsics::_bigIntegerRightShiftWorker:
619 return inline_bigIntegerShift(true);
620 case vmIntrinsics::_bigIntegerLeftShiftWorker:
621 return inline_bigIntegerShift(false);
622
623 case vmIntrinsics::_vectorizedMismatch:
624 return inline_vectorizedMismatch();
625
626 case vmIntrinsics::_ghash_processBlocks:
627 return inline_ghash_processBlocks();
628 case vmIntrinsics::_chacha20Block:
629 return inline_chacha20Block();
630 case vmIntrinsics::_kyberNtt:
631 return inline_kyberNtt();
632 case vmIntrinsics::_kyberInverseNtt:
633 return inline_kyberInverseNtt();
634 case vmIntrinsics::_kyberNttMult:
635 return inline_kyberNttMult();
636 case vmIntrinsics::_kyberAddPoly_2:
637 return inline_kyberAddPoly_2();
638 case vmIntrinsics::_kyberAddPoly_3:
639 return inline_kyberAddPoly_3();
640 case vmIntrinsics::_kyber12To16:
641 return inline_kyber12To16();
642 case vmIntrinsics::_kyberBarrettReduce:
643 return inline_kyberBarrettReduce();
644 case vmIntrinsics::_dilithiumAlmostNtt:
645 return inline_dilithiumAlmostNtt();
646 case vmIntrinsics::_dilithiumAlmostInverseNtt:
647 return inline_dilithiumAlmostInverseNtt();
648 case vmIntrinsics::_dilithiumNttMult:
649 return inline_dilithiumNttMult();
650 case vmIntrinsics::_dilithiumMontMulByConstant:
651 return inline_dilithiumMontMulByConstant();
652 case vmIntrinsics::_dilithiumDecomposePoly:
653 return inline_dilithiumDecomposePoly();
654 case vmIntrinsics::_base64_encodeBlock:
655 return inline_base64_encodeBlock();
656 case vmIntrinsics::_base64_decodeBlock:
657 return inline_base64_decodeBlock();
658 case vmIntrinsics::_poly1305_processBlocks:
659 return inline_poly1305_processBlocks();
660 case vmIntrinsics::_intpoly_montgomeryMult_P256:
661 return inline_intpoly_montgomeryMult_P256();
662 case vmIntrinsics::_intpoly_assign:
663 return inline_intpoly_assign();
664 case vmIntrinsics::_encodeISOArray:
665 case vmIntrinsics::_encodeByteISOArray:
666 return inline_encodeISOArray(false);
667 case vmIntrinsics::_encodeAsciiArray:
668 return inline_encodeISOArray(true);
669
670 case vmIntrinsics::_updateCRC32:
671 return inline_updateCRC32();
672 case vmIntrinsics::_updateBytesCRC32:
673 return inline_updateBytesCRC32();
674 case vmIntrinsics::_updateByteBufferCRC32:
675 return inline_updateByteBufferCRC32();
676
677 case vmIntrinsics::_updateBytesCRC32C:
678 return inline_updateBytesCRC32C();
679 case vmIntrinsics::_updateDirectByteBufferCRC32C:
680 return inline_updateDirectByteBufferCRC32C();
681
682 case vmIntrinsics::_updateBytesAdler32:
683 return inline_updateBytesAdler32();
684 case vmIntrinsics::_updateByteBufferAdler32:
685 return inline_updateByteBufferAdler32();
686
687 case vmIntrinsics::_profileBoolean:
688 return inline_profileBoolean();
689 case vmIntrinsics::_isCompileConstant:
690 return inline_isCompileConstant();
691
692 case vmIntrinsics::_countPositives:
693 return inline_countPositives();
694
695 case vmIntrinsics::_fmaD:
696 case vmIntrinsics::_fmaF:
697 return inline_fma(intrinsic_id());
698
699 case vmIntrinsics::_isDigit:
700 case vmIntrinsics::_isLowerCase:
701 case vmIntrinsics::_isUpperCase:
702 case vmIntrinsics::_isWhitespace:
703 return inline_character_compare(intrinsic_id());
704
705 case vmIntrinsics::_min:
706 case vmIntrinsics::_max:
707 case vmIntrinsics::_min_strict:
708 case vmIntrinsics::_max_strict:
709 case vmIntrinsics::_minL:
710 case vmIntrinsics::_maxL:
711 case vmIntrinsics::_minF:
712 case vmIntrinsics::_maxF:
713 case vmIntrinsics::_minD:
714 case vmIntrinsics::_maxD:
715 case vmIntrinsics::_minF_strict:
716 case vmIntrinsics::_maxF_strict:
717 case vmIntrinsics::_minD_strict:
718 case vmIntrinsics::_maxD_strict:
719 return inline_min_max(intrinsic_id());
720
721 case vmIntrinsics::_VectorUnaryOp:
722 return inline_vector_nary_operation(1);
723 case vmIntrinsics::_VectorBinaryOp:
724 return inline_vector_nary_operation(2);
725 case vmIntrinsics::_VectorUnaryLibOp:
726 return inline_vector_call(1);
727 case vmIntrinsics::_VectorBinaryLibOp:
728 return inline_vector_call(2);
729 case vmIntrinsics::_VectorTernaryOp:
730 return inline_vector_nary_operation(3);
731 case vmIntrinsics::_VectorFromBitsCoerced:
732 return inline_vector_frombits_coerced();
733 case vmIntrinsics::_VectorMaskOp:
734 return inline_vector_mask_operation();
735 case vmIntrinsics::_VectorLoadOp:
736 return inline_vector_mem_operation(/*is_store=*/false);
737 case vmIntrinsics::_VectorLoadMaskedOp:
738 return inline_vector_mem_masked_operation(/*is_store*/false);
739 case vmIntrinsics::_VectorStoreOp:
740 return inline_vector_mem_operation(/*is_store=*/true);
741 case vmIntrinsics::_VectorStoreMaskedOp:
742 return inline_vector_mem_masked_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorGatherOp:
744 return inline_vector_gather_scatter(/*is_scatter*/ false);
745 case vmIntrinsics::_VectorScatterOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ true);
747 case vmIntrinsics::_VectorReductionCoerced:
748 return inline_vector_reduction();
749 case vmIntrinsics::_VectorTest:
750 return inline_vector_test();
751 case vmIntrinsics::_VectorBlend:
752 return inline_vector_blend();
753 case vmIntrinsics::_VectorRearrange:
754 return inline_vector_rearrange();
755 case vmIntrinsics::_VectorSelectFrom:
756 return inline_vector_select_from();
757 case vmIntrinsics::_VectorCompare:
758 return inline_vector_compare();
759 case vmIntrinsics::_VectorBroadcastInt:
760 return inline_vector_broadcast_int();
761 case vmIntrinsics::_VectorConvert:
762 return inline_vector_convert();
763 case vmIntrinsics::_VectorInsert:
764 return inline_vector_insert();
765 case vmIntrinsics::_VectorExtract:
766 return inline_vector_extract();
767 case vmIntrinsics::_VectorCompressExpand:
768 return inline_vector_compress_expand();
769 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
770 return inline_vector_select_from_two_vectors();
771 case vmIntrinsics::_IndexVector:
772 return inline_index_vector();
773 case vmIntrinsics::_IndexPartiallyInUpperRange:
774 return inline_index_partially_in_upper_range();
775
776 case vmIntrinsics::_getObjectSize:
777 return inline_getObjectSize();
778
779 case vmIntrinsics::_blackhole:
780 return inline_blackhole();
781
782 default:
783 // If you get here, it may be that someone has added a new intrinsic
784 // to the list in vmIntrinsics.hpp without implementing it here.
785 #ifndef PRODUCT
786 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
787 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
788 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
789 }
790 #endif
791 return false;
792 }
793 }
794
795 Node* LibraryCallKit::try_to_predicate(int predicate) {
796 if (!jvms()->has_method()) {
797 // Root JVMState has a null method.
798 assert(map()->memory()->Opcode() == Op_Parm, "");
799 // Insert the memory aliasing node
800 set_all_memory(reset_memory());
801 }
802 assert(merged_memory(), "");
803
804 switch (intrinsic_id()) {
805 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
806 return inline_cipherBlockChaining_AESCrypt_predicate(false);
807 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(true);
809 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
810 return inline_electronicCodeBook_AESCrypt_predicate(false);
811 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(true);
813 case vmIntrinsics::_counterMode_AESCrypt:
814 return inline_counterMode_AESCrypt_predicate();
815 case vmIntrinsics::_digestBase_implCompressMB:
816 return inline_digestBase_implCompressMB_predicate(predicate);
817 case vmIntrinsics::_galoisCounterMode_AESCrypt:
818 return inline_galoisCounterMode_AESCrypt_predicate();
819
820 default:
821 // If you get here, it may be that someone has added a new intrinsic
822 // to the list in vmIntrinsics.hpp without implementing it here.
823 #ifndef PRODUCT
824 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
825 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
826 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
827 }
828 #endif
829 Node* slow_ctl = control();
830 set_control(top()); // No fast path intrinsic
831 return slow_ctl;
832 }
833 }
834
835 //------------------------------set_result-------------------------------
836 // Helper function for finishing intrinsics.
837 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
838 record_for_igvn(region);
839 set_control(_gvn.transform(region));
840 set_result( _gvn.transform(value));
841 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
842 }
843
844 //------------------------------generate_guard---------------------------
845 // Helper function for generating guarded fast-slow graph structures.
846 // The given 'test', if true, guards a slow path. If the test fails
847 // then a fast path can be taken. (We generally hope it fails.)
848 // In all cases, GraphKit::control() is updated to the fast path.
849 // The returned value represents the control for the slow path.
850 // The return value is never 'top'; it is either a valid control
851 // or null if it is obvious that the slow path can never be taken.
852 // Also, if region and the slow control are not null, the slow edge
853 // is appended to the region.
854 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
855 if (stopped()) {
856 // Already short circuited.
857 return nullptr;
858 }
859
860 // Build an if node and its projections.
861 // If test is true we take the slow path, which we assume is uncommon.
862 if (_gvn.type(test) == TypeInt::ZERO) {
863 // The slow branch is never taken. No need to build this guard.
864 return nullptr;
865 }
866
867 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
868
869 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
870 if (if_slow == top()) {
871 // The slow branch is never taken. No need to build this guard.
872 return nullptr;
873 }
874
875 if (region != nullptr)
876 region->add_req(if_slow);
877
878 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
879 set_control(if_fast);
880
881 return if_slow;
882 }
883
884 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
885 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
886 }
887 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
888 return generate_guard(test, region, PROB_FAIR);
889 }
890
891 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
892 Node* *pos_index) {
893 if (stopped())
894 return nullptr; // already stopped
895 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
896 return nullptr; // index is already adequately typed
897 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
898 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
899 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
900 if (is_neg != nullptr && pos_index != nullptr) {
901 // Emulate effect of Parse::adjust_map_after_if.
902 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
903 (*pos_index) = _gvn.transform(ccast);
904 }
905 return is_neg;
906 }
907
908 // Make sure that 'position' is a valid limit index, in [0..length].
909 // There are two equivalent plans for checking this:
910 // A. (offset + copyLength) unsigned<= arrayLength
911 // B. offset <= (arrayLength - copyLength)
912 // We require that all of the values above, except for the sum and
913 // difference, are already known to be non-negative.
914 // Plan A is robust in the face of overflow, if offset and copyLength
915 // are both hugely positive.
916 //
917 // Plan B is less direct and intuitive, but it does not overflow at
918 // all, since the difference of two non-negatives is always
919 // representable. Whenever Java methods must perform the equivalent
920 // check they generally use Plan B instead of Plan A.
921 // For the moment we use Plan A.
922 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
923 Node* subseq_length,
924 Node* array_length,
925 RegionNode* region) {
926 if (stopped())
927 return nullptr; // already stopped
928 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
929 if (zero_offset && subseq_length->eqv_uncast(array_length))
930 return nullptr; // common case of whole-array copy
931 Node* last = subseq_length;
932 if (!zero_offset) // last += offset
933 last = _gvn.transform(new AddINode(last, offset));
934 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
935 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
936 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
937 return is_over;
938 }
939
940 // Emit range checks for the given String.value byte array
941 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
942 if (stopped()) {
943 return; // already stopped
944 }
945 RegionNode* bailout = new RegionNode(1);
946 record_for_igvn(bailout);
947 if (char_count) {
948 // Convert char count to byte count
949 count = _gvn.transform(new LShiftINode(count, intcon(1)));
950 }
951
952 // Offset and count must not be negative
953 generate_negative_guard(offset, bailout);
954 generate_negative_guard(count, bailout);
955 // Offset + count must not exceed length of array
956 generate_limit_guard(offset, count, load_array_length(array), bailout);
957
958 if (bailout->req() > 1) {
959 PreserveJVMState pjvms(this);
960 set_control(_gvn.transform(bailout));
961 uncommon_trap(Deoptimization::Reason_intrinsic,
962 Deoptimization::Action_maybe_recompile);
963 }
964 }
965
966 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
967 bool is_immutable) {
968 ciKlass* thread_klass = env()->Thread_klass();
969 const Type* thread_type
970 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
971
972 Node* thread = _gvn.transform(new ThreadLocalNode());
973 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
974 tls_output = thread;
975
976 Node* thread_obj_handle
977 = (is_immutable
978 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
979 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
980 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
981 thread_obj_handle = _gvn.transform(thread_obj_handle);
982
983 DecoratorSet decorators = IN_NATIVE;
984 if (is_immutable) {
985 decorators |= C2_IMMUTABLE_MEMORY;
986 }
987 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
988 }
989
990 //--------------------------generate_current_thread--------------------
991 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
992 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
993 /*is_immutable*/false);
994 }
995
996 //--------------------------generate_virtual_thread--------------------
997 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
998 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
999 !C->method()->changes_current_thread());
1000 }
1001
1002 //------------------------------make_string_method_node------------------------
1003 // Helper method for String intrinsic functions. This version is called with
1004 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1005 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1006 // containing the lengths of str1 and str2.
1007 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1008 Node* result = nullptr;
1009 switch (opcode) {
1010 case Op_StrIndexOf:
1011 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1012 str1_start, cnt1, str2_start, cnt2, ae);
1013 break;
1014 case Op_StrComp:
1015 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1016 str1_start, cnt1, str2_start, cnt2, ae);
1017 break;
1018 case Op_StrEquals:
1019 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1020 // Use the constant length if there is one because optimized match rule may exist.
1021 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1022 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1023 break;
1024 default:
1025 ShouldNotReachHere();
1026 return nullptr;
1027 }
1028
1029 // All these intrinsics have checks.
1030 C->set_has_split_ifs(true); // Has chance for split-if optimization
1031 clear_upper_avx();
1032
1033 return _gvn.transform(result);
1034 }
1035
1036 //------------------------------inline_string_compareTo------------------------
1037 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1038 Node* arg1 = argument(0);
1039 Node* arg2 = argument(1);
1040
1041 arg1 = must_be_not_null(arg1, true);
1042 arg2 = must_be_not_null(arg2, true);
1043
1044 // Get start addr and length of first argument
1045 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1046 Node* arg1_cnt = load_array_length(arg1);
1047
1048 // Get start addr and length of second argument
1049 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1050 Node* arg2_cnt = load_array_length(arg2);
1051
1052 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1053 set_result(result);
1054 return true;
1055 }
1056
1057 //------------------------------inline_string_equals------------------------
1058 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1059 Node* arg1 = argument(0);
1060 Node* arg2 = argument(1);
1061
1062 // paths (plus control) merge
1063 RegionNode* region = new RegionNode(3);
1064 Node* phi = new PhiNode(region, TypeInt::BOOL);
1065
1066 if (!stopped()) {
1067
1068 arg1 = must_be_not_null(arg1, true);
1069 arg2 = must_be_not_null(arg2, true);
1070
1071 // Get start addr and length of first argument
1072 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1073 Node* arg1_cnt = load_array_length(arg1);
1074
1075 // Get start addr and length of second argument
1076 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1077 Node* arg2_cnt = load_array_length(arg2);
1078
1079 // Check for arg1_cnt != arg2_cnt
1080 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1081 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1082 Node* if_ne = generate_slow_guard(bol, nullptr);
1083 if (if_ne != nullptr) {
1084 phi->init_req(2, intcon(0));
1085 region->init_req(2, if_ne);
1086 }
1087
1088 // Check for count == 0 is done by assembler code for StrEquals.
1089
1090 if (!stopped()) {
1091 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1092 phi->init_req(1, equals);
1093 region->init_req(1, control());
1094 }
1095 }
1096
1097 // post merge
1098 set_control(_gvn.transform(region));
1099 record_for_igvn(region);
1100
1101 set_result(_gvn.transform(phi));
1102 return true;
1103 }
1104
1105 //------------------------------inline_array_equals----------------------------
1106 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1107 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1108 Node* arg1 = argument(0);
1109 Node* arg2 = argument(1);
1110
1111 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1112 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1113 clear_upper_avx();
1114
1115 return true;
1116 }
1117
1118
1119 //------------------------------inline_countPositives------------------------------
1120 bool LibraryCallKit::inline_countPositives() {
1121 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1122 return false;
1123 }
1124
1125 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1126 // no receiver since it is static method
1127 Node* ba = argument(0);
1128 Node* offset = argument(1);
1129 Node* len = argument(2);
1130
1131 ba = must_be_not_null(ba, true);
1132
1133 // Range checks
1134 generate_string_range_check(ba, offset, len, false);
1135 if (stopped()) {
1136 return true;
1137 }
1138 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1139 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1140 set_result(_gvn.transform(result));
1141 clear_upper_avx();
1142 return true;
1143 }
1144
1145 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1146 Node* index = argument(0);
1147 Node* length = bt == T_INT ? argument(1) : argument(2);
1148 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1149 return false;
1150 }
1151
1152 // check that length is positive
1153 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1154 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1155
1156 {
1157 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1158 uncommon_trap(Deoptimization::Reason_intrinsic,
1159 Deoptimization::Action_make_not_entrant);
1160 }
1161
1162 if (stopped()) {
1163 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1164 return true;
1165 }
1166
1167 // length is now known positive, add a cast node to make this explicit
1168 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1169 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1170 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1171 ConstraintCastNode::RegularDependency, bt);
1172 casted_length = _gvn.transform(casted_length);
1173 replace_in_map(length, casted_length);
1174 length = casted_length;
1175
1176 // Use an unsigned comparison for the range check itself
1177 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1178 BoolTest::mask btest = BoolTest::lt;
1179 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1180 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1181 _gvn.set_type(rc, rc->Value(&_gvn));
1182 if (!rc_bool->is_Con()) {
1183 record_for_igvn(rc);
1184 }
1185 set_control(_gvn.transform(new IfTrueNode(rc)));
1186 {
1187 PreserveJVMState pjvms(this);
1188 set_control(_gvn.transform(new IfFalseNode(rc)));
1189 uncommon_trap(Deoptimization::Reason_range_check,
1190 Deoptimization::Action_make_not_entrant);
1191 }
1192
1193 if (stopped()) {
1194 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1195 return true;
1196 }
1197
1198 // index is now known to be >= 0 and < length, cast it
1199 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1200 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1201 ConstraintCastNode::RegularDependency, bt);
1202 result = _gvn.transform(result);
1203 set_result(result);
1204 replace_in_map(index, result);
1205 return true;
1206 }
1207
1208 //------------------------------inline_string_indexOf------------------------
1209 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1210 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1211 return false;
1212 }
1213 Node* src = argument(0);
1214 Node* tgt = argument(1);
1215
1216 // Make the merge point
1217 RegionNode* result_rgn = new RegionNode(4);
1218 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1219
1220 src = must_be_not_null(src, true);
1221 tgt = must_be_not_null(tgt, true);
1222
1223 // Get start addr and length of source string
1224 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1225 Node* src_count = load_array_length(src);
1226
1227 // Get start addr and length of substring
1228 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1229 Node* tgt_count = load_array_length(tgt);
1230
1231 Node* result = nullptr;
1232 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1233
1234 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1235 // Divide src size by 2 if String is UTF16 encoded
1236 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1237 }
1238 if (ae == StrIntrinsicNode::UU) {
1239 // Divide substring size by 2 if String is UTF16 encoded
1240 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1241 }
1242
1243 if (call_opt_stub) {
1244 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1245 StubRoutines::_string_indexof_array[ae],
1246 "stringIndexOf", TypePtr::BOTTOM, src_start,
1247 src_count, tgt_start, tgt_count);
1248 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1249 } else {
1250 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1251 result_rgn, result_phi, ae);
1252 }
1253 if (result != nullptr) {
1254 result_phi->init_req(3, result);
1255 result_rgn->init_req(3, control());
1256 }
1257 set_control(_gvn.transform(result_rgn));
1258 record_for_igvn(result_rgn);
1259 set_result(_gvn.transform(result_phi));
1260
1261 return true;
1262 }
1263
1264 //-----------------------------inline_string_indexOfI-----------------------
1265 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1266 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1267 return false;
1268 }
1269 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1270 return false;
1271 }
1272
1273 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1274 Node* src = argument(0); // byte[]
1275 Node* src_count = argument(1); // char count
1276 Node* tgt = argument(2); // byte[]
1277 Node* tgt_count = argument(3); // char count
1278 Node* from_index = argument(4); // char index
1279
1280 src = must_be_not_null(src, true);
1281 tgt = must_be_not_null(tgt, true);
1282
1283 // Multiply byte array index by 2 if String is UTF16 encoded
1284 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1285 src_count = _gvn.transform(new SubINode(src_count, from_index));
1286 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1287 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1288
1289 // Range checks
1290 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1291 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1292 if (stopped()) {
1293 return true;
1294 }
1295
1296 RegionNode* region = new RegionNode(5);
1297 Node* phi = new PhiNode(region, TypeInt::INT);
1298 Node* result = nullptr;
1299
1300 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1301
1302 if (call_opt_stub) {
1303 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1304 StubRoutines::_string_indexof_array[ae],
1305 "stringIndexOf", TypePtr::BOTTOM, src_start,
1306 src_count, tgt_start, tgt_count);
1307 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1308 } else {
1309 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1310 region, phi, ae);
1311 }
1312 if (result != nullptr) {
1313 // The result is index relative to from_index if substring was found, -1 otherwise.
1314 // Generate code which will fold into cmove.
1315 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1316 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1317
1318 Node* if_lt = generate_slow_guard(bol, nullptr);
1319 if (if_lt != nullptr) {
1320 // result == -1
1321 phi->init_req(3, result);
1322 region->init_req(3, if_lt);
1323 }
1324 if (!stopped()) {
1325 result = _gvn.transform(new AddINode(result, from_index));
1326 phi->init_req(4, result);
1327 region->init_req(4, control());
1328 }
1329 }
1330
1331 set_control(_gvn.transform(region));
1332 record_for_igvn(region);
1333 set_result(_gvn.transform(phi));
1334 clear_upper_avx();
1335
1336 return true;
1337 }
1338
1339 // Create StrIndexOfNode with fast path checks
1340 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1341 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1342 // Check for substr count > string count
1343 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1344 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1345 Node* if_gt = generate_slow_guard(bol, nullptr);
1346 if (if_gt != nullptr) {
1347 phi->init_req(1, intcon(-1));
1348 region->init_req(1, if_gt);
1349 }
1350 if (!stopped()) {
1351 // Check for substr count == 0
1352 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1353 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1354 Node* if_zero = generate_slow_guard(bol, nullptr);
1355 if (if_zero != nullptr) {
1356 phi->init_req(2, intcon(0));
1357 region->init_req(2, if_zero);
1358 }
1359 }
1360 if (!stopped()) {
1361 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1362 }
1363 return nullptr;
1364 }
1365
1366 //-----------------------------inline_string_indexOfChar-----------------------
1367 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1368 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1369 return false;
1370 }
1371 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1372 return false;
1373 }
1374 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1375 Node* src = argument(0); // byte[]
1376 Node* int_ch = argument(1);
1377 Node* from_index = argument(2);
1378 Node* max = argument(3);
1379
1380 src = must_be_not_null(src, true);
1381
1382 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1383 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1384 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1385
1386 // Range checks
1387 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1388
1389 // Check for int_ch >= 0
1390 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1391 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1392 {
1393 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1394 uncommon_trap(Deoptimization::Reason_intrinsic,
1395 Deoptimization::Action_maybe_recompile);
1396 }
1397 if (stopped()) {
1398 return true;
1399 }
1400
1401 RegionNode* region = new RegionNode(3);
1402 Node* phi = new PhiNode(region, TypeInt::INT);
1403
1404 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1405 C->set_has_split_ifs(true); // Has chance for split-if optimization
1406 _gvn.transform(result);
1407
1408 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1409 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1410
1411 Node* if_lt = generate_slow_guard(bol, nullptr);
1412 if (if_lt != nullptr) {
1413 // result == -1
1414 phi->init_req(2, result);
1415 region->init_req(2, if_lt);
1416 }
1417 if (!stopped()) {
1418 result = _gvn.transform(new AddINode(result, from_index));
1419 phi->init_req(1, result);
1420 region->init_req(1, control());
1421 }
1422 set_control(_gvn.transform(region));
1423 record_for_igvn(region);
1424 set_result(_gvn.transform(phi));
1425 clear_upper_avx();
1426
1427 return true;
1428 }
1429 //---------------------------inline_string_copy---------------------
1430 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1431 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1432 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1433 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1434 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1435 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1436 bool LibraryCallKit::inline_string_copy(bool compress) {
1437 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1438 return false;
1439 }
1440 int nargs = 5; // 2 oops, 3 ints
1441 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1442
1443 Node* src = argument(0);
1444 Node* src_offset = argument(1);
1445 Node* dst = argument(2);
1446 Node* dst_offset = argument(3);
1447 Node* length = argument(4);
1448
1449 // Check for allocation before we add nodes that would confuse
1450 // tightly_coupled_allocation()
1451 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1452
1453 // Figure out the size and type of the elements we will be copying.
1454 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1455 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1456 if (src_type == nullptr || dst_type == nullptr) {
1457 return false;
1458 }
1459 BasicType src_elem = src_type->elem()->array_element_basic_type();
1460 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1461 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1462 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1463 "Unsupported array types for inline_string_copy");
1464
1465 src = must_be_not_null(src, true);
1466 dst = must_be_not_null(dst, true);
1467
1468 // Convert char[] offsets to byte[] offsets
1469 bool convert_src = (compress && src_elem == T_BYTE);
1470 bool convert_dst = (!compress && dst_elem == T_BYTE);
1471 if (convert_src) {
1472 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1473 } else if (convert_dst) {
1474 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1475 }
1476
1477 // Range checks
1478 generate_string_range_check(src, src_offset, length, convert_src);
1479 generate_string_range_check(dst, dst_offset, length, convert_dst);
1480 if (stopped()) {
1481 return true;
1482 }
1483
1484 Node* src_start = array_element_address(src, src_offset, src_elem);
1485 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1486 // 'src_start' points to src array + scaled offset
1487 // 'dst_start' points to dst array + scaled offset
1488 Node* count = nullptr;
1489 if (compress) {
1490 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1491 } else {
1492 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1493 }
1494
1495 if (alloc != nullptr) {
1496 if (alloc->maybe_set_complete(&_gvn)) {
1497 // "You break it, you buy it."
1498 InitializeNode* init = alloc->initialization();
1499 assert(init->is_complete(), "we just did this");
1500 init->set_complete_with_arraycopy();
1501 assert(dst->is_CheckCastPP(), "sanity");
1502 assert(dst->in(0)->in(0) == init, "dest pinned");
1503 }
1504 // Do not let stores that initialize this object be reordered with
1505 // a subsequent store that would make this object accessible by
1506 // other threads.
1507 // Record what AllocateNode this StoreStore protects so that
1508 // escape analysis can go from the MemBarStoreStoreNode to the
1509 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1510 // based on the escape status of the AllocateNode.
1511 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1512 }
1513 if (compress) {
1514 set_result(_gvn.transform(count));
1515 }
1516 clear_upper_avx();
1517
1518 return true;
1519 }
1520
1521 #ifdef _LP64
1522 #define XTOP ,top() /*additional argument*/
1523 #else //_LP64
1524 #define XTOP /*no additional argument*/
1525 #endif //_LP64
1526
1527 //------------------------inline_string_toBytesU--------------------------
1528 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1529 bool LibraryCallKit::inline_string_toBytesU() {
1530 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1531 return false;
1532 }
1533 // Get the arguments.
1534 Node* value = argument(0);
1535 Node* offset = argument(1);
1536 Node* length = argument(2);
1537
1538 Node* newcopy = nullptr;
1539
1540 // Set the original stack and the reexecute bit for the interpreter to reexecute
1541 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1542 { PreserveReexecuteState preexecs(this);
1543 jvms()->set_should_reexecute(true);
1544
1545 // Check if a null path was taken unconditionally.
1546 value = null_check(value);
1547
1548 RegionNode* bailout = new RegionNode(1);
1549 record_for_igvn(bailout);
1550
1551 // Range checks
1552 generate_negative_guard(offset, bailout);
1553 generate_negative_guard(length, bailout);
1554 generate_limit_guard(offset, length, load_array_length(value), bailout);
1555 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1556 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1557
1558 if (bailout->req() > 1) {
1559 PreserveJVMState pjvms(this);
1560 set_control(_gvn.transform(bailout));
1561 uncommon_trap(Deoptimization::Reason_intrinsic,
1562 Deoptimization::Action_maybe_recompile);
1563 }
1564 if (stopped()) {
1565 return true;
1566 }
1567
1568 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1569 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1570 newcopy = new_array(klass_node, size, 0); // no arguments to push
1571 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1572 guarantee(alloc != nullptr, "created above");
1573
1574 // Calculate starting addresses.
1575 Node* src_start = array_element_address(value, offset, T_CHAR);
1576 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1577
1578 // Check if dst array address is aligned to HeapWordSize
1579 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1580 // If true, then check if src array address is aligned to HeapWordSize
1581 if (aligned) {
1582 const TypeInt* toffset = gvn().type(offset)->is_int();
1583 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1584 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1585 }
1586
1587 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1588 const char* copyfunc_name = "arraycopy";
1589 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1590 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1591 OptoRuntime::fast_arraycopy_Type(),
1592 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1593 src_start, dst_start, ConvI2X(length) XTOP);
1594 // Do not let reads from the cloned object float above the arraycopy.
1595 if (alloc->maybe_set_complete(&_gvn)) {
1596 // "You break it, you buy it."
1597 InitializeNode* init = alloc->initialization();
1598 assert(init->is_complete(), "we just did this");
1599 init->set_complete_with_arraycopy();
1600 assert(newcopy->is_CheckCastPP(), "sanity");
1601 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1602 }
1603 // Do not let stores that initialize this object be reordered with
1604 // a subsequent store that would make this object accessible by
1605 // other threads.
1606 // Record what AllocateNode this StoreStore protects so that
1607 // escape analysis can go from the MemBarStoreStoreNode to the
1608 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1609 // based on the escape status of the AllocateNode.
1610 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1611 } // original reexecute is set back here
1612
1613 C->set_has_split_ifs(true); // Has chance for split-if optimization
1614 if (!stopped()) {
1615 set_result(newcopy);
1616 }
1617 clear_upper_avx();
1618
1619 return true;
1620 }
1621
1622 //------------------------inline_string_getCharsU--------------------------
1623 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1624 bool LibraryCallKit::inline_string_getCharsU() {
1625 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1626 return false;
1627 }
1628
1629 // Get the arguments.
1630 Node* src = argument(0);
1631 Node* src_begin = argument(1);
1632 Node* src_end = argument(2); // exclusive offset (i < src_end)
1633 Node* dst = argument(3);
1634 Node* dst_begin = argument(4);
1635
1636 // Check for allocation before we add nodes that would confuse
1637 // tightly_coupled_allocation()
1638 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1639
1640 // Check if a null path was taken unconditionally.
1641 src = null_check(src);
1642 dst = null_check(dst);
1643 if (stopped()) {
1644 return true;
1645 }
1646
1647 // Get length and convert char[] offset to byte[] offset
1648 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1649 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1650
1651 // Range checks
1652 generate_string_range_check(src, src_begin, length, true);
1653 generate_string_range_check(dst, dst_begin, length, false);
1654 if (stopped()) {
1655 return true;
1656 }
1657
1658 if (!stopped()) {
1659 // Calculate starting addresses.
1660 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1661 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1662
1663 // Check if array addresses are aligned to HeapWordSize
1664 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1665 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1666 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1667 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1668
1669 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1670 const char* copyfunc_name = "arraycopy";
1671 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1672 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1673 OptoRuntime::fast_arraycopy_Type(),
1674 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1675 src_start, dst_start, ConvI2X(length) XTOP);
1676 // Do not let reads from the cloned object float above the arraycopy.
1677 if (alloc != nullptr) {
1678 if (alloc->maybe_set_complete(&_gvn)) {
1679 // "You break it, you buy it."
1680 InitializeNode* init = alloc->initialization();
1681 assert(init->is_complete(), "we just did this");
1682 init->set_complete_with_arraycopy();
1683 assert(dst->is_CheckCastPP(), "sanity");
1684 assert(dst->in(0)->in(0) == init, "dest pinned");
1685 }
1686 // Do not let stores that initialize this object be reordered with
1687 // a subsequent store that would make this object accessible by
1688 // other threads.
1689 // Record what AllocateNode this StoreStore protects so that
1690 // escape analysis can go from the MemBarStoreStoreNode to the
1691 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1692 // based on the escape status of the AllocateNode.
1693 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1694 } else {
1695 insert_mem_bar(Op_MemBarCPUOrder);
1696 }
1697 }
1698
1699 C->set_has_split_ifs(true); // Has chance for split-if optimization
1700 return true;
1701 }
1702
1703 //----------------------inline_string_char_access----------------------------
1704 // Store/Load char to/from byte[] array.
1705 // static void StringUTF16.putChar(byte[] val, int index, int c)
1706 // static char StringUTF16.getChar(byte[] val, int index)
1707 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1708 Node* value = argument(0);
1709 Node* index = argument(1);
1710 Node* ch = is_store ? argument(2) : nullptr;
1711
1712 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1713 // correctly requires matched array shapes.
1714 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1715 "sanity: byte[] and char[] bases agree");
1716 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1717 "sanity: byte[] and char[] scales agree");
1718
1719 // Bail when getChar over constants is requested: constant folding would
1720 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1721 // Java method would constant fold nicely instead.
1722 if (!is_store && value->is_Con() && index->is_Con()) {
1723 return false;
1724 }
1725
1726 // Save state and restore on bailout
1727 uint old_sp = sp();
1728 SafePointNode* old_map = clone_map();
1729
1730 value = must_be_not_null(value, true);
1731
1732 Node* adr = array_element_address(value, index, T_CHAR);
1733 if (adr->is_top()) {
1734 set_map(old_map);
1735 set_sp(old_sp);
1736 return false;
1737 }
1738 destruct_map_clone(old_map);
1739 if (is_store) {
1740 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1741 } else {
1742 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);
1743 set_result(ch);
1744 }
1745 return true;
1746 }
1747
1748
1749 //------------------------------inline_math-----------------------------------
1750 // public static double Math.abs(double)
1751 // public static double Math.sqrt(double)
1752 // public static double Math.log(double)
1753 // public static double Math.log10(double)
1754 // public static double Math.round(double)
1755 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1756 Node* arg = argument(0);
1757 Node* n = nullptr;
1758 switch (id) {
1759 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1760 case vmIntrinsics::_dsqrt:
1761 case vmIntrinsics::_dsqrt_strict:
1762 n = new SqrtDNode(C, control(), arg); break;
1763 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1764 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1765 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1766 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1767 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1768 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1769 default: fatal_unexpected_iid(id); break;
1770 }
1771 set_result(_gvn.transform(n));
1772 return true;
1773 }
1774
1775 //------------------------------inline_math-----------------------------------
1776 // public static float Math.abs(float)
1777 // public static int Math.abs(int)
1778 // public static long Math.abs(long)
1779 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1780 Node* arg = argument(0);
1781 Node* n = nullptr;
1782 switch (id) {
1783 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1784 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1785 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1786 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1787 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1788 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1789 default: fatal_unexpected_iid(id); break;
1790 }
1791 set_result(_gvn.transform(n));
1792 return true;
1793 }
1794
1795 //------------------------------runtime_math-----------------------------
1796 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1797 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1798 "must be (DD)D or (D)D type");
1799
1800 // Inputs
1801 Node* a = argument(0);
1802 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1803
1804 const TypePtr* no_memory_effects = nullptr;
1805 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1806 no_memory_effects,
1807 a, top(), b, b ? top() : nullptr);
1808 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1809 #ifdef ASSERT
1810 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1811 assert(value_top == top(), "second value must be top");
1812 #endif
1813
1814 set_result(value);
1815 return true;
1816 }
1817
1818 //------------------------------inline_math_pow-----------------------------
1819 bool LibraryCallKit::inline_math_pow() {
1820 Node* exp = argument(2);
1821 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1822 if (d != nullptr) {
1823 if (d->getd() == 2.0) {
1824 // Special case: pow(x, 2.0) => x * x
1825 Node* base = argument(0);
1826 set_result(_gvn.transform(new MulDNode(base, base)));
1827 return true;
1828 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1829 // Special case: pow(x, 0.5) => sqrt(x)
1830 Node* base = argument(0);
1831 Node* zero = _gvn.zerocon(T_DOUBLE);
1832
1833 RegionNode* region = new RegionNode(3);
1834 Node* phi = new PhiNode(region, Type::DOUBLE);
1835
1836 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1837 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1838 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1839 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1840 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1841
1842 Node* if_pow = generate_slow_guard(test, nullptr);
1843 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1844 phi->init_req(1, value_sqrt);
1845 region->init_req(1, control());
1846
1847 if (if_pow != nullptr) {
1848 set_control(if_pow);
1849 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1850 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1851 const TypePtr* no_memory_effects = nullptr;
1852 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1853 no_memory_effects, base, top(), exp, top());
1854 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1855 #ifdef ASSERT
1856 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1857 assert(value_top == top(), "second value must be top");
1858 #endif
1859 phi->init_req(2, value_pow);
1860 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1861 }
1862
1863 C->set_has_split_ifs(true); // Has chance for split-if optimization
1864 set_control(_gvn.transform(region));
1865 record_for_igvn(region);
1866 set_result(_gvn.transform(phi));
1867
1868 return true;
1869 }
1870 }
1871
1872 return StubRoutines::dpow() != nullptr ?
1873 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1874 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1875 }
1876
1877 //------------------------------inline_math_native-----------------------------
1878 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1879 switch (id) {
1880 case vmIntrinsics::_dsin:
1881 return StubRoutines::dsin() != nullptr ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1883 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1884 case vmIntrinsics::_dcos:
1885 return StubRoutines::dcos() != nullptr ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1887 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1888 case vmIntrinsics::_dtan:
1889 return StubRoutines::dtan() != nullptr ?
1890 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1891 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1892 case vmIntrinsics::_dtanh:
1893 return StubRoutines::dtanh() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1895 case vmIntrinsics::_dcbrt:
1896 return StubRoutines::dcbrt() != nullptr ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1898 case vmIntrinsics::_dexp:
1899 return StubRoutines::dexp() != nullptr ?
1900 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1901 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1902 case vmIntrinsics::_dlog:
1903 return StubRoutines::dlog() != nullptr ?
1904 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1905 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1906 case vmIntrinsics::_dlog10:
1907 return StubRoutines::dlog10() != nullptr ?
1908 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1909 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1910
1911 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1912 case vmIntrinsics::_ceil:
1913 case vmIntrinsics::_floor:
1914 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1915
1916 case vmIntrinsics::_dsqrt:
1917 case vmIntrinsics::_dsqrt_strict:
1918 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1919 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1920 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1921 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1922 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1923
1924 case vmIntrinsics::_dpow: return inline_math_pow();
1925 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1926 case vmIntrinsics::_fcopySign: return inline_math(id);
1927 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1928 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1929 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1930
1931 // These intrinsics are not yet correctly implemented
1932 case vmIntrinsics::_datan2:
1933 return false;
1934
1935 default:
1936 fatal_unexpected_iid(id);
1937 return false;
1938 }
1939 }
1940
1941 //----------------------------inline_notify-----------------------------------*
1942 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1943 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1944 address func;
1945 if (id == vmIntrinsics::_notify) {
1946 func = OptoRuntime::monitor_notify_Java();
1947 } else {
1948 func = OptoRuntime::monitor_notifyAll_Java();
1949 }
1950 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1951 make_slow_call_ex(call, env()->Throwable_klass(), false);
1952 return true;
1953 }
1954
1955
1956 //----------------------------inline_min_max-----------------------------------
1957 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1958 Node* a = nullptr;
1959 Node* b = nullptr;
1960 Node* n = nullptr;
1961 switch (id) {
1962 case vmIntrinsics::_min:
1963 case vmIntrinsics::_max:
1964 case vmIntrinsics::_minF:
1965 case vmIntrinsics::_maxF:
1966 case vmIntrinsics::_minF_strict:
1967 case vmIntrinsics::_maxF_strict:
1968 case vmIntrinsics::_min_strict:
1969 case vmIntrinsics::_max_strict:
1970 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1971 a = argument(0);
1972 b = argument(1);
1973 break;
1974 case vmIntrinsics::_minD:
1975 case vmIntrinsics::_maxD:
1976 case vmIntrinsics::_minD_strict:
1977 case vmIntrinsics::_maxD_strict:
1978 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1979 a = argument(0);
1980 b = argument(2);
1981 break;
1982 case vmIntrinsics::_minL:
1983 case vmIntrinsics::_maxL:
1984 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1985 a = argument(0);
1986 b = argument(2);
1987 break;
1988 default:
1989 fatal_unexpected_iid(id);
1990 break;
1991 }
1992
1993 switch (id) {
1994 case vmIntrinsics::_min:
1995 case vmIntrinsics::_min_strict:
1996 n = new MinINode(a, b);
1997 break;
1998 case vmIntrinsics::_max:
1999 case vmIntrinsics::_max_strict:
2000 n = new MaxINode(a, b);
2001 break;
2002 case vmIntrinsics::_minF:
2003 case vmIntrinsics::_minF_strict:
2004 n = new MinFNode(a, b);
2005 break;
2006 case vmIntrinsics::_maxF:
2007 case vmIntrinsics::_maxF_strict:
2008 n = new MaxFNode(a, b);
2009 break;
2010 case vmIntrinsics::_minD:
2011 case vmIntrinsics::_minD_strict:
2012 n = new MinDNode(a, b);
2013 break;
2014 case vmIntrinsics::_maxD:
2015 case vmIntrinsics::_maxD_strict:
2016 n = new MaxDNode(a, b);
2017 break;
2018 case vmIntrinsics::_minL:
2019 n = new MinLNode(_gvn.C, a, b);
2020 break;
2021 case vmIntrinsics::_maxL:
2022 n = new MaxLNode(_gvn.C, a, b);
2023 break;
2024 default:
2025 fatal_unexpected_iid(id);
2026 break;
2027 }
2028
2029 set_result(_gvn.transform(n));
2030 return true;
2031 }
2032
2033 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2034 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2035 env()->ArithmeticException_instance())) {
2036 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2037 // so let's bail out intrinsic rather than risking deopting again.
2038 return false;
2039 }
2040
2041 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2042 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2043 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2044 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2045
2046 {
2047 PreserveJVMState pjvms(this);
2048 PreserveReexecuteState preexecs(this);
2049 jvms()->set_should_reexecute(true);
2050
2051 set_control(slow_path);
2052 set_i_o(i_o());
2053
2054 builtin_throw(Deoptimization::Reason_intrinsic,
2055 env()->ArithmeticException_instance(),
2056 /*allow_too_many_traps*/ false);
2057 }
2058
2059 set_control(fast_path);
2060 set_result(math);
2061 return true;
2062 }
2063
2064 template <typename OverflowOp>
2065 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2066 typedef typename OverflowOp::MathOp MathOp;
2067
2068 MathOp* mathOp = new MathOp(arg1, arg2);
2069 Node* operation = _gvn.transform( mathOp );
2070 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2071 return inline_math_mathExact(operation, ofcheck);
2072 }
2073
2074 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2075 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2076 }
2077
2078 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2079 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2080 }
2081
2082 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2083 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2084 }
2085
2086 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2087 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2088 }
2089
2090 bool LibraryCallKit::inline_math_negateExactI() {
2091 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2092 }
2093
2094 bool LibraryCallKit::inline_math_negateExactL() {
2095 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2096 }
2097
2098 bool LibraryCallKit::inline_math_multiplyExactI() {
2099 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2100 }
2101
2102 bool LibraryCallKit::inline_math_multiplyExactL() {
2103 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2104 }
2105
2106 bool LibraryCallKit::inline_math_multiplyHigh() {
2107 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2108 return true;
2109 }
2110
2111 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2112 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2113 return true;
2114 }
2115
2116 inline int
2117 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2118 const TypePtr* base_type = TypePtr::NULL_PTR;
2119 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2120 if (base_type == nullptr) {
2121 // Unknown type.
2122 return Type::AnyPtr;
2123 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2124 // Since this is a null+long form, we have to switch to a rawptr.
2125 base = _gvn.transform(new CastX2PNode(offset));
2126 offset = MakeConX(0);
2127 return Type::RawPtr;
2128 } else if (base_type->base() == Type::RawPtr) {
2129 return Type::RawPtr;
2130 } else if (base_type->isa_oopptr()) {
2131 // Base is never null => always a heap address.
2132 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2133 return Type::OopPtr;
2134 }
2135 // Offset is small => always a heap address.
2136 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2137 if (offset_type != nullptr &&
2138 base_type->offset() == 0 && // (should always be?)
2139 offset_type->_lo >= 0 &&
2140 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2141 return Type::OopPtr;
2142 } else if (type == T_OBJECT) {
2143 // off heap access to an oop doesn't make any sense. Has to be on
2144 // heap.
2145 return Type::OopPtr;
2146 }
2147 // Otherwise, it might either be oop+off or null+addr.
2148 return Type::AnyPtr;
2149 } else {
2150 // No information:
2151 return Type::AnyPtr;
2152 }
2153 }
2154
2155 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2156 Node* uncasted_base = base;
2157 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2158 if (kind == Type::RawPtr) {
2159 return basic_plus_adr(top(), uncasted_base, offset);
2160 } else if (kind == Type::AnyPtr) {
2161 assert(base == uncasted_base, "unexpected base change");
2162 if (can_cast) {
2163 if (!_gvn.type(base)->speculative_maybe_null() &&
2164 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2165 // According to profiling, this access is always on
2166 // heap. Casting the base to not null and thus avoiding membars
2167 // around the access should allow better optimizations
2168 Node* null_ctl = top();
2169 base = null_check_oop(base, &null_ctl, true, true, true);
2170 assert(null_ctl->is_top(), "no null control here");
2171 return basic_plus_adr(base, offset);
2172 } else if (_gvn.type(base)->speculative_always_null() &&
2173 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2174 // According to profiling, this access is always off
2175 // heap.
2176 base = null_assert(base);
2177 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2178 offset = MakeConX(0);
2179 return basic_plus_adr(top(), raw_base, offset);
2180 }
2181 }
2182 // We don't know if it's an on heap or off heap access. Fall back
2183 // to raw memory access.
2184 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2185 return basic_plus_adr(top(), raw, offset);
2186 } else {
2187 assert(base == uncasted_base, "unexpected base change");
2188 // We know it's an on heap access so base can't be null
2189 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2190 base = must_be_not_null(base, true);
2191 }
2192 return basic_plus_adr(base, offset);
2193 }
2194 }
2195
2196 //--------------------------inline_number_methods-----------------------------
2197 // inline int Integer.numberOfLeadingZeros(int)
2198 // inline int Long.numberOfLeadingZeros(long)
2199 //
2200 // inline int Integer.numberOfTrailingZeros(int)
2201 // inline int Long.numberOfTrailingZeros(long)
2202 //
2203 // inline int Integer.bitCount(int)
2204 // inline int Long.bitCount(long)
2205 //
2206 // inline char Character.reverseBytes(char)
2207 // inline short Short.reverseBytes(short)
2208 // inline int Integer.reverseBytes(int)
2209 // inline long Long.reverseBytes(long)
2210 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2211 Node* arg = argument(0);
2212 Node* n = nullptr;
2213 switch (id) {
2214 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2215 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2216 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2217 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2218 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2219 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2220 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2221 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2222 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2223 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2224 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2225 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2226 default: fatal_unexpected_iid(id); break;
2227 }
2228 set_result(_gvn.transform(n));
2229 return true;
2230 }
2231
2232 //--------------------------inline_bitshuffle_methods-----------------------------
2233 // inline int Integer.compress(int, int)
2234 // inline int Integer.expand(int, int)
2235 // inline long Long.compress(long, long)
2236 // inline long Long.expand(long, long)
2237 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2238 Node* n = nullptr;
2239 switch (id) {
2240 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2241 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2242 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2243 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2244 default: fatal_unexpected_iid(id); break;
2245 }
2246 set_result(_gvn.transform(n));
2247 return true;
2248 }
2249
2250 //--------------------------inline_number_methods-----------------------------
2251 // inline int Integer.compareUnsigned(int, int)
2252 // inline int Long.compareUnsigned(long, long)
2253 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2254 Node* arg1 = argument(0);
2255 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2256 Node* n = nullptr;
2257 switch (id) {
2258 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2259 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2260 default: fatal_unexpected_iid(id); break;
2261 }
2262 set_result(_gvn.transform(n));
2263 return true;
2264 }
2265
2266 //--------------------------inline_unsigned_divmod_methods-----------------------------
2267 // inline int Integer.divideUnsigned(int, int)
2268 // inline int Integer.remainderUnsigned(int, int)
2269 // inline long Long.divideUnsigned(long, long)
2270 // inline long Long.remainderUnsigned(long, long)
2271 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2272 Node* n = nullptr;
2273 switch (id) {
2274 case vmIntrinsics::_divideUnsigned_i: {
2275 zero_check_int(argument(1));
2276 // Compile-time detect of null-exception
2277 if (stopped()) {
2278 return true; // keep the graph constructed so far
2279 }
2280 n = new UDivINode(control(), argument(0), argument(1));
2281 break;
2282 }
2283 case vmIntrinsics::_divideUnsigned_l: {
2284 zero_check_long(argument(2));
2285 // Compile-time detect of null-exception
2286 if (stopped()) {
2287 return true; // keep the graph constructed so far
2288 }
2289 n = new UDivLNode(control(), argument(0), argument(2));
2290 break;
2291 }
2292 case vmIntrinsics::_remainderUnsigned_i: {
2293 zero_check_int(argument(1));
2294 // Compile-time detect of null-exception
2295 if (stopped()) {
2296 return true; // keep the graph constructed so far
2297 }
2298 n = new UModINode(control(), argument(0), argument(1));
2299 break;
2300 }
2301 case vmIntrinsics::_remainderUnsigned_l: {
2302 zero_check_long(argument(2));
2303 // Compile-time detect of null-exception
2304 if (stopped()) {
2305 return true; // keep the graph constructed so far
2306 }
2307 n = new UModLNode(control(), argument(0), argument(2));
2308 break;
2309 }
2310 default: fatal_unexpected_iid(id); break;
2311 }
2312 set_result(_gvn.transform(n));
2313 return true;
2314 }
2315
2316 //----------------------------inline_unsafe_access----------------------------
2317
2318 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2319 // Attempt to infer a sharper value type from the offset and base type.
2320 ciKlass* sharpened_klass = nullptr;
2321
2322 // See if it is an instance field, with an object type.
2323 if (alias_type->field() != nullptr) {
2324 if (alias_type->field()->type()->is_klass()) {
2325 sharpened_klass = alias_type->field()->type()->as_klass();
2326 }
2327 }
2328
2329 const TypeOopPtr* result = nullptr;
2330 // See if it is a narrow oop array.
2331 if (adr_type->isa_aryptr()) {
2332 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2333 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2334 if (elem_type != nullptr && elem_type->is_loaded()) {
2335 // Sharpen the value type.
2336 result = elem_type;
2337 }
2338 }
2339 }
2340
2341 // The sharpened class might be unloaded if there is no class loader
2342 // contraint in place.
2343 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2344 // Sharpen the value type.
2345 result = TypeOopPtr::make_from_klass(sharpened_klass);
2346 }
2347 if (result != nullptr) {
2348 #ifndef PRODUCT
2349 if (C->print_intrinsics() || C->print_inlining()) {
2350 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2351 tty->print(" sharpened value: "); result->dump(); tty->cr();
2352 }
2353 #endif
2354 }
2355 return result;
2356 }
2357
2358 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2359 switch (kind) {
2360 case Relaxed:
2361 return MO_UNORDERED;
2362 case Opaque:
2363 return MO_RELAXED;
2364 case Acquire:
2365 return MO_ACQUIRE;
2366 case Release:
2367 return MO_RELEASE;
2368 case Volatile:
2369 return MO_SEQ_CST;
2370 default:
2371 ShouldNotReachHere();
2372 return 0;
2373 }
2374 }
2375
2376 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2377 if (callee()->is_static()) return false; // caller must have the capability!
2378 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2379 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2380 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2381 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2382
2383 if (is_reference_type(type)) {
2384 decorators |= ON_UNKNOWN_OOP_REF;
2385 }
2386
2387 if (unaligned) {
2388 decorators |= C2_UNALIGNED;
2389 }
2390
2391 #ifndef PRODUCT
2392 {
2393 ResourceMark rm;
2394 // Check the signatures.
2395 ciSignature* sig = callee()->signature();
2396 #ifdef ASSERT
2397 if (!is_store) {
2398 // Object getReference(Object base, int/long offset), etc.
2399 BasicType rtype = sig->return_type()->basic_type();
2400 assert(rtype == type, "getter must return the expected value");
2401 assert(sig->count() == 2, "oop getter has 2 arguments");
2402 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2403 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2404 } else {
2405 // void putReference(Object base, int/long offset, Object x), etc.
2406 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2407 assert(sig->count() == 3, "oop putter has 3 arguments");
2408 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2409 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2410 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2411 assert(vtype == type, "putter must accept the expected value");
2412 }
2413 #endif // ASSERT
2414 }
2415 #endif //PRODUCT
2416
2417 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2418
2419 Node* receiver = argument(0); // type: oop
2420
2421 // Build address expression.
2422 Node* heap_base_oop = top();
2423
2424 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2425 Node* base = argument(1); // type: oop
2426 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2427 Node* offset = argument(2); // type: long
2428 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2429 // to be plain byte offsets, which are also the same as those accepted
2430 // by oopDesc::field_addr.
2431 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2432 "fieldOffset must be byte-scaled");
2433 // 32-bit machines ignore the high half!
2434 offset = ConvL2X(offset);
2435
2436 // Save state and restore on bailout
2437 uint old_sp = sp();
2438 SafePointNode* old_map = clone_map();
2439
2440 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2441 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2442
2443 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2444 if (type != T_OBJECT) {
2445 decorators |= IN_NATIVE; // off-heap primitive access
2446 } else {
2447 set_map(old_map);
2448 set_sp(old_sp);
2449 return false; // off-heap oop accesses are not supported
2450 }
2451 } else {
2452 heap_base_oop = base; // on-heap or mixed access
2453 }
2454
2455 // Can base be null? Otherwise, always on-heap access.
2456 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2457
2458 if (!can_access_non_heap) {
2459 decorators |= IN_HEAP;
2460 }
2461
2462 Node* val = is_store ? argument(4) : nullptr;
2463
2464 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2465 if (adr_type == TypePtr::NULL_PTR) {
2466 set_map(old_map);
2467 set_sp(old_sp);
2468 return false; // off-heap access with zero address
2469 }
2470
2471 // Try to categorize the address.
2472 Compile::AliasType* alias_type = C->alias_type(adr_type);
2473 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2474
2475 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2476 alias_type->adr_type() == TypeAryPtr::RANGE) {
2477 set_map(old_map);
2478 set_sp(old_sp);
2479 return false; // not supported
2480 }
2481
2482 bool mismatched = false;
2483 BasicType bt = alias_type->basic_type();
2484 if (bt != T_ILLEGAL) {
2485 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2486 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2487 // Alias type doesn't differentiate between byte[] and boolean[]).
2488 // Use address type to get the element type.
2489 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2490 }
2491 if (is_reference_type(bt, true)) {
2492 // accessing an array field with getReference is not a mismatch
2493 bt = T_OBJECT;
2494 }
2495 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2496 // Don't intrinsify mismatched object accesses
2497 set_map(old_map);
2498 set_sp(old_sp);
2499 return false;
2500 }
2501 mismatched = (bt != type);
2502 } else if (alias_type->adr_type()->isa_oopptr()) {
2503 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2504 }
2505
2506 destruct_map_clone(old_map);
2507 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2508
2509 if (mismatched) {
2510 decorators |= C2_MISMATCHED;
2511 }
2512
2513 // First guess at the value type.
2514 const Type *value_type = Type::get_const_basic_type(type);
2515
2516 // Figure out the memory ordering.
2517 decorators |= mo_decorator_for_access_kind(kind);
2518
2519 if (!is_store && type == T_OBJECT) {
2520 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2521 if (tjp != nullptr) {
2522 value_type = tjp;
2523 }
2524 }
2525
2526 receiver = null_check(receiver);
2527 if (stopped()) {
2528 return true;
2529 }
2530 // Heap pointers get a null-check from the interpreter,
2531 // as a courtesy. However, this is not guaranteed by Unsafe,
2532 // and it is not possible to fully distinguish unintended nulls
2533 // from intended ones in this API.
2534
2535 if (!is_store) {
2536 Node* p = nullptr;
2537 // Try to constant fold a load from a constant field
2538 ciField* field = alias_type->field();
2539 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2540 // final or stable field
2541 p = make_constant_from_field(field, heap_base_oop);
2542 }
2543
2544 if (p == nullptr) { // Could not constant fold the load
2545 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2546 // Normalize the value returned by getBoolean in the following cases
2547 if (type == T_BOOLEAN &&
2548 (mismatched ||
2549 heap_base_oop == top() || // - heap_base_oop is null or
2550 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2551 // and the unsafe access is made to large offset
2552 // (i.e., larger than the maximum offset necessary for any
2553 // field access)
2554 ) {
2555 IdealKit ideal = IdealKit(this);
2556 #define __ ideal.
2557 IdealVariable normalized_result(ideal);
2558 __ declarations_done();
2559 __ set(normalized_result, p);
2560 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2561 __ set(normalized_result, ideal.ConI(1));
2562 ideal.end_if();
2563 final_sync(ideal);
2564 p = __ value(normalized_result);
2565 #undef __
2566 }
2567 }
2568 if (type == T_ADDRESS) {
2569 p = gvn().transform(new CastP2XNode(nullptr, p));
2570 p = ConvX2UL(p);
2571 }
2572 // The load node has the control of the preceding MemBarCPUOrder. All
2573 // following nodes will have the control of the MemBarCPUOrder inserted at
2574 // the end of this method. So, pushing the load onto the stack at a later
2575 // point is fine.
2576 set_result(p);
2577 } else {
2578 if (bt == T_ADDRESS) {
2579 // Repackage the long as a pointer.
2580 val = ConvL2X(val);
2581 val = gvn().transform(new CastX2PNode(val));
2582 }
2583 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2584 }
2585
2586 return true;
2587 }
2588
2589 //----------------------------inline_unsafe_load_store----------------------------
2590 // This method serves a couple of different customers (depending on LoadStoreKind):
2591 //
2592 // LS_cmp_swap:
2593 //
2594 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2595 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2596 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2597 //
2598 // LS_cmp_swap_weak:
2599 //
2600 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2601 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2602 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2603 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2604 //
2605 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2606 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2607 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2608 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2609 //
2610 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2611 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2612 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2613 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2614 //
2615 // LS_cmp_exchange:
2616 //
2617 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2620 //
2621 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2622 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2624 //
2625 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2626 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2627 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2628 //
2629 // LS_get_add:
2630 //
2631 // int getAndAddInt( Object o, long offset, int delta)
2632 // long getAndAddLong(Object o, long offset, long delta)
2633 //
2634 // LS_get_set:
2635 //
2636 // int getAndSet(Object o, long offset, int newValue)
2637 // long getAndSet(Object o, long offset, long newValue)
2638 // Object getAndSet(Object o, long offset, Object newValue)
2639 //
2640 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2641 // This basic scheme here is the same as inline_unsafe_access, but
2642 // differs in enough details that combining them would make the code
2643 // overly confusing. (This is a true fact! I originally combined
2644 // them, but even I was confused by it!) As much code/comments as
2645 // possible are retained from inline_unsafe_access though to make
2646 // the correspondences clearer. - dl
2647
2648 if (callee()->is_static()) return false; // caller must have the capability!
2649
2650 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2651 decorators |= mo_decorator_for_access_kind(access_kind);
2652
2653 #ifndef PRODUCT
2654 BasicType rtype;
2655 {
2656 ResourceMark rm;
2657 // Check the signatures.
2658 ciSignature* sig = callee()->signature();
2659 rtype = sig->return_type()->basic_type();
2660 switch(kind) {
2661 case LS_get_add:
2662 case LS_get_set: {
2663 // Check the signatures.
2664 #ifdef ASSERT
2665 assert(rtype == type, "get and set must return the expected type");
2666 assert(sig->count() == 3, "get and set has 3 arguments");
2667 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2668 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2669 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2670 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2671 #endif // ASSERT
2672 break;
2673 }
2674 case LS_cmp_swap:
2675 case LS_cmp_swap_weak: {
2676 // Check the signatures.
2677 #ifdef ASSERT
2678 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2679 assert(sig->count() == 4, "CAS has 4 arguments");
2680 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2681 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2682 #endif // ASSERT
2683 break;
2684 }
2685 case LS_cmp_exchange: {
2686 // Check the signatures.
2687 #ifdef ASSERT
2688 assert(rtype == type, "CAS must return the expected type");
2689 assert(sig->count() == 4, "CAS has 4 arguments");
2690 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2691 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2692 #endif // ASSERT
2693 break;
2694 }
2695 default:
2696 ShouldNotReachHere();
2697 }
2698 }
2699 #endif //PRODUCT
2700
2701 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2702
2703 // Get arguments:
2704 Node* receiver = nullptr;
2705 Node* base = nullptr;
2706 Node* offset = nullptr;
2707 Node* oldval = nullptr;
2708 Node* newval = nullptr;
2709 switch(kind) {
2710 case LS_cmp_swap:
2711 case LS_cmp_swap_weak:
2712 case LS_cmp_exchange: {
2713 const bool two_slot_type = type2size[type] == 2;
2714 receiver = argument(0); // type: oop
2715 base = argument(1); // type: oop
2716 offset = argument(2); // type: long
2717 oldval = argument(4); // type: oop, int, or long
2718 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2719 break;
2720 }
2721 case LS_get_add:
2722 case LS_get_set: {
2723 receiver = argument(0); // type: oop
2724 base = argument(1); // type: oop
2725 offset = argument(2); // type: long
2726 oldval = nullptr;
2727 newval = argument(4); // type: oop, int, or long
2728 break;
2729 }
2730 default:
2731 ShouldNotReachHere();
2732 }
2733
2734 // Build field offset expression.
2735 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2736 // to be plain byte offsets, which are also the same as those accepted
2737 // by oopDesc::field_addr.
2738 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2739 // 32-bit machines ignore the high half of long offsets
2740 offset = ConvL2X(offset);
2741 // Save state and restore on bailout
2742 uint old_sp = sp();
2743 SafePointNode* old_map = clone_map();
2744 Node* adr = make_unsafe_address(base, offset,type, false);
2745 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2746
2747 Compile::AliasType* alias_type = C->alias_type(adr_type);
2748 BasicType bt = alias_type->basic_type();
2749 if (bt != T_ILLEGAL &&
2750 (is_reference_type(bt) != (type == T_OBJECT))) {
2751 // Don't intrinsify mismatched object accesses.
2752 set_map(old_map);
2753 set_sp(old_sp);
2754 return false;
2755 }
2756
2757 destruct_map_clone(old_map);
2758
2759 // For CAS, unlike inline_unsafe_access, there seems no point in
2760 // trying to refine types. Just use the coarse types here.
2761 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2762 const Type *value_type = Type::get_const_basic_type(type);
2763
2764 switch (kind) {
2765 case LS_get_set:
2766 case LS_cmp_exchange: {
2767 if (type == T_OBJECT) {
2768 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2769 if (tjp != nullptr) {
2770 value_type = tjp;
2771 }
2772 }
2773 break;
2774 }
2775 case LS_cmp_swap:
2776 case LS_cmp_swap_weak:
2777 case LS_get_add:
2778 break;
2779 default:
2780 ShouldNotReachHere();
2781 }
2782
2783 // Null check receiver.
2784 receiver = null_check(receiver);
2785 if (stopped()) {
2786 return true;
2787 }
2788
2789 int alias_idx = C->get_alias_index(adr_type);
2790
2791 if (is_reference_type(type)) {
2792 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2793
2794 // Transformation of a value which could be null pointer (CastPP #null)
2795 // could be delayed during Parse (for example, in adjust_map_after_if()).
2796 // Execute transformation here to avoid barrier generation in such case.
2797 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2798 newval = _gvn.makecon(TypePtr::NULL_PTR);
2799
2800 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2801 // Refine the value to a null constant, when it is known to be null
2802 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2803 }
2804 }
2805
2806 Node* result = nullptr;
2807 switch (kind) {
2808 case LS_cmp_exchange: {
2809 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2810 oldval, newval, value_type, type, decorators);
2811 break;
2812 }
2813 case LS_cmp_swap_weak:
2814 decorators |= C2_WEAK_CMPXCHG;
2815 case LS_cmp_swap: {
2816 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2817 oldval, newval, value_type, type, decorators);
2818 break;
2819 }
2820 case LS_get_set: {
2821 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2822 newval, value_type, type, decorators);
2823 break;
2824 }
2825 case LS_get_add: {
2826 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2827 newval, value_type, type, decorators);
2828 break;
2829 }
2830 default:
2831 ShouldNotReachHere();
2832 }
2833
2834 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2835 set_result(result);
2836 return true;
2837 }
2838
2839 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2840 // Regardless of form, don't allow previous ld/st to move down,
2841 // then issue acquire, release, or volatile mem_bar.
2842 insert_mem_bar(Op_MemBarCPUOrder);
2843 switch(id) {
2844 case vmIntrinsics::_loadFence:
2845 insert_mem_bar(Op_LoadFence);
2846 return true;
2847 case vmIntrinsics::_storeFence:
2848 insert_mem_bar(Op_StoreFence);
2849 return true;
2850 case vmIntrinsics::_storeStoreFence:
2851 insert_mem_bar(Op_StoreStoreFence);
2852 return true;
2853 case vmIntrinsics::_fullFence:
2854 insert_mem_bar(Op_MemBarVolatile);
2855 return true;
2856 default:
2857 fatal_unexpected_iid(id);
2858 return false;
2859 }
2860 }
2861
2862 bool LibraryCallKit::inline_onspinwait() {
2863 insert_mem_bar(Op_OnSpinWait);
2864 return true;
2865 }
2866
2867 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2868 if (!kls->is_Con()) {
2869 return true;
2870 }
2871 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2872 if (klsptr == nullptr) {
2873 return true;
2874 }
2875 ciInstanceKlass* ik = klsptr->instance_klass();
2876 // don't need a guard for a klass that is already initialized
2877 return !ik->is_initialized();
2878 }
2879
2880 //----------------------------inline_unsafe_writeback0-------------------------
2881 // public native void Unsafe.writeback0(long address)
2882 bool LibraryCallKit::inline_unsafe_writeback0() {
2883 if (!Matcher::has_match_rule(Op_CacheWB)) {
2884 return false;
2885 }
2886 #ifndef PRODUCT
2887 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2888 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2889 ciSignature* sig = callee()->signature();
2890 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2891 #endif
2892 null_check_receiver(); // null-check, then ignore
2893 Node *addr = argument(1);
2894 addr = new CastX2PNode(addr);
2895 addr = _gvn.transform(addr);
2896 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2897 flush = _gvn.transform(flush);
2898 set_memory(flush, TypeRawPtr::BOTTOM);
2899 return true;
2900 }
2901
2902 //----------------------------inline_unsafe_writeback0-------------------------
2903 // public native void Unsafe.writeback0(long address)
2904 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2905 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2906 return false;
2907 }
2908 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2909 return false;
2910 }
2911 #ifndef PRODUCT
2912 assert(Matcher::has_match_rule(Op_CacheWB),
2913 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2914 : "found match rule for CacheWBPostSync but not CacheWB"));
2915
2916 #endif
2917 null_check_receiver(); // null-check, then ignore
2918 Node *sync;
2919 if (is_pre) {
2920 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2921 } else {
2922 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2923 }
2924 sync = _gvn.transform(sync);
2925 set_memory(sync, TypeRawPtr::BOTTOM);
2926 return true;
2927 }
2928
2929 //----------------------------inline_unsafe_allocate---------------------------
2930 // public native Object Unsafe.allocateInstance(Class<?> cls);
2931 bool LibraryCallKit::inline_unsafe_allocate() {
2932
2933 #if INCLUDE_JVMTI
2934 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2935 return false;
2936 }
2937 #endif //INCLUDE_JVMTI
2938
2939 if (callee()->is_static()) return false; // caller must have the capability!
2940
2941 null_check_receiver(); // null-check, then ignore
2942 Node* cls = null_check(argument(1));
2943 if (stopped()) return true;
2944
2945 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2946 kls = null_check(kls);
2947 if (stopped()) return true; // argument was like int.class
2948
2949 #if INCLUDE_JVMTI
2950 // Don't try to access new allocated obj in the intrinsic.
2951 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2952 // Deoptimize and allocate in interpreter instead.
2953 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2954 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2955 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2956 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2957 {
2958 BuildCutout unless(this, tst, PROB_MAX);
2959 uncommon_trap(Deoptimization::Reason_intrinsic,
2960 Deoptimization::Action_make_not_entrant);
2961 }
2962 if (stopped()) {
2963 return true;
2964 }
2965 #endif //INCLUDE_JVMTI
2966
2967 Node* test = nullptr;
2968 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2969 // Note: The argument might still be an illegal value like
2970 // Serializable.class or Object[].class. The runtime will handle it.
2971 // But we must make an explicit check for initialization.
2972 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2973 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2974 // can generate code to load it as unsigned byte.
2975 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2976 Node* bits = intcon(InstanceKlass::fully_initialized);
2977 test = _gvn.transform(new SubINode(inst, bits));
2978 // The 'test' is non-zero if we need to take a slow path.
2979 }
2980
2981 Node* obj = new_instance(kls, test);
2982 set_result(obj);
2983 return true;
2984 }
2985
2986 //------------------------inline_native_time_funcs--------------
2987 // inline code for System.currentTimeMillis() and System.nanoTime()
2988 // these have the same type and signature
2989 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2990 const TypeFunc* tf = OptoRuntime::void_long_Type();
2991 const TypePtr* no_memory_effects = nullptr;
2992 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2993 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2994 #ifdef ASSERT
2995 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2996 assert(value_top == top(), "second value must be top");
2997 #endif
2998 set_result(value);
2999 return true;
3000 }
3001
3002
3003 #if INCLUDE_JVMTI
3004
3005 // When notifications are disabled then just update the VTMS transition bit and return.
3006 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3007 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3008 if (!DoJVMTIVirtualThreadTransitions) {
3009 return true;
3010 }
3011 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3012 IdealKit ideal(this);
3013
3014 Node* ONE = ideal.ConI(1);
3015 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3016 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3017 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3018
3019 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3020 sync_kit(ideal);
3021 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3022 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3023 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3024 ideal.sync_kit(this);
3025 } ideal.else_(); {
3026 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3027 Node* thread = ideal.thread();
3028 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3029 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3030
3031 sync_kit(ideal);
3032 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3033 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3034
3035 ideal.sync_kit(this);
3036 } ideal.end_if();
3037 final_sync(ideal);
3038
3039 return true;
3040 }
3041
3042 // Always update the is_disable_suspend bit.
3043 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3044 if (!DoJVMTIVirtualThreadTransitions) {
3045 return true;
3046 }
3047 IdealKit ideal(this);
3048
3049 {
3050 // unconditionally update the is_disable_suspend bit in current JavaThread
3051 Node* thread = ideal.thread();
3052 Node* arg = _gvn.transform(argument(0)); // argument for notification
3053 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3054 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3055
3056 sync_kit(ideal);
3057 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3058 ideal.sync_kit(this);
3059 }
3060 final_sync(ideal);
3061
3062 return true;
3063 }
3064
3065 #endif // INCLUDE_JVMTI
3066
3067 #ifdef JFR_HAVE_INTRINSICS
3068
3069 /**
3070 * if oop->klass != null
3071 * // normal class
3072 * epoch = _epoch_state ? 2 : 1
3073 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3074 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3075 * }
3076 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3077 * else
3078 * // primitive class
3079 * if oop->array_klass != null
3080 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3081 * else
3082 * id = LAST_TYPE_ID + 1 // void class path
3083 * if (!signaled)
3084 * signaled = true
3085 */
3086 bool LibraryCallKit::inline_native_classID() {
3087 Node* cls = argument(0);
3088
3089 IdealKit ideal(this);
3090 #define __ ideal.
3091 IdealVariable result(ideal); __ declarations_done();
3092 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3093 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3094 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3095
3096
3097 __ if_then(kls, BoolTest::ne, null()); {
3098 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3099 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3100
3101 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3102 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3103 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3104 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3105 mask = _gvn.transform(new OrLNode(mask, epoch));
3106 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3107
3108 float unlikely = PROB_UNLIKELY(0.999);
3109 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3110 sync_kit(ideal);
3111 make_runtime_call(RC_LEAF,
3112 OptoRuntime::class_id_load_barrier_Type(),
3113 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3114 "class id load barrier",
3115 TypePtr::BOTTOM,
3116 kls);
3117 ideal.sync_kit(this);
3118 } __ end_if();
3119
3120 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3121 } __ else_(); {
3122 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3123 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3124 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3125 __ if_then(array_kls, BoolTest::ne, null()); {
3126 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3127 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3128 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3129 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3130 } __ else_(); {
3131 // void class case
3132 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3133 } __ end_if();
3134
3135 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3136 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3137 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3138 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3139 } __ end_if();
3140 } __ end_if();
3141
3142 final_sync(ideal);
3143 set_result(ideal.value(result));
3144 #undef __
3145 return true;
3146 }
3147
3148 //------------------------inline_native_jvm_commit------------------
3149 bool LibraryCallKit::inline_native_jvm_commit() {
3150 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3151
3152 // Save input memory and i_o state.
3153 Node* input_memory_state = reset_memory();
3154 set_all_memory(input_memory_state);
3155 Node* input_io_state = i_o();
3156
3157 // TLS.
3158 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3159 // Jfr java buffer.
3160 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3161 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3162 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3163
3164 // Load the current value of the notified field in the JfrThreadLocal.
3165 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3166 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3167
3168 // Test for notification.
3169 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3170 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3171 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3172
3173 // True branch, is notified.
3174 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3175 set_control(is_notified);
3176
3177 // Reset notified state.
3178 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3179 Node* notified_reset_memory = reset_memory();
3180
3181 // 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.
3182 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3183 // Convert the machine-word to a long.
3184 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3185
3186 // False branch, not notified.
3187 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3188 set_control(not_notified);
3189 set_all_memory(input_memory_state);
3190
3191 // Arg is the next position as a long.
3192 Node* arg = argument(0);
3193 // Convert long to machine-word.
3194 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3195
3196 // Store the next_position to the underlying jfr java buffer.
3197 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3198
3199 Node* commit_memory = reset_memory();
3200 set_all_memory(commit_memory);
3201
3202 // 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.
3203 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3204 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3205 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3206
3207 // And flags with lease constant.
3208 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3209
3210 // Branch on lease to conditionalize returning the leased java buffer.
3211 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3212 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3213 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3214
3215 // False branch, not a lease.
3216 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3217
3218 // True branch, is lease.
3219 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3220 set_control(is_lease);
3221
3222 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3223 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3224 OptoRuntime::void_void_Type(),
3225 SharedRuntime::jfr_return_lease(),
3226 "return_lease", TypePtr::BOTTOM);
3227 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3228
3229 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3230 record_for_igvn(lease_compare_rgn);
3231 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3232 record_for_igvn(lease_compare_mem);
3233 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3234 record_for_igvn(lease_compare_io);
3235 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3236 record_for_igvn(lease_result_value);
3237
3238 // Update control and phi nodes.
3239 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3240 lease_compare_rgn->init_req(_false_path, not_lease);
3241
3242 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3243 lease_compare_mem->init_req(_false_path, commit_memory);
3244
3245 lease_compare_io->init_req(_true_path, i_o());
3246 lease_compare_io->init_req(_false_path, input_io_state);
3247
3248 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3249 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3250
3251 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3252 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3253 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3254 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3255
3256 // Update control and phi nodes.
3257 result_rgn->init_req(_true_path, is_notified);
3258 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3259
3260 result_mem->init_req(_true_path, notified_reset_memory);
3261 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3262
3263 result_io->init_req(_true_path, input_io_state);
3264 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3265
3266 result_value->init_req(_true_path, current_pos);
3267 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3268
3269 // Set output state.
3270 set_control(_gvn.transform(result_rgn));
3271 set_all_memory(_gvn.transform(result_mem));
3272 set_i_o(_gvn.transform(result_io));
3273 set_result(result_rgn, result_value);
3274 return true;
3275 }
3276
3277 /*
3278 * The intrinsic is a model of this pseudo-code:
3279 *
3280 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3281 * jobject h_event_writer = tl->java_event_writer();
3282 * if (h_event_writer == nullptr) {
3283 * return nullptr;
3284 * }
3285 * oop threadObj = Thread::threadObj();
3286 * oop vthread = java_lang_Thread::vthread(threadObj);
3287 * traceid tid;
3288 * bool pinVirtualThread;
3289 * bool excluded;
3290 * if (vthread != threadObj) { // i.e. current thread is virtual
3291 * tid = java_lang_Thread::tid(vthread);
3292 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3293 * pinVirtualThread = VMContinuations;
3294 * excluded = vthread_epoch_raw & excluded_mask;
3295 * if (!excluded) {
3296 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3297 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3298 * if (vthread_epoch != current_epoch) {
3299 * write_checkpoint();
3300 * }
3301 * }
3302 * } else {
3303 * tid = java_lang_Thread::tid(threadObj);
3304 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3305 * pinVirtualThread = false;
3306 * excluded = thread_epoch_raw & excluded_mask;
3307 * }
3308 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3309 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3310 * if (tid_in_event_writer != tid) {
3311 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3312 * setField(event_writer, "excluded", excluded);
3313 * setField(event_writer, "threadID", tid);
3314 * }
3315 * return event_writer
3316 */
3317 bool LibraryCallKit::inline_native_getEventWriter() {
3318 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3319
3320 // Save input memory and i_o state.
3321 Node* input_memory_state = reset_memory();
3322 set_all_memory(input_memory_state);
3323 Node* input_io_state = i_o();
3324
3325 // The most significant bit of the u2 is used to denote thread exclusion
3326 Node* excluded_shift = _gvn.intcon(15);
3327 Node* excluded_mask = _gvn.intcon(1 << 15);
3328 // The epoch generation is the range [1-32767]
3329 Node* epoch_mask = _gvn.intcon(32767);
3330
3331 // TLS
3332 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3333
3334 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3335 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3336
3337 // Load the eventwriter jobject handle.
3338 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3339
3340 // Null check the jobject handle.
3341 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3342 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3343 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3344
3345 // False path, jobj is null.
3346 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3347
3348 // True path, jobj is not null.
3349 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3350
3351 set_control(jobj_is_not_null);
3352
3353 // Load the threadObj for the CarrierThread.
3354 Node* threadObj = generate_current_thread(tls_ptr);
3355
3356 // Load the vthread.
3357 Node* vthread = generate_virtual_thread(tls_ptr);
3358
3359 // If vthread != threadObj, this is a virtual thread.
3360 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3361 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3362 IfNode* iff_vthread_not_equal_threadObj =
3363 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3364
3365 // False branch, fallback to threadObj.
3366 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3367 set_control(vthread_equal_threadObj);
3368
3369 // Load the tid field from the vthread object.
3370 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3371
3372 // Load the raw epoch value from the threadObj.
3373 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3374 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3375 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3376 TypeInt::CHAR, T_CHAR,
3377 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3378
3379 // Mask off the excluded information from the epoch.
3380 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3381
3382 // True branch, this is a virtual thread.
3383 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3384 set_control(vthread_not_equal_threadObj);
3385
3386 // Load the tid field from the vthread object.
3387 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3388
3389 // Continuation support determines if a virtual thread should be pinned.
3390 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3391 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3392
3393 // Load the raw epoch value from the vthread.
3394 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3395 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3396 TypeInt::CHAR, T_CHAR,
3397 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3398
3399 // Mask off the excluded information from the epoch.
3400 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3401
3402 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3403 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3404 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3405 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3406
3407 // False branch, vthread is excluded, no need to write epoch info.
3408 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3409
3410 // True branch, vthread is included, update epoch info.
3411 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3412 set_control(included);
3413
3414 // Get epoch value.
3415 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3416
3417 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3418 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3419 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3420
3421 // Compare the epoch in the vthread to the current epoch generation.
3422 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3423 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3424 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3425
3426 // False path, epoch is equal, checkpoint information is valid.
3427 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3428
3429 // True path, epoch is not equal, write a checkpoint for the vthread.
3430 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3431
3432 set_control(epoch_is_not_equal);
3433
3434 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3435 // The call also updates the native thread local thread id and the vthread with the current epoch.
3436 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3437 OptoRuntime::jfr_write_checkpoint_Type(),
3438 SharedRuntime::jfr_write_checkpoint(),
3439 "write_checkpoint", TypePtr::BOTTOM);
3440 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3441
3442 // vthread epoch != current epoch
3443 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3444 record_for_igvn(epoch_compare_rgn);
3445 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3446 record_for_igvn(epoch_compare_mem);
3447 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3448 record_for_igvn(epoch_compare_io);
3449
3450 // Update control and phi nodes.
3451 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3452 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3453 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3454 epoch_compare_mem->init_req(_false_path, input_memory_state);
3455 epoch_compare_io->init_req(_true_path, i_o());
3456 epoch_compare_io->init_req(_false_path, input_io_state);
3457
3458 // excluded != true
3459 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3460 record_for_igvn(exclude_compare_rgn);
3461 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3462 record_for_igvn(exclude_compare_mem);
3463 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3464 record_for_igvn(exclude_compare_io);
3465
3466 // Update control and phi nodes.
3467 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3468 exclude_compare_rgn->init_req(_false_path, excluded);
3469 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3470 exclude_compare_mem->init_req(_false_path, input_memory_state);
3471 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3472 exclude_compare_io->init_req(_false_path, input_io_state);
3473
3474 // vthread != threadObj
3475 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3476 record_for_igvn(vthread_compare_rgn);
3477 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3478 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3479 record_for_igvn(vthread_compare_io);
3480 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3481 record_for_igvn(tid);
3482 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3483 record_for_igvn(exclusion);
3484 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3485 record_for_igvn(pinVirtualThread);
3486
3487 // Update control and phi nodes.
3488 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3489 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3490 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3491 vthread_compare_mem->init_req(_false_path, input_memory_state);
3492 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3493 vthread_compare_io->init_req(_false_path, input_io_state);
3494 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3495 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3496 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3497 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3498 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3499 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3500
3501 // Update branch state.
3502 set_control(_gvn.transform(vthread_compare_rgn));
3503 set_all_memory(_gvn.transform(vthread_compare_mem));
3504 set_i_o(_gvn.transform(vthread_compare_io));
3505
3506 // Load the event writer oop by dereferencing the jobject handle.
3507 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3508 assert(klass_EventWriter->is_loaded(), "invariant");
3509 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3510 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3511 const TypeOopPtr* const xtype = aklass->as_instance_type();
3512 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3513 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3514
3515 // Load the current thread id from the event writer object.
3516 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3517 // Get the field offset to, conditionally, store an updated tid value later.
3518 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3519 // Get the field offset to, conditionally, store an updated exclusion value later.
3520 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3521 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3522 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3523
3524 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3525 record_for_igvn(event_writer_tid_compare_rgn);
3526 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3527 record_for_igvn(event_writer_tid_compare_mem);
3528 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3529 record_for_igvn(event_writer_tid_compare_io);
3530
3531 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3532 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3533 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3534 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3535
3536 // False path, tids are the same.
3537 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3538
3539 // True path, tid is not equal, need to update the tid in the event writer.
3540 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3541 record_for_igvn(tid_is_not_equal);
3542
3543 // Store the pin state to the event writer.
3544 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3545
3546 // Store the exclusion state to the event writer.
3547 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3548 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3549
3550 // Store the tid to the event writer.
3551 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3552
3553 // Update control and phi nodes.
3554 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3555 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3556 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3557 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3558 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3559 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3560
3561 // Result of top level CFG, Memory, IO and Value.
3562 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3563 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3564 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3565 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3566
3567 // Result control.
3568 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3569 result_rgn->init_req(_false_path, jobj_is_null);
3570
3571 // Result memory.
3572 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3573 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3574
3575 // Result IO.
3576 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3577 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3578
3579 // Result value.
3580 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3581 result_value->init_req(_false_path, null()); // return null
3582
3583 // Set output state.
3584 set_control(_gvn.transform(result_rgn));
3585 set_all_memory(_gvn.transform(result_mem));
3586 set_i_o(_gvn.transform(result_io));
3587 set_result(result_rgn, result_value);
3588 return true;
3589 }
3590
3591 /*
3592 * The intrinsic is a model of this pseudo-code:
3593 *
3594 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3595 * if (carrierThread != thread) { // is virtual thread
3596 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3597 * bool excluded = vthread_epoch_raw & excluded_mask;
3598 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3599 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3600 * if (!excluded) {
3601 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3602 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3603 * }
3604 * Atomic::release_store(&tl->_vthread, true);
3605 * return;
3606 * }
3607 * Atomic::release_store(&tl->_vthread, false);
3608 */
3609 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3610 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3611
3612 Node* input_memory_state = reset_memory();
3613 set_all_memory(input_memory_state);
3614
3615 // The most significant bit of the u2 is used to denote thread exclusion
3616 Node* excluded_mask = _gvn.intcon(1 << 15);
3617 // The epoch generation is the range [1-32767]
3618 Node* epoch_mask = _gvn.intcon(32767);
3619
3620 Node* const carrierThread = generate_current_thread(jt);
3621 // If thread != carrierThread, this is a virtual thread.
3622 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3623 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3624 IfNode* iff_thread_not_equal_carrierThread =
3625 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3626
3627 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3628
3629 // False branch, is carrierThread.
3630 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3631 // Store release
3632 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3633
3634 set_all_memory(input_memory_state);
3635
3636 // True branch, is virtual thread.
3637 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3638 set_control(thread_not_equal_carrierThread);
3639
3640 // Load the raw epoch value from the vthread.
3641 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3642 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3643 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3644
3645 // Mask off the excluded information from the epoch.
3646 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3647
3648 // Load the tid field from the thread.
3649 Node* tid = load_field_from_object(thread, "tid", "J");
3650
3651 // Store the vthread tid to the jfr thread local.
3652 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3653 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3654
3655 // Branch is_excluded to conditionalize updating the epoch .
3656 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3657 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3658 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3659
3660 // True branch, vthread is excluded, no need to write epoch info.
3661 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3662 set_control(excluded);
3663 Node* vthread_is_excluded = _gvn.intcon(1);
3664
3665 // False branch, vthread is included, update epoch info.
3666 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3667 set_control(included);
3668 Node* vthread_is_included = _gvn.intcon(0);
3669
3670 // Get epoch value.
3671 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3672
3673 // Store the vthread epoch to the jfr thread local.
3674 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3675 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3676
3677 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3678 record_for_igvn(excluded_rgn);
3679 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3680 record_for_igvn(excluded_mem);
3681 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3682 record_for_igvn(exclusion);
3683
3684 // Merge the excluded control and memory.
3685 excluded_rgn->init_req(_true_path, excluded);
3686 excluded_rgn->init_req(_false_path, included);
3687 excluded_mem->init_req(_true_path, tid_memory);
3688 excluded_mem->init_req(_false_path, included_memory);
3689 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3690 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3691
3692 // Set intermediate state.
3693 set_control(_gvn.transform(excluded_rgn));
3694 set_all_memory(excluded_mem);
3695
3696 // Store the vthread exclusion state to the jfr thread local.
3697 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3698 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3699
3700 // Store release
3701 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3702
3703 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3704 record_for_igvn(thread_compare_rgn);
3705 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3706 record_for_igvn(thread_compare_mem);
3707 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3708 record_for_igvn(vthread);
3709
3710 // Merge the thread_compare control and memory.
3711 thread_compare_rgn->init_req(_true_path, control());
3712 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3713 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3714 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3715
3716 // Set output state.
3717 set_control(_gvn.transform(thread_compare_rgn));
3718 set_all_memory(_gvn.transform(thread_compare_mem));
3719 }
3720
3721 #endif // JFR_HAVE_INTRINSICS
3722
3723 //------------------------inline_native_currentCarrierThread------------------
3724 bool LibraryCallKit::inline_native_currentCarrierThread() {
3725 Node* junk = nullptr;
3726 set_result(generate_current_thread(junk));
3727 return true;
3728 }
3729
3730 //------------------------inline_native_currentThread------------------
3731 bool LibraryCallKit::inline_native_currentThread() {
3732 Node* junk = nullptr;
3733 set_result(generate_virtual_thread(junk));
3734 return true;
3735 }
3736
3737 //------------------------inline_native_setVthread------------------
3738 bool LibraryCallKit::inline_native_setCurrentThread() {
3739 assert(C->method()->changes_current_thread(),
3740 "method changes current Thread but is not annotated ChangesCurrentThread");
3741 Node* arr = argument(1);
3742 Node* thread = _gvn.transform(new ThreadLocalNode());
3743 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3744 Node* thread_obj_handle
3745 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3746 thread_obj_handle = _gvn.transform(thread_obj_handle);
3747 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3748 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3749
3750 // Change the _monitor_owner_id of the JavaThread
3751 Node* tid = load_field_from_object(arr, "tid", "J");
3752 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3753 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3754
3755 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3756 return true;
3757 }
3758
3759 const Type* LibraryCallKit::scopedValueCache_type() {
3760 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3761 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3762 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3763
3764 // Because we create the scopedValue cache lazily we have to make the
3765 // type of the result BotPTR.
3766 bool xk = etype->klass_is_exact();
3767 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3768 return objects_type;
3769 }
3770
3771 Node* LibraryCallKit::scopedValueCache_helper() {
3772 Node* thread = _gvn.transform(new ThreadLocalNode());
3773 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3774 // We cannot use immutable_memory() because we might flip onto a
3775 // different carrier thread, at which point we'll need to use that
3776 // carrier thread's cache.
3777 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3778 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3779 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3780 }
3781
3782 //------------------------inline_native_scopedValueCache------------------
3783 bool LibraryCallKit::inline_native_scopedValueCache() {
3784 Node* cache_obj_handle = scopedValueCache_helper();
3785 const Type* objects_type = scopedValueCache_type();
3786 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3787
3788 return true;
3789 }
3790
3791 //------------------------inline_native_setScopedValueCache------------------
3792 bool LibraryCallKit::inline_native_setScopedValueCache() {
3793 Node* arr = argument(0);
3794 Node* cache_obj_handle = scopedValueCache_helper();
3795 const Type* objects_type = scopedValueCache_type();
3796
3797 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3798 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3799
3800 return true;
3801 }
3802
3803 //------------------------inline_native_Continuation_pin and unpin-----------
3804
3805 // Shared implementation routine for both pin and unpin.
3806 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3807 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3808
3809 // Save input memory.
3810 Node* input_memory_state = reset_memory();
3811 set_all_memory(input_memory_state);
3812
3813 // TLS
3814 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3815 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3816 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3817
3818 // Null check the last continuation object.
3819 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3820 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3821 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3822
3823 // False path, last continuation is null.
3824 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3825
3826 // True path, last continuation is not null.
3827 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3828
3829 set_control(continuation_is_not_null);
3830
3831 // Load the pin count from the last continuation.
3832 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3833 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3834
3835 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3836 Node* pin_count_rhs;
3837 if (unpin) {
3838 pin_count_rhs = _gvn.intcon(0);
3839 } else {
3840 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3841 }
3842 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3843 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3844 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3845
3846 // True branch, pin count over/underflow.
3847 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3848 {
3849 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3850 // which will throw IllegalStateException for pin count over/underflow.
3851 // No memory changed so far - we can use memory create by reset_memory()
3852 // at the beginning of this intrinsic. No need to call reset_memory() again.
3853 PreserveJVMState pjvms(this);
3854 set_control(pin_count_over_underflow);
3855 uncommon_trap(Deoptimization::Reason_intrinsic,
3856 Deoptimization::Action_none);
3857 assert(stopped(), "invariant");
3858 }
3859
3860 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3861 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3862 set_control(valid_pin_count);
3863
3864 Node* next_pin_count;
3865 if (unpin) {
3866 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3867 } else {
3868 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3869 }
3870
3871 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3872
3873 // Result of top level CFG and Memory.
3874 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3875 record_for_igvn(result_rgn);
3876 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3877 record_for_igvn(result_mem);
3878
3879 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3880 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3881 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3882 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3883
3884 // Set output state.
3885 set_control(_gvn.transform(result_rgn));
3886 set_all_memory(_gvn.transform(result_mem));
3887
3888 return true;
3889 }
3890
3891 //---------------------------load_mirror_from_klass----------------------------
3892 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3893 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3894 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3895 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3896 // mirror = ((OopHandle)mirror)->resolve();
3897 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3898 }
3899
3900 //-----------------------load_klass_from_mirror_common-------------------------
3901 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3902 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3903 // and branch to the given path on the region.
3904 // If never_see_null, take an uncommon trap on null, so we can optimistically
3905 // compile for the non-null case.
3906 // If the region is null, force never_see_null = true.
3907 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3908 bool never_see_null,
3909 RegionNode* region,
3910 int null_path,
3911 int offset) {
3912 if (region == nullptr) never_see_null = true;
3913 Node* p = basic_plus_adr(mirror, offset);
3914 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3915 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3916 Node* null_ctl = top();
3917 kls = null_check_oop(kls, &null_ctl, never_see_null);
3918 if (region != nullptr) {
3919 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3920 region->init_req(null_path, null_ctl);
3921 } else {
3922 assert(null_ctl == top(), "no loose ends");
3923 }
3924 return kls;
3925 }
3926
3927 //--------------------(inline_native_Class_query helpers)---------------------
3928 // Use this for JVM_ACC_INTERFACE.
3929 // Fall through if (mods & mask) == bits, take the guard otherwise.
3930 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3931 ByteSize offset, const Type* type, BasicType bt) {
3932 // Branch around if the given klass has the given modifier bit set.
3933 // Like generate_guard, adds a new path onto the region.
3934 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3935 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3936 Node* mask = intcon(modifier_mask);
3937 Node* bits = intcon(modifier_bits);
3938 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3939 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3940 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3941 return generate_fair_guard(bol, region);
3942 }
3943 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3944 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3945 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3946 }
3947
3948 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3949 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3950 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3951 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3952 }
3953
3954 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3955 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3956 }
3957
3958 //-------------------------inline_native_Class_query-------------------
3959 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3960 const Type* return_type = TypeInt::BOOL;
3961 Node* prim_return_value = top(); // what happens if it's a primitive class?
3962 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3963 bool expect_prim = false; // most of these guys expect to work on refs
3964
3965 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3966
3967 Node* mirror = argument(0);
3968 Node* obj = top();
3969
3970 switch (id) {
3971 case vmIntrinsics::_isInstance:
3972 // nothing is an instance of a primitive type
3973 prim_return_value = intcon(0);
3974 obj = argument(1);
3975 break;
3976 case vmIntrinsics::_isHidden:
3977 prim_return_value = intcon(0);
3978 break;
3979 case vmIntrinsics::_getSuperclass:
3980 prim_return_value = null();
3981 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3982 break;
3983 case vmIntrinsics::_getClassAccessFlags:
3984 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3985 return_type = TypeInt::CHAR;
3986 break;
3987 default:
3988 fatal_unexpected_iid(id);
3989 break;
3990 }
3991
3992 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3993 if (mirror_con == nullptr) return false; // cannot happen?
3994
3995 #ifndef PRODUCT
3996 if (C->print_intrinsics() || C->print_inlining()) {
3997 ciType* k = mirror_con->java_mirror_type();
3998 if (k) {
3999 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4000 k->print_name();
4001 tty->cr();
4002 }
4003 }
4004 #endif
4005
4006 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4007 RegionNode* region = new RegionNode(PATH_LIMIT);
4008 record_for_igvn(region);
4009 PhiNode* phi = new PhiNode(region, return_type);
4010
4011 // The mirror will never be null of Reflection.getClassAccessFlags, however
4012 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4013 // if it is. See bug 4774291.
4014
4015 // For Reflection.getClassAccessFlags(), the null check occurs in
4016 // the wrong place; see inline_unsafe_access(), above, for a similar
4017 // situation.
4018 mirror = null_check(mirror);
4019 // If mirror or obj is dead, only null-path is taken.
4020 if (stopped()) return true;
4021
4022 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4023
4024 // Now load the mirror's klass metaobject, and null-check it.
4025 // Side-effects region with the control path if the klass is null.
4026 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4027 // If kls is null, we have a primitive mirror.
4028 phi->init_req(_prim_path, prim_return_value);
4029 if (stopped()) { set_result(region, phi); return true; }
4030 bool safe_for_replace = (region->in(_prim_path) == top());
4031
4032 Node* p; // handy temp
4033 Node* null_ctl;
4034
4035 // Now that we have the non-null klass, we can perform the real query.
4036 // For constant classes, the query will constant-fold in LoadNode::Value.
4037 Node* query_value = top();
4038 switch (id) {
4039 case vmIntrinsics::_isInstance:
4040 // nothing is an instance of a primitive type
4041 query_value = gen_instanceof(obj, kls, safe_for_replace);
4042 break;
4043
4044 case vmIntrinsics::_isHidden:
4045 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4046 if (generate_hidden_class_guard(kls, region) != nullptr)
4047 // A guard was added. If the guard is taken, it was an hidden class.
4048 phi->add_req(intcon(1));
4049 // If we fall through, it's a plain class.
4050 query_value = intcon(0);
4051 break;
4052
4053
4054 case vmIntrinsics::_getSuperclass:
4055 // The rules here are somewhat unfortunate, but we can still do better
4056 // with random logic than with a JNI call.
4057 // Interfaces store null or Object as _super, but must report null.
4058 // Arrays store an intermediate super as _super, but must report Object.
4059 // Other types can report the actual _super.
4060 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4061 if (generate_interface_guard(kls, region) != nullptr)
4062 // A guard was added. If the guard is taken, it was an interface.
4063 phi->add_req(null());
4064 if (generate_array_guard(kls, region) != nullptr)
4065 // A guard was added. If the guard is taken, it was an array.
4066 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4067 // If we fall through, it's a plain class. Get its _super.
4068 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4069 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4070 null_ctl = top();
4071 kls = null_check_oop(kls, &null_ctl);
4072 if (null_ctl != top()) {
4073 // If the guard is taken, Object.superClass is null (both klass and mirror).
4074 region->add_req(null_ctl);
4075 phi ->add_req(null());
4076 }
4077 if (!stopped()) {
4078 query_value = load_mirror_from_klass(kls);
4079 }
4080 break;
4081
4082 case vmIntrinsics::_getClassAccessFlags:
4083 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4084 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4085 break;
4086
4087 default:
4088 fatal_unexpected_iid(id);
4089 break;
4090 }
4091
4092 // Fall-through is the normal case of a query to a real class.
4093 phi->init_req(1, query_value);
4094 region->init_req(1, control());
4095
4096 C->set_has_split_ifs(true); // Has chance for split-if optimization
4097 set_result(region, phi);
4098 return true;
4099 }
4100
4101 //-------------------------inline_Class_cast-------------------
4102 bool LibraryCallKit::inline_Class_cast() {
4103 Node* mirror = argument(0); // Class
4104 Node* obj = argument(1);
4105 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4106 if (mirror_con == nullptr) {
4107 return false; // dead path (mirror->is_top()).
4108 }
4109 if (obj == nullptr || obj->is_top()) {
4110 return false; // dead path
4111 }
4112 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4113
4114 // First, see if Class.cast() can be folded statically.
4115 // java_mirror_type() returns non-null for compile-time Class constants.
4116 ciType* tm = mirror_con->java_mirror_type();
4117 if (tm != nullptr && tm->is_klass() &&
4118 tp != nullptr) {
4119 if (!tp->is_loaded()) {
4120 // Don't use intrinsic when class is not loaded.
4121 return false;
4122 } else {
4123 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4124 if (static_res == Compile::SSC_always_true) {
4125 // isInstance() is true - fold the code.
4126 set_result(obj);
4127 return true;
4128 } else if (static_res == Compile::SSC_always_false) {
4129 // Don't use intrinsic, have to throw ClassCastException.
4130 // If the reference is null, the non-intrinsic bytecode will
4131 // be optimized appropriately.
4132 return false;
4133 }
4134 }
4135 }
4136
4137 // Bailout intrinsic and do normal inlining if exception path is frequent.
4138 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4139 return false;
4140 }
4141
4142 // Generate dynamic checks.
4143 // Class.cast() is java implementation of _checkcast bytecode.
4144 // Do checkcast (Parse::do_checkcast()) optimizations here.
4145
4146 mirror = null_check(mirror);
4147 // If mirror is dead, only null-path is taken.
4148 if (stopped()) {
4149 return true;
4150 }
4151
4152 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4153 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4154 RegionNode* region = new RegionNode(PATH_LIMIT);
4155 record_for_igvn(region);
4156
4157 // Now load the mirror's klass metaobject, and null-check it.
4158 // If kls is null, we have a primitive mirror and
4159 // nothing is an instance of a primitive type.
4160 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4161
4162 Node* res = top();
4163 if (!stopped()) {
4164 Node* bad_type_ctrl = top();
4165 // Do checkcast optimizations.
4166 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4167 region->init_req(_bad_type_path, bad_type_ctrl);
4168 }
4169 if (region->in(_prim_path) != top() ||
4170 region->in(_bad_type_path) != top()) {
4171 // Let Interpreter throw ClassCastException.
4172 PreserveJVMState pjvms(this);
4173 set_control(_gvn.transform(region));
4174 uncommon_trap(Deoptimization::Reason_intrinsic,
4175 Deoptimization::Action_maybe_recompile);
4176 }
4177 if (!stopped()) {
4178 set_result(res);
4179 }
4180 return true;
4181 }
4182
4183
4184 //--------------------------inline_native_subtype_check------------------------
4185 // This intrinsic takes the JNI calls out of the heart of
4186 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4187 bool LibraryCallKit::inline_native_subtype_check() {
4188 // Pull both arguments off the stack.
4189 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4190 args[0] = argument(0);
4191 args[1] = argument(1);
4192 Node* klasses[2]; // corresponding Klasses: superk, subk
4193 klasses[0] = klasses[1] = top();
4194
4195 enum {
4196 // A full decision tree on {superc is prim, subc is prim}:
4197 _prim_0_path = 1, // {P,N} => false
4198 // {P,P} & superc!=subc => false
4199 _prim_same_path, // {P,P} & superc==subc => true
4200 _prim_1_path, // {N,P} => false
4201 _ref_subtype_path, // {N,N} & subtype check wins => true
4202 _both_ref_path, // {N,N} & subtype check loses => false
4203 PATH_LIMIT
4204 };
4205
4206 RegionNode* region = new RegionNode(PATH_LIMIT);
4207 Node* phi = new PhiNode(region, TypeInt::BOOL);
4208 record_for_igvn(region);
4209
4210 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4211 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4212 int class_klass_offset = java_lang_Class::klass_offset();
4213
4214 // First null-check both mirrors and load each mirror's klass metaobject.
4215 int which_arg;
4216 for (which_arg = 0; which_arg <= 1; which_arg++) {
4217 Node* arg = args[which_arg];
4218 arg = null_check(arg);
4219 if (stopped()) break;
4220 args[which_arg] = arg;
4221
4222 Node* p = basic_plus_adr(arg, class_klass_offset);
4223 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4224 klasses[which_arg] = _gvn.transform(kls);
4225 }
4226
4227 // Having loaded both klasses, test each for null.
4228 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4229 for (which_arg = 0; which_arg <= 1; which_arg++) {
4230 Node* kls = klasses[which_arg];
4231 Node* null_ctl = top();
4232 kls = null_check_oop(kls, &null_ctl, never_see_null);
4233 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4234 region->init_req(prim_path, null_ctl);
4235 if (stopped()) break;
4236 klasses[which_arg] = kls;
4237 }
4238
4239 if (!stopped()) {
4240 // now we have two reference types, in klasses[0..1]
4241 Node* subk = klasses[1]; // the argument to isAssignableFrom
4242 Node* superk = klasses[0]; // the receiver
4243 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4244 // now we have a successful reference subtype check
4245 region->set_req(_ref_subtype_path, control());
4246 }
4247
4248 // If both operands are primitive (both klasses null), then
4249 // we must return true when they are identical primitives.
4250 // It is convenient to test this after the first null klass check.
4251 set_control(region->in(_prim_0_path)); // go back to first null check
4252 if (!stopped()) {
4253 // Since superc is primitive, make a guard for the superc==subc case.
4254 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4255 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4256 generate_guard(bol_eq, region, PROB_FAIR);
4257 if (region->req() == PATH_LIMIT+1) {
4258 // A guard was added. If the added guard is taken, superc==subc.
4259 region->swap_edges(PATH_LIMIT, _prim_same_path);
4260 region->del_req(PATH_LIMIT);
4261 }
4262 region->set_req(_prim_0_path, control()); // Not equal after all.
4263 }
4264
4265 // these are the only paths that produce 'true':
4266 phi->set_req(_prim_same_path, intcon(1));
4267 phi->set_req(_ref_subtype_path, intcon(1));
4268
4269 // pull together the cases:
4270 assert(region->req() == PATH_LIMIT, "sane region");
4271 for (uint i = 1; i < region->req(); i++) {
4272 Node* ctl = region->in(i);
4273 if (ctl == nullptr || ctl == top()) {
4274 region->set_req(i, top());
4275 phi ->set_req(i, top());
4276 } else if (phi->in(i) == nullptr) {
4277 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4278 }
4279 }
4280
4281 set_control(_gvn.transform(region));
4282 set_result(_gvn.transform(phi));
4283 return true;
4284 }
4285
4286 //---------------------generate_array_guard_common------------------------
4287 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4288 bool obj_array, bool not_array, Node** obj) {
4289
4290 if (stopped()) {
4291 return nullptr;
4292 }
4293
4294 // If obj_array/non_array==false/false:
4295 // Branch around if the given klass is in fact an array (either obj or prim).
4296 // If obj_array/non_array==false/true:
4297 // Branch around if the given klass is not an array klass of any kind.
4298 // If obj_array/non_array==true/true:
4299 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4300 // If obj_array/non_array==true/false:
4301 // Branch around if the kls is an oop array (Object[] or subtype)
4302 //
4303 // Like generate_guard, adds a new path onto the region.
4304 jint layout_con = 0;
4305 Node* layout_val = get_layout_helper(kls, layout_con);
4306 if (layout_val == nullptr) {
4307 bool query = (obj_array
4308 ? Klass::layout_helper_is_objArray(layout_con)
4309 : Klass::layout_helper_is_array(layout_con));
4310 if (query == not_array) {
4311 return nullptr; // never a branch
4312 } else { // always a branch
4313 Node* always_branch = control();
4314 if (region != nullptr)
4315 region->add_req(always_branch);
4316 set_control(top());
4317 return always_branch;
4318 }
4319 }
4320 // Now test the correct condition.
4321 jint nval = (obj_array
4322 ? (jint)(Klass::_lh_array_tag_type_value
4323 << Klass::_lh_array_tag_shift)
4324 : Klass::_lh_neutral_value);
4325 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4326 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4327 // invert the test if we are looking for a non-array
4328 if (not_array) btest = BoolTest(btest).negate();
4329 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4330 Node* ctrl = generate_fair_guard(bol, region);
4331 Node* is_array_ctrl = not_array ? control() : ctrl;
4332 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4333 // Keep track of the fact that 'obj' is an array to prevent
4334 // array specific accesses from floating above the guard.
4335 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4336 }
4337 return ctrl;
4338 }
4339
4340
4341 //-----------------------inline_native_newArray--------------------------
4342 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4343 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4344 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4345 Node* mirror;
4346 Node* count_val;
4347 if (uninitialized) {
4348 null_check_receiver();
4349 mirror = argument(1);
4350 count_val = argument(2);
4351 } else {
4352 mirror = argument(0);
4353 count_val = argument(1);
4354 }
4355
4356 mirror = null_check(mirror);
4357 // If mirror or obj is dead, only null-path is taken.
4358 if (stopped()) return true;
4359
4360 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4361 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4362 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4363 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4364 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4365
4366 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4367 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4368 result_reg, _slow_path);
4369 Node* normal_ctl = control();
4370 Node* no_array_ctl = result_reg->in(_slow_path);
4371
4372 // Generate code for the slow case. We make a call to newArray().
4373 set_control(no_array_ctl);
4374 if (!stopped()) {
4375 // Either the input type is void.class, or else the
4376 // array klass has not yet been cached. Either the
4377 // ensuing call will throw an exception, or else it
4378 // will cache the array klass for next time.
4379 PreserveJVMState pjvms(this);
4380 CallJavaNode* slow_call = nullptr;
4381 if (uninitialized) {
4382 // Generate optimized virtual call (holder class 'Unsafe' is final)
4383 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4384 } else {
4385 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4386 }
4387 Node* slow_result = set_results_for_java_call(slow_call);
4388 // this->control() comes from set_results_for_java_call
4389 result_reg->set_req(_slow_path, control());
4390 result_val->set_req(_slow_path, slow_result);
4391 result_io ->set_req(_slow_path, i_o());
4392 result_mem->set_req(_slow_path, reset_memory());
4393 }
4394
4395 set_control(normal_ctl);
4396 if (!stopped()) {
4397 // Normal case: The array type has been cached in the java.lang.Class.
4398 // The following call works fine even if the array type is polymorphic.
4399 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4400 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4401 result_reg->init_req(_normal_path, control());
4402 result_val->init_req(_normal_path, obj);
4403 result_io ->init_req(_normal_path, i_o());
4404 result_mem->init_req(_normal_path, reset_memory());
4405
4406 if (uninitialized) {
4407 // Mark the allocation so that zeroing is skipped
4408 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4409 alloc->maybe_set_complete(&_gvn);
4410 }
4411 }
4412
4413 // Return the combined state.
4414 set_i_o( _gvn.transform(result_io) );
4415 set_all_memory( _gvn.transform(result_mem));
4416
4417 C->set_has_split_ifs(true); // Has chance for split-if optimization
4418 set_result(result_reg, result_val);
4419 return true;
4420 }
4421
4422 //----------------------inline_native_getLength--------------------------
4423 // public static native int java.lang.reflect.Array.getLength(Object array);
4424 bool LibraryCallKit::inline_native_getLength() {
4425 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4426
4427 Node* array = null_check(argument(0));
4428 // If array is dead, only null-path is taken.
4429 if (stopped()) return true;
4430
4431 // Deoptimize if it is a non-array.
4432 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4433
4434 if (non_array != nullptr) {
4435 PreserveJVMState pjvms(this);
4436 set_control(non_array);
4437 uncommon_trap(Deoptimization::Reason_intrinsic,
4438 Deoptimization::Action_maybe_recompile);
4439 }
4440
4441 // If control is dead, only non-array-path is taken.
4442 if (stopped()) return true;
4443
4444 // The works fine even if the array type is polymorphic.
4445 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4446 Node* result = load_array_length(array);
4447
4448 C->set_has_split_ifs(true); // Has chance for split-if optimization
4449 set_result(result);
4450 return true;
4451 }
4452
4453 //------------------------inline_array_copyOf----------------------------
4454 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4455 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4456 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4457 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4458
4459 // Get the arguments.
4460 Node* original = argument(0);
4461 Node* start = is_copyOfRange? argument(1): intcon(0);
4462 Node* end = is_copyOfRange? argument(2): argument(1);
4463 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4464
4465 Node* newcopy = nullptr;
4466
4467 // Set the original stack and the reexecute bit for the interpreter to reexecute
4468 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4469 { PreserveReexecuteState preexecs(this);
4470 jvms()->set_should_reexecute(true);
4471
4472 array_type_mirror = null_check(array_type_mirror);
4473 original = null_check(original);
4474
4475 // Check if a null path was taken unconditionally.
4476 if (stopped()) return true;
4477
4478 Node* orig_length = load_array_length(original);
4479
4480 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4481 klass_node = null_check(klass_node);
4482
4483 RegionNode* bailout = new RegionNode(1);
4484 record_for_igvn(bailout);
4485
4486 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4487 // Bail out if that is so.
4488 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4489 if (not_objArray != nullptr) {
4490 // Improve the klass node's type from the new optimistic assumption:
4491 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4492 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4493 Node* cast = new CastPPNode(control(), klass_node, akls);
4494 klass_node = _gvn.transform(cast);
4495 }
4496
4497 // Bail out if either start or end is negative.
4498 generate_negative_guard(start, bailout, &start);
4499 generate_negative_guard(end, bailout, &end);
4500
4501 Node* length = end;
4502 if (_gvn.type(start) != TypeInt::ZERO) {
4503 length = _gvn.transform(new SubINode(end, start));
4504 }
4505
4506 // Bail out if length is negative (i.e., if start > end).
4507 // Without this the new_array would throw
4508 // NegativeArraySizeException but IllegalArgumentException is what
4509 // should be thrown
4510 generate_negative_guard(length, bailout, &length);
4511
4512 // Bail out if start is larger than the original length
4513 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4514 generate_negative_guard(orig_tail, bailout, &orig_tail);
4515
4516 if (bailout->req() > 1) {
4517 PreserveJVMState pjvms(this);
4518 set_control(_gvn.transform(bailout));
4519 uncommon_trap(Deoptimization::Reason_intrinsic,
4520 Deoptimization::Action_maybe_recompile);
4521 }
4522
4523 if (!stopped()) {
4524 // How many elements will we copy from the original?
4525 // The answer is MinI(orig_tail, length).
4526 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4527
4528 // Generate a direct call to the right arraycopy function(s).
4529 // We know the copy is disjoint but we might not know if the
4530 // oop stores need checking.
4531 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4532 // This will fail a store-check if x contains any non-nulls.
4533
4534 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4535 // loads/stores but it is legal only if we're sure the
4536 // Arrays.copyOf would succeed. So we need all input arguments
4537 // to the copyOf to be validated, including that the copy to the
4538 // new array won't trigger an ArrayStoreException. That subtype
4539 // check can be optimized if we know something on the type of
4540 // the input array from type speculation.
4541 if (_gvn.type(klass_node)->singleton()) {
4542 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4543 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4544
4545 int test = C->static_subtype_check(superk, subk);
4546 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4547 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4548 if (t_original->speculative_type() != nullptr) {
4549 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4550 }
4551 }
4552 }
4553
4554 bool validated = false;
4555 // Reason_class_check rather than Reason_intrinsic because we
4556 // want to intrinsify even if this traps.
4557 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4558 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4559
4560 if (not_subtype_ctrl != top()) {
4561 PreserveJVMState pjvms(this);
4562 set_control(not_subtype_ctrl);
4563 uncommon_trap(Deoptimization::Reason_class_check,
4564 Deoptimization::Action_make_not_entrant);
4565 assert(stopped(), "Should be stopped");
4566 }
4567 validated = true;
4568 }
4569
4570 if (!stopped()) {
4571 newcopy = new_array(klass_node, length, 0); // no arguments to push
4572
4573 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4574 load_object_klass(original), klass_node);
4575 if (!is_copyOfRange) {
4576 ac->set_copyof(validated);
4577 } else {
4578 ac->set_copyofrange(validated);
4579 }
4580 Node* n = _gvn.transform(ac);
4581 if (n == ac) {
4582 ac->connect_outputs(this);
4583 } else {
4584 assert(validated, "shouldn't transform if all arguments not validated");
4585 set_all_memory(n);
4586 }
4587 }
4588 }
4589 } // original reexecute is set back here
4590
4591 C->set_has_split_ifs(true); // Has chance for split-if optimization
4592 if (!stopped()) {
4593 set_result(newcopy);
4594 }
4595 return true;
4596 }
4597
4598
4599 //----------------------generate_virtual_guard---------------------------
4600 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4601 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4602 RegionNode* slow_region) {
4603 ciMethod* method = callee();
4604 int vtable_index = method->vtable_index();
4605 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4606 "bad index %d", vtable_index);
4607 // Get the Method* out of the appropriate vtable entry.
4608 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4609 vtable_index*vtableEntry::size_in_bytes() +
4610 in_bytes(vtableEntry::method_offset());
4611 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4612 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4613
4614 // Compare the target method with the expected method (e.g., Object.hashCode).
4615 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4616
4617 Node* native_call = makecon(native_call_addr);
4618 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4619 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4620
4621 return generate_slow_guard(test_native, slow_region);
4622 }
4623
4624 //-----------------------generate_method_call----------------------------
4625 // Use generate_method_call to make a slow-call to the real
4626 // method if the fast path fails. An alternative would be to
4627 // use a stub like OptoRuntime::slow_arraycopy_Java.
4628 // This only works for expanding the current library call,
4629 // not another intrinsic. (E.g., don't use this for making an
4630 // arraycopy call inside of the copyOf intrinsic.)
4631 CallJavaNode*
4632 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4633 // When compiling the intrinsic method itself, do not use this technique.
4634 guarantee(callee() != C->method(), "cannot make slow-call to self");
4635
4636 ciMethod* method = callee();
4637 // ensure the JVMS we have will be correct for this call
4638 guarantee(method_id == method->intrinsic_id(), "must match");
4639
4640 const TypeFunc* tf = TypeFunc::make(method);
4641 if (res_not_null) {
4642 assert(tf->return_type() == T_OBJECT, "");
4643 const TypeTuple* range = tf->range();
4644 const Type** fields = TypeTuple::fields(range->cnt());
4645 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4646 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4647 tf = TypeFunc::make(tf->domain(), new_range);
4648 }
4649 CallJavaNode* slow_call;
4650 if (is_static) {
4651 assert(!is_virtual, "");
4652 slow_call = new CallStaticJavaNode(C, tf,
4653 SharedRuntime::get_resolve_static_call_stub(), method);
4654 } else if (is_virtual) {
4655 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4656 int vtable_index = Method::invalid_vtable_index;
4657 if (UseInlineCaches) {
4658 // Suppress the vtable call
4659 } else {
4660 // hashCode and clone are not a miranda methods,
4661 // so the vtable index is fixed.
4662 // No need to use the linkResolver to get it.
4663 vtable_index = method->vtable_index();
4664 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4665 "bad index %d", vtable_index);
4666 }
4667 slow_call = new CallDynamicJavaNode(tf,
4668 SharedRuntime::get_resolve_virtual_call_stub(),
4669 method, vtable_index);
4670 } else { // neither virtual nor static: opt_virtual
4671 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4672 slow_call = new CallStaticJavaNode(C, tf,
4673 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4674 slow_call->set_optimized_virtual(true);
4675 }
4676 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4677 // To be able to issue a direct call (optimized virtual or virtual)
4678 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4679 // about the method being invoked should be attached to the call site to
4680 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4681 slow_call->set_override_symbolic_info(true);
4682 }
4683 set_arguments_for_java_call(slow_call);
4684 set_edges_for_java_call(slow_call);
4685 return slow_call;
4686 }
4687
4688
4689 /**
4690 * Build special case code for calls to hashCode on an object. This call may
4691 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4692 * slightly different code.
4693 */
4694 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4695 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4696 assert(!(is_virtual && is_static), "either virtual, special, or static");
4697
4698 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4699
4700 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4701 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4702 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4703 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4704 Node* obj = nullptr;
4705 if (!is_static) {
4706 // Check for hashing null object
4707 obj = null_check_receiver();
4708 if (stopped()) return true; // unconditionally null
4709 result_reg->init_req(_null_path, top());
4710 result_val->init_req(_null_path, top());
4711 } else {
4712 // Do a null check, and return zero if null.
4713 // System.identityHashCode(null) == 0
4714 obj = argument(0);
4715 Node* null_ctl = top();
4716 obj = null_check_oop(obj, &null_ctl);
4717 result_reg->init_req(_null_path, null_ctl);
4718 result_val->init_req(_null_path, _gvn.intcon(0));
4719 }
4720
4721 // Unconditionally null? Then return right away.
4722 if (stopped()) {
4723 set_control( result_reg->in(_null_path));
4724 if (!stopped())
4725 set_result(result_val->in(_null_path));
4726 return true;
4727 }
4728
4729 // We only go to the fast case code if we pass a number of guards. The
4730 // paths which do not pass are accumulated in the slow_region.
4731 RegionNode* slow_region = new RegionNode(1);
4732 record_for_igvn(slow_region);
4733
4734 // If this is a virtual call, we generate a funny guard. We pull out
4735 // the vtable entry corresponding to hashCode() from the target object.
4736 // If the target method which we are calling happens to be the native
4737 // Object hashCode() method, we pass the guard. We do not need this
4738 // guard for non-virtual calls -- the caller is known to be the native
4739 // Object hashCode().
4740 if (is_virtual) {
4741 // After null check, get the object's klass.
4742 Node* obj_klass = load_object_klass(obj);
4743 generate_virtual_guard(obj_klass, slow_region);
4744 }
4745
4746 // Get the header out of the object, use LoadMarkNode when available
4747 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4748 // The control of the load must be null. Otherwise, the load can move before
4749 // the null check after castPP removal.
4750 Node* no_ctrl = nullptr;
4751 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4752
4753 if (!UseObjectMonitorTable) {
4754 // Test the header to see if it is safe to read w.r.t. locking.
4755 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4756 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4757 if (LockingMode == LM_LIGHTWEIGHT) {
4758 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4759 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4760 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4761
4762 generate_slow_guard(test_monitor, slow_region);
4763 } else {
4764 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4765 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4766 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4767
4768 generate_slow_guard(test_not_unlocked, slow_region);
4769 }
4770 }
4771
4772 // Get the hash value and check to see that it has been properly assigned.
4773 // We depend on hash_mask being at most 32 bits and avoid the use of
4774 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4775 // vm: see markWord.hpp.
4776 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4777 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4778 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4779 // This hack lets the hash bits live anywhere in the mark object now, as long
4780 // as the shift drops the relevant bits into the low 32 bits. Note that
4781 // Java spec says that HashCode is an int so there's no point in capturing
4782 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4783 hshifted_header = ConvX2I(hshifted_header);
4784 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4785
4786 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4787 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4788 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4789
4790 generate_slow_guard(test_assigned, slow_region);
4791
4792 Node* init_mem = reset_memory();
4793 // fill in the rest of the null path:
4794 result_io ->init_req(_null_path, i_o());
4795 result_mem->init_req(_null_path, init_mem);
4796
4797 result_val->init_req(_fast_path, hash_val);
4798 result_reg->init_req(_fast_path, control());
4799 result_io ->init_req(_fast_path, i_o());
4800 result_mem->init_req(_fast_path, init_mem);
4801
4802 // Generate code for the slow case. We make a call to hashCode().
4803 set_control(_gvn.transform(slow_region));
4804 if (!stopped()) {
4805 // No need for PreserveJVMState, because we're using up the present state.
4806 set_all_memory(init_mem);
4807 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4808 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4809 Node* slow_result = set_results_for_java_call(slow_call);
4810 // this->control() comes from set_results_for_java_call
4811 result_reg->init_req(_slow_path, control());
4812 result_val->init_req(_slow_path, slow_result);
4813 result_io ->set_req(_slow_path, i_o());
4814 result_mem ->set_req(_slow_path, reset_memory());
4815 }
4816
4817 // Return the combined state.
4818 set_i_o( _gvn.transform(result_io) );
4819 set_all_memory( _gvn.transform(result_mem));
4820
4821 set_result(result_reg, result_val);
4822 return true;
4823 }
4824
4825 //---------------------------inline_native_getClass----------------------------
4826 // public final native Class<?> java.lang.Object.getClass();
4827 //
4828 // Build special case code for calls to getClass on an object.
4829 bool LibraryCallKit::inline_native_getClass() {
4830 Node* obj = null_check_receiver();
4831 if (stopped()) return true;
4832 set_result(load_mirror_from_klass(load_object_klass(obj)));
4833 return true;
4834 }
4835
4836 //-----------------inline_native_Reflection_getCallerClass---------------------
4837 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4838 //
4839 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4840 //
4841 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4842 // in that it must skip particular security frames and checks for
4843 // caller sensitive methods.
4844 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4845 #ifndef PRODUCT
4846 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4847 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4848 }
4849 #endif
4850
4851 if (!jvms()->has_method()) {
4852 #ifndef PRODUCT
4853 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4854 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4855 }
4856 #endif
4857 return false;
4858 }
4859
4860 // Walk back up the JVM state to find the caller at the required
4861 // depth.
4862 JVMState* caller_jvms = jvms();
4863
4864 // Cf. JVM_GetCallerClass
4865 // NOTE: Start the loop at depth 1 because the current JVM state does
4866 // not include the Reflection.getCallerClass() frame.
4867 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4868 ciMethod* m = caller_jvms->method();
4869 switch (n) {
4870 case 0:
4871 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4872 break;
4873 case 1:
4874 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4875 if (!m->caller_sensitive()) {
4876 #ifndef PRODUCT
4877 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4878 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4879 }
4880 #endif
4881 return false; // bail-out; let JVM_GetCallerClass do the work
4882 }
4883 break;
4884 default:
4885 if (!m->is_ignored_by_security_stack_walk()) {
4886 // We have reached the desired frame; return the holder class.
4887 // Acquire method holder as java.lang.Class and push as constant.
4888 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4889 ciInstance* caller_mirror = caller_klass->java_mirror();
4890 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4891
4892 #ifndef PRODUCT
4893 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4894 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());
4895 tty->print_cr(" JVM state at this point:");
4896 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4897 ciMethod* m = jvms()->of_depth(i)->method();
4898 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4899 }
4900 }
4901 #endif
4902 return true;
4903 }
4904 break;
4905 }
4906 }
4907
4908 #ifndef PRODUCT
4909 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4910 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4911 tty->print_cr(" JVM state at this point:");
4912 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4913 ciMethod* m = jvms()->of_depth(i)->method();
4914 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4915 }
4916 }
4917 #endif
4918
4919 return false; // bail-out; let JVM_GetCallerClass do the work
4920 }
4921
4922 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4923 Node* arg = argument(0);
4924 Node* result = nullptr;
4925
4926 switch (id) {
4927 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4928 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4929 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4930 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4931 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4932 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4933
4934 case vmIntrinsics::_doubleToLongBits: {
4935 // two paths (plus control) merge in a wood
4936 RegionNode *r = new RegionNode(3);
4937 Node *phi = new PhiNode(r, TypeLong::LONG);
4938
4939 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4940 // Build the boolean node
4941 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4942
4943 // Branch either way.
4944 // NaN case is less traveled, which makes all the difference.
4945 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4946 Node *opt_isnan = _gvn.transform(ifisnan);
4947 assert( opt_isnan->is_If(), "Expect an IfNode");
4948 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4949 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4950
4951 set_control(iftrue);
4952
4953 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4954 Node *slow_result = longcon(nan_bits); // return NaN
4955 phi->init_req(1, _gvn.transform( slow_result ));
4956 r->init_req(1, iftrue);
4957
4958 // Else fall through
4959 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4960 set_control(iffalse);
4961
4962 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4963 r->init_req(2, iffalse);
4964
4965 // Post merge
4966 set_control(_gvn.transform(r));
4967 record_for_igvn(r);
4968
4969 C->set_has_split_ifs(true); // Has chance for split-if optimization
4970 result = phi;
4971 assert(result->bottom_type()->isa_long(), "must be");
4972 break;
4973 }
4974
4975 case vmIntrinsics::_floatToIntBits: {
4976 // two paths (plus control) merge in a wood
4977 RegionNode *r = new RegionNode(3);
4978 Node *phi = new PhiNode(r, TypeInt::INT);
4979
4980 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4981 // Build the boolean node
4982 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4983
4984 // Branch either way.
4985 // NaN case is less traveled, which makes all the difference.
4986 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4987 Node *opt_isnan = _gvn.transform(ifisnan);
4988 assert( opt_isnan->is_If(), "Expect an IfNode");
4989 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4990 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4991
4992 set_control(iftrue);
4993
4994 static const jint nan_bits = 0x7fc00000;
4995 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4996 phi->init_req(1, _gvn.transform( slow_result ));
4997 r->init_req(1, iftrue);
4998
4999 // Else fall through
5000 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5001 set_control(iffalse);
5002
5003 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5004 r->init_req(2, iffalse);
5005
5006 // Post merge
5007 set_control(_gvn.transform(r));
5008 record_for_igvn(r);
5009
5010 C->set_has_split_ifs(true); // Has chance for split-if optimization
5011 result = phi;
5012 assert(result->bottom_type()->isa_int(), "must be");
5013 break;
5014 }
5015
5016 default:
5017 fatal_unexpected_iid(id);
5018 break;
5019 }
5020 set_result(_gvn.transform(result));
5021 return true;
5022 }
5023
5024 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5025 Node* arg = argument(0);
5026 Node* result = nullptr;
5027
5028 switch (id) {
5029 case vmIntrinsics::_floatIsInfinite:
5030 result = new IsInfiniteFNode(arg);
5031 break;
5032 case vmIntrinsics::_floatIsFinite:
5033 result = new IsFiniteFNode(arg);
5034 break;
5035 case vmIntrinsics::_doubleIsInfinite:
5036 result = new IsInfiniteDNode(arg);
5037 break;
5038 case vmIntrinsics::_doubleIsFinite:
5039 result = new IsFiniteDNode(arg);
5040 break;
5041 default:
5042 fatal_unexpected_iid(id);
5043 break;
5044 }
5045 set_result(_gvn.transform(result));
5046 return true;
5047 }
5048
5049 //----------------------inline_unsafe_copyMemory-------------------------
5050 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5051
5052 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5053 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5054 const Type* base_t = gvn.type(base);
5055
5056 bool in_native = (base_t == TypePtr::NULL_PTR);
5057 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5058 bool is_mixed = !in_heap && !in_native;
5059
5060 if (is_mixed) {
5061 return true; // mixed accesses can touch both on-heap and off-heap memory
5062 }
5063 if (in_heap) {
5064 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5065 if (!is_prim_array) {
5066 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5067 // there's not enough type information available to determine proper memory slice for it.
5068 return true;
5069 }
5070 }
5071 return false;
5072 }
5073
5074 bool LibraryCallKit::inline_unsafe_copyMemory() {
5075 if (callee()->is_static()) return false; // caller must have the capability!
5076 null_check_receiver(); // null-check receiver
5077 if (stopped()) return true;
5078
5079 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5080
5081 Node* src_base = argument(1); // type: oop
5082 Node* src_off = ConvL2X(argument(2)); // type: long
5083 Node* dst_base = argument(4); // type: oop
5084 Node* dst_off = ConvL2X(argument(5)); // type: long
5085 Node* size = ConvL2X(argument(7)); // type: long
5086
5087 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5088 "fieldOffset must be byte-scaled");
5089
5090 Node* src_addr = make_unsafe_address(src_base, src_off);
5091 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5092
5093 Node* thread = _gvn.transform(new ThreadLocalNode());
5094 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5095 BasicType doing_unsafe_access_bt = T_BYTE;
5096 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5097
5098 // update volatile field
5099 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5100
5101 int flags = RC_LEAF | RC_NO_FP;
5102
5103 const TypePtr* dst_type = TypePtr::BOTTOM;
5104
5105 // Adjust memory effects of the runtime call based on input values.
5106 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5107 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5108 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5109
5110 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5111 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5112 flags |= RC_NARROW_MEM; // narrow in memory
5113 }
5114 }
5115
5116 // Call it. Note that the length argument is not scaled.
5117 make_runtime_call(flags,
5118 OptoRuntime::fast_arraycopy_Type(),
5119 StubRoutines::unsafe_arraycopy(),
5120 "unsafe_arraycopy",
5121 dst_type,
5122 src_addr, dst_addr, size XTOP);
5123
5124 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5125
5126 return true;
5127 }
5128
5129 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5130 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5131 bool LibraryCallKit::inline_unsafe_setMemory() {
5132 if (callee()->is_static()) return false; // caller must have the capability!
5133 null_check_receiver(); // null-check receiver
5134 if (stopped()) return true;
5135
5136 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5137
5138 Node* dst_base = argument(1); // type: oop
5139 Node* dst_off = ConvL2X(argument(2)); // type: long
5140 Node* size = ConvL2X(argument(4)); // type: long
5141 Node* byte = argument(6); // type: byte
5142
5143 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5144 "fieldOffset must be byte-scaled");
5145
5146 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5147
5148 Node* thread = _gvn.transform(new ThreadLocalNode());
5149 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5150 BasicType doing_unsafe_access_bt = T_BYTE;
5151 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5152
5153 // update volatile field
5154 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5155
5156 int flags = RC_LEAF | RC_NO_FP;
5157
5158 const TypePtr* dst_type = TypePtr::BOTTOM;
5159
5160 // Adjust memory effects of the runtime call based on input values.
5161 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5162 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5163
5164 flags |= RC_NARROW_MEM; // narrow in memory
5165 }
5166
5167 // Call it. Note that the length argument is not scaled.
5168 make_runtime_call(flags,
5169 OptoRuntime::unsafe_setmemory_Type(),
5170 StubRoutines::unsafe_setmemory(),
5171 "unsafe_setmemory",
5172 dst_type,
5173 dst_addr, size XTOP, byte);
5174
5175 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5176
5177 return true;
5178 }
5179
5180 #undef XTOP
5181
5182 //------------------------clone_coping-----------------------------------
5183 // Helper function for inline_native_clone.
5184 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5185 assert(obj_size != nullptr, "");
5186 Node* raw_obj = alloc_obj->in(1);
5187 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5188
5189 AllocateNode* alloc = nullptr;
5190 if (ReduceBulkZeroing &&
5191 // If we are implementing an array clone without knowing its source type
5192 // (can happen when compiling the array-guarded branch of a reflective
5193 // Object.clone() invocation), initialize the array within the allocation.
5194 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5195 // to a runtime clone call that assumes fully initialized source arrays.
5196 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5197 // We will be completely responsible for initializing this object -
5198 // mark Initialize node as complete.
5199 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5200 // The object was just allocated - there should be no any stores!
5201 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5202 // Mark as complete_with_arraycopy so that on AllocateNode
5203 // expansion, we know this AllocateNode is initialized by an array
5204 // copy and a StoreStore barrier exists after the array copy.
5205 alloc->initialization()->set_complete_with_arraycopy();
5206 }
5207
5208 Node* size = _gvn.transform(obj_size);
5209 access_clone(obj, alloc_obj, size, is_array);
5210
5211 // Do not let reads from the cloned object float above the arraycopy.
5212 if (alloc != nullptr) {
5213 // Do not let stores that initialize this object be reordered with
5214 // a subsequent store that would make this object accessible by
5215 // other threads.
5216 // Record what AllocateNode this StoreStore protects so that
5217 // escape analysis can go from the MemBarStoreStoreNode to the
5218 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5219 // based on the escape status of the AllocateNode.
5220 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5221 } else {
5222 insert_mem_bar(Op_MemBarCPUOrder);
5223 }
5224 }
5225
5226 //------------------------inline_native_clone----------------------------
5227 // protected native Object java.lang.Object.clone();
5228 //
5229 // Here are the simple edge cases:
5230 // null receiver => normal trap
5231 // virtual and clone was overridden => slow path to out-of-line clone
5232 // not cloneable or finalizer => slow path to out-of-line Object.clone
5233 //
5234 // The general case has two steps, allocation and copying.
5235 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5236 //
5237 // Copying also has two cases, oop arrays and everything else.
5238 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5239 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5240 //
5241 // These steps fold up nicely if and when the cloned object's klass
5242 // can be sharply typed as an object array, a type array, or an instance.
5243 //
5244 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5245 PhiNode* result_val;
5246
5247 // Set the reexecute bit for the interpreter to reexecute
5248 // the bytecode that invokes Object.clone if deoptimization happens.
5249 { PreserveReexecuteState preexecs(this);
5250 jvms()->set_should_reexecute(true);
5251
5252 Node* obj = null_check_receiver();
5253 if (stopped()) return true;
5254
5255 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5256
5257 // If we are going to clone an instance, we need its exact type to
5258 // know the number and types of fields to convert the clone to
5259 // loads/stores. Maybe a speculative type can help us.
5260 if (!obj_type->klass_is_exact() &&
5261 obj_type->speculative_type() != nullptr &&
5262 obj_type->speculative_type()->is_instance_klass()) {
5263 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5264 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5265 !spec_ik->has_injected_fields()) {
5266 if (!obj_type->isa_instptr() ||
5267 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5268 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5269 }
5270 }
5271 }
5272
5273 // Conservatively insert a memory barrier on all memory slices.
5274 // Do not let writes into the original float below the clone.
5275 insert_mem_bar(Op_MemBarCPUOrder);
5276
5277 // paths into result_reg:
5278 enum {
5279 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5280 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5281 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5282 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5283 PATH_LIMIT
5284 };
5285 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5286 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5287 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5288 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5289 record_for_igvn(result_reg);
5290
5291 Node* obj_klass = load_object_klass(obj);
5292 Node* array_obj = obj;
5293 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5294 if (array_ctl != nullptr) {
5295 // It's an array.
5296 PreserveJVMState pjvms(this);
5297 set_control(array_ctl);
5298 Node* obj_length = load_array_length(array_obj);
5299 Node* array_size = nullptr; // Size of the array without object alignment padding.
5300 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5301
5302 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5303 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5304 // If it is an oop array, it requires very special treatment,
5305 // because gc barriers are required when accessing the array.
5306 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5307 if (is_obja != nullptr) {
5308 PreserveJVMState pjvms2(this);
5309 set_control(is_obja);
5310 // Generate a direct call to the right arraycopy function(s).
5311 // Clones are always tightly coupled.
5312 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5313 ac->set_clone_oop_array();
5314 Node* n = _gvn.transform(ac);
5315 assert(n == ac, "cannot disappear");
5316 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5317
5318 result_reg->init_req(_objArray_path, control());
5319 result_val->init_req(_objArray_path, alloc_obj);
5320 result_i_o ->set_req(_objArray_path, i_o());
5321 result_mem ->set_req(_objArray_path, reset_memory());
5322 }
5323 }
5324 // Otherwise, there are no barriers to worry about.
5325 // (We can dispense with card marks if we know the allocation
5326 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5327 // causes the non-eden paths to take compensating steps to
5328 // simulate a fresh allocation, so that no further
5329 // card marks are required in compiled code to initialize
5330 // the object.)
5331
5332 if (!stopped()) {
5333 copy_to_clone(array_obj, alloc_obj, array_size, true);
5334
5335 // Present the results of the copy.
5336 result_reg->init_req(_array_path, control());
5337 result_val->init_req(_array_path, alloc_obj);
5338 result_i_o ->set_req(_array_path, i_o());
5339 result_mem ->set_req(_array_path, reset_memory());
5340 }
5341 }
5342
5343 // We only go to the instance fast case code if we pass a number of guards.
5344 // The paths which do not pass are accumulated in the slow_region.
5345 RegionNode* slow_region = new RegionNode(1);
5346 record_for_igvn(slow_region);
5347 if (!stopped()) {
5348 // It's an instance (we did array above). Make the slow-path tests.
5349 // If this is a virtual call, we generate a funny guard. We grab
5350 // the vtable entry corresponding to clone() from the target object.
5351 // If the target method which we are calling happens to be the
5352 // Object clone() method, we pass the guard. We do not need this
5353 // guard for non-virtual calls; the caller is known to be the native
5354 // Object clone().
5355 if (is_virtual) {
5356 generate_virtual_guard(obj_klass, slow_region);
5357 }
5358
5359 // The object must be easily cloneable and must not have a finalizer.
5360 // Both of these conditions may be checked in a single test.
5361 // We could optimize the test further, but we don't care.
5362 generate_misc_flags_guard(obj_klass,
5363 // Test both conditions:
5364 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5365 // Must be cloneable but not finalizer:
5366 KlassFlags::_misc_is_cloneable_fast,
5367 slow_region);
5368 }
5369
5370 if (!stopped()) {
5371 // It's an instance, and it passed the slow-path tests.
5372 PreserveJVMState pjvms(this);
5373 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5374 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5375 // is reexecuted if deoptimization occurs and there could be problems when merging
5376 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5377 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5378
5379 copy_to_clone(obj, alloc_obj, obj_size, false);
5380
5381 // Present the results of the slow call.
5382 result_reg->init_req(_instance_path, control());
5383 result_val->init_req(_instance_path, alloc_obj);
5384 result_i_o ->set_req(_instance_path, i_o());
5385 result_mem ->set_req(_instance_path, reset_memory());
5386 }
5387
5388 // Generate code for the slow case. We make a call to clone().
5389 set_control(_gvn.transform(slow_region));
5390 if (!stopped()) {
5391 PreserveJVMState pjvms(this);
5392 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5393 // We need to deoptimize on exception (see comment above)
5394 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5395 // this->control() comes from set_results_for_java_call
5396 result_reg->init_req(_slow_path, control());
5397 result_val->init_req(_slow_path, slow_result);
5398 result_i_o ->set_req(_slow_path, i_o());
5399 result_mem ->set_req(_slow_path, reset_memory());
5400 }
5401
5402 // Return the combined state.
5403 set_control( _gvn.transform(result_reg));
5404 set_i_o( _gvn.transform(result_i_o));
5405 set_all_memory( _gvn.transform(result_mem));
5406 } // original reexecute is set back here
5407
5408 set_result(_gvn.transform(result_val));
5409 return true;
5410 }
5411
5412 // If we have a tightly coupled allocation, the arraycopy may take care
5413 // of the array initialization. If one of the guards we insert between
5414 // the allocation and the arraycopy causes a deoptimization, an
5415 // uninitialized array will escape the compiled method. To prevent that
5416 // we set the JVM state for uncommon traps between the allocation and
5417 // the arraycopy to the state before the allocation so, in case of
5418 // deoptimization, we'll reexecute the allocation and the
5419 // initialization.
5420 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5421 if (alloc != nullptr) {
5422 ciMethod* trap_method = alloc->jvms()->method();
5423 int trap_bci = alloc->jvms()->bci();
5424
5425 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5426 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5427 // Make sure there's no store between the allocation and the
5428 // arraycopy otherwise visible side effects could be rexecuted
5429 // in case of deoptimization and cause incorrect execution.
5430 bool no_interfering_store = true;
5431 Node* mem = alloc->in(TypeFunc::Memory);
5432 if (mem->is_MergeMem()) {
5433 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5434 Node* n = mms.memory();
5435 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5436 assert(n->is_Store(), "what else?");
5437 no_interfering_store = false;
5438 break;
5439 }
5440 }
5441 } else {
5442 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5443 Node* n = mms.memory();
5444 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5445 assert(n->is_Store(), "what else?");
5446 no_interfering_store = false;
5447 break;
5448 }
5449 }
5450 }
5451
5452 if (no_interfering_store) {
5453 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5454
5455 JVMState* saved_jvms = jvms();
5456 saved_reexecute_sp = _reexecute_sp;
5457
5458 set_jvms(sfpt->jvms());
5459 _reexecute_sp = jvms()->sp();
5460
5461 return saved_jvms;
5462 }
5463 }
5464 }
5465 return nullptr;
5466 }
5467
5468 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5469 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5470 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5471 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5472 uint size = alloc->req();
5473 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5474 old_jvms->set_map(sfpt);
5475 for (uint i = 0; i < size; i++) {
5476 sfpt->init_req(i, alloc->in(i));
5477 }
5478 // re-push array length for deoptimization
5479 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5480 old_jvms->set_sp(old_jvms->sp()+1);
5481 old_jvms->set_monoff(old_jvms->monoff()+1);
5482 old_jvms->set_scloff(old_jvms->scloff()+1);
5483 old_jvms->set_endoff(old_jvms->endoff()+1);
5484 old_jvms->set_should_reexecute(true);
5485
5486 sfpt->set_i_o(map()->i_o());
5487 sfpt->set_memory(map()->memory());
5488 sfpt->set_control(map()->control());
5489 return sfpt;
5490 }
5491
5492 // In case of a deoptimization, we restart execution at the
5493 // allocation, allocating a new array. We would leave an uninitialized
5494 // array in the heap that GCs wouldn't expect. Move the allocation
5495 // after the traps so we don't allocate the array if we
5496 // deoptimize. This is possible because tightly_coupled_allocation()
5497 // guarantees there's no observer of the allocated array at this point
5498 // and the control flow is simple enough.
5499 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5500 int saved_reexecute_sp, uint new_idx) {
5501 if (saved_jvms_before_guards != nullptr && !stopped()) {
5502 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5503
5504 assert(alloc != nullptr, "only with a tightly coupled allocation");
5505 // restore JVM state to the state at the arraycopy
5506 saved_jvms_before_guards->map()->set_control(map()->control());
5507 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5508 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5509 // If we've improved the types of some nodes (null check) while
5510 // emitting the guards, propagate them to the current state
5511 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5512 set_jvms(saved_jvms_before_guards);
5513 _reexecute_sp = saved_reexecute_sp;
5514
5515 // Remove the allocation from above the guards
5516 CallProjections callprojs;
5517 alloc->extract_projections(&callprojs, true);
5518 InitializeNode* init = alloc->initialization();
5519 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5520 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5521 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5522
5523 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5524 // the allocation (i.e. is only valid if the allocation succeeds):
5525 // 1) replace CastIINode with AllocateArrayNode's length here
5526 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5527 //
5528 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5529 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5530 Node* init_control = init->proj_out(TypeFunc::Control);
5531 Node* alloc_length = alloc->Ideal_length();
5532 #ifdef ASSERT
5533 Node* prev_cast = nullptr;
5534 #endif
5535 for (uint i = 0; i < init_control->outcnt(); i++) {
5536 Node* init_out = init_control->raw_out(i);
5537 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5538 #ifdef ASSERT
5539 if (prev_cast == nullptr) {
5540 prev_cast = init_out;
5541 } else {
5542 if (prev_cast->cmp(*init_out) == false) {
5543 prev_cast->dump();
5544 init_out->dump();
5545 assert(false, "not equal CastIINode");
5546 }
5547 }
5548 #endif
5549 C->gvn_replace_by(init_out, alloc_length);
5550 }
5551 }
5552 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5553
5554 // move the allocation here (after the guards)
5555 _gvn.hash_delete(alloc);
5556 alloc->set_req(TypeFunc::Control, control());
5557 alloc->set_req(TypeFunc::I_O, i_o());
5558 Node *mem = reset_memory();
5559 set_all_memory(mem);
5560 alloc->set_req(TypeFunc::Memory, mem);
5561 set_control(init->proj_out_or_null(TypeFunc::Control));
5562 set_i_o(callprojs.fallthrough_ioproj);
5563
5564 // Update memory as done in GraphKit::set_output_for_allocation()
5565 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5566 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5567 if (ary_type->isa_aryptr() && length_type != nullptr) {
5568 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5569 }
5570 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5571 int elemidx = C->get_alias_index(telemref);
5572 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5573 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5574
5575 Node* allocx = _gvn.transform(alloc);
5576 assert(allocx == alloc, "where has the allocation gone?");
5577 assert(dest->is_CheckCastPP(), "not an allocation result?");
5578
5579 _gvn.hash_delete(dest);
5580 dest->set_req(0, control());
5581 Node* destx = _gvn.transform(dest);
5582 assert(destx == dest, "where has the allocation result gone?");
5583
5584 array_ideal_length(alloc, ary_type, true);
5585 }
5586 }
5587
5588 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5589 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5590 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5591 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5592 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5593 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5594 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5595 JVMState* saved_jvms_before_guards) {
5596 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5597 // There is at least one unrelated uncommon trap which needs to be replaced.
5598 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5599
5600 JVMState* saved_jvms = jvms();
5601 const int saved_reexecute_sp = _reexecute_sp;
5602 set_jvms(sfpt->jvms());
5603 _reexecute_sp = jvms()->sp();
5604
5605 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5606
5607 // Restore state
5608 set_jvms(saved_jvms);
5609 _reexecute_sp = saved_reexecute_sp;
5610 }
5611 }
5612
5613 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5614 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5615 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5616 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5617 while (if_proj->is_IfProj()) {
5618 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5619 if (uncommon_trap != nullptr) {
5620 create_new_uncommon_trap(uncommon_trap);
5621 }
5622 assert(if_proj->in(0)->is_If(), "must be If");
5623 if_proj = if_proj->in(0)->in(0);
5624 }
5625 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5626 "must have reached control projection of init node");
5627 }
5628
5629 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5630 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5631 assert(trap_request != 0, "no valid UCT trap request");
5632 PreserveJVMState pjvms(this);
5633 set_control(uncommon_trap_call->in(0));
5634 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5635 Deoptimization::trap_request_action(trap_request));
5636 assert(stopped(), "Should be stopped");
5637 _gvn.hash_delete(uncommon_trap_call);
5638 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5639 }
5640
5641 // Common checks for array sorting intrinsics arguments.
5642 // Returns `true` if checks passed.
5643 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5644 // check address of the class
5645 if (elementType == nullptr || elementType->is_top()) {
5646 return false; // dead path
5647 }
5648 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5649 if (elem_klass == nullptr) {
5650 return false; // dead path
5651 }
5652 // java_mirror_type() returns non-null for compile-time Class constants only
5653 ciType* elem_type = elem_klass->java_mirror_type();
5654 if (elem_type == nullptr) {
5655 return false;
5656 }
5657 bt = elem_type->basic_type();
5658 // Disable the intrinsic if the CPU does not support SIMD sort
5659 if (!Matcher::supports_simd_sort(bt)) {
5660 return false;
5661 }
5662 // check address of the array
5663 if (obj == nullptr || obj->is_top()) {
5664 return false; // dead path
5665 }
5666 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5667 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5668 return false; // failed input validation
5669 }
5670 return true;
5671 }
5672
5673 //------------------------------inline_array_partition-----------------------
5674 bool LibraryCallKit::inline_array_partition() {
5675 address stubAddr = StubRoutines::select_array_partition_function();
5676 if (stubAddr == nullptr) {
5677 return false; // Intrinsic's stub is not implemented on this platform
5678 }
5679 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5680
5681 // no receiver because it is a static method
5682 Node* elementType = argument(0);
5683 Node* obj = argument(1);
5684 Node* offset = argument(2); // long
5685 Node* fromIndex = argument(4);
5686 Node* toIndex = argument(5);
5687 Node* indexPivot1 = argument(6);
5688 Node* indexPivot2 = argument(7);
5689 // PartitionOperation: argument(8) is ignored
5690
5691 Node* pivotIndices = nullptr;
5692 BasicType bt = T_ILLEGAL;
5693
5694 if (!check_array_sort_arguments(elementType, obj, bt)) {
5695 return false;
5696 }
5697 null_check(obj);
5698 // If obj is dead, only null-path is taken.
5699 if (stopped()) {
5700 return true;
5701 }
5702 // Set the original stack and the reexecute bit for the interpreter to reexecute
5703 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5704 { PreserveReexecuteState preexecs(this);
5705 jvms()->set_should_reexecute(true);
5706
5707 Node* obj_adr = make_unsafe_address(obj, offset);
5708
5709 // create the pivotIndices array of type int and size = 2
5710 Node* size = intcon(2);
5711 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5712 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5713 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5714 guarantee(alloc != nullptr, "created above");
5715 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5716
5717 // pass the basic type enum to the stub
5718 Node* elemType = intcon(bt);
5719
5720 // Call the stub
5721 const char *stubName = "array_partition_stub";
5722 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5723 stubAddr, stubName, TypePtr::BOTTOM,
5724 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5725 indexPivot1, indexPivot2);
5726
5727 } // original reexecute is set back here
5728
5729 if (!stopped()) {
5730 set_result(pivotIndices);
5731 }
5732
5733 return true;
5734 }
5735
5736
5737 //------------------------------inline_array_sort-----------------------
5738 bool LibraryCallKit::inline_array_sort() {
5739 address stubAddr = StubRoutines::select_arraysort_function();
5740 if (stubAddr == nullptr) {
5741 return false; // Intrinsic's stub is not implemented on this platform
5742 }
5743 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5744
5745 // no receiver because it is a static method
5746 Node* elementType = argument(0);
5747 Node* obj = argument(1);
5748 Node* offset = argument(2); // long
5749 Node* fromIndex = argument(4);
5750 Node* toIndex = argument(5);
5751 // SortOperation: argument(6) is ignored
5752
5753 BasicType bt = T_ILLEGAL;
5754
5755 if (!check_array_sort_arguments(elementType, obj, bt)) {
5756 return false;
5757 }
5758 null_check(obj);
5759 // If obj is dead, only null-path is taken.
5760 if (stopped()) {
5761 return true;
5762 }
5763 Node* obj_adr = make_unsafe_address(obj, offset);
5764
5765 // pass the basic type enum to the stub
5766 Node* elemType = intcon(bt);
5767
5768 // Call the stub.
5769 const char *stubName = "arraysort_stub";
5770 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5771 stubAddr, stubName, TypePtr::BOTTOM,
5772 obj_adr, elemType, fromIndex, toIndex);
5773
5774 return true;
5775 }
5776
5777
5778 //------------------------------inline_arraycopy-----------------------
5779 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5780 // Object dest, int destPos,
5781 // int length);
5782 bool LibraryCallKit::inline_arraycopy() {
5783 // Get the arguments.
5784 Node* src = argument(0); // type: oop
5785 Node* src_offset = argument(1); // type: int
5786 Node* dest = argument(2); // type: oop
5787 Node* dest_offset = argument(3); // type: int
5788 Node* length = argument(4); // type: int
5789
5790 uint new_idx = C->unique();
5791
5792 // Check for allocation before we add nodes that would confuse
5793 // tightly_coupled_allocation()
5794 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5795
5796 int saved_reexecute_sp = -1;
5797 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5798 // See arraycopy_restore_alloc_state() comment
5799 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5800 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5801 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5802 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5803
5804 // The following tests must be performed
5805 // (1) src and dest are arrays.
5806 // (2) src and dest arrays must have elements of the same BasicType
5807 // (3) src and dest must not be null.
5808 // (4) src_offset must not be negative.
5809 // (5) dest_offset must not be negative.
5810 // (6) length must not be negative.
5811 // (7) src_offset + length must not exceed length of src.
5812 // (8) dest_offset + length must not exceed length of dest.
5813 // (9) each element of an oop array must be assignable
5814
5815 // (3) src and dest must not be null.
5816 // always do this here because we need the JVM state for uncommon traps
5817 Node* null_ctl = top();
5818 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5819 assert(null_ctl->is_top(), "no null control here");
5820 dest = null_check(dest, T_ARRAY);
5821
5822 if (!can_emit_guards) {
5823 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5824 // guards but the arraycopy node could still take advantage of a
5825 // tightly allocated allocation. tightly_coupled_allocation() is
5826 // called again to make sure it takes the null check above into
5827 // account: the null check is mandatory and if it caused an
5828 // uncommon trap to be emitted then the allocation can't be
5829 // considered tightly coupled in this context.
5830 alloc = tightly_coupled_allocation(dest);
5831 }
5832
5833 bool validated = false;
5834
5835 const Type* src_type = _gvn.type(src);
5836 const Type* dest_type = _gvn.type(dest);
5837 const TypeAryPtr* top_src = src_type->isa_aryptr();
5838 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5839
5840 // Do we have the type of src?
5841 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5842 // Do we have the type of dest?
5843 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5844 // Is the type for src from speculation?
5845 bool src_spec = false;
5846 // Is the type for dest from speculation?
5847 bool dest_spec = false;
5848
5849 if ((!has_src || !has_dest) && can_emit_guards) {
5850 // We don't have sufficient type information, let's see if
5851 // speculative types can help. We need to have types for both src
5852 // and dest so that it pays off.
5853
5854 // Do we already have or could we have type information for src
5855 bool could_have_src = has_src;
5856 // Do we already have or could we have type information for dest
5857 bool could_have_dest = has_dest;
5858
5859 ciKlass* src_k = nullptr;
5860 if (!has_src) {
5861 src_k = src_type->speculative_type_not_null();
5862 if (src_k != nullptr && src_k->is_array_klass()) {
5863 could_have_src = true;
5864 }
5865 }
5866
5867 ciKlass* dest_k = nullptr;
5868 if (!has_dest) {
5869 dest_k = dest_type->speculative_type_not_null();
5870 if (dest_k != nullptr && dest_k->is_array_klass()) {
5871 could_have_dest = true;
5872 }
5873 }
5874
5875 if (could_have_src && could_have_dest) {
5876 // This is going to pay off so emit the required guards
5877 if (!has_src) {
5878 src = maybe_cast_profiled_obj(src, src_k, true);
5879 src_type = _gvn.type(src);
5880 top_src = src_type->isa_aryptr();
5881 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5882 src_spec = true;
5883 }
5884 if (!has_dest) {
5885 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5886 dest_type = _gvn.type(dest);
5887 top_dest = dest_type->isa_aryptr();
5888 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5889 dest_spec = true;
5890 }
5891 }
5892 }
5893
5894 if (has_src && has_dest && can_emit_guards) {
5895 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5896 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5897 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5898 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5899
5900 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5901 // If both arrays are object arrays then having the exact types
5902 // for both will remove the need for a subtype check at runtime
5903 // before the call and may make it possible to pick a faster copy
5904 // routine (without a subtype check on every element)
5905 // Do we have the exact type of src?
5906 bool could_have_src = src_spec;
5907 // Do we have the exact type of dest?
5908 bool could_have_dest = dest_spec;
5909 ciKlass* src_k = nullptr;
5910 ciKlass* dest_k = nullptr;
5911 if (!src_spec) {
5912 src_k = src_type->speculative_type_not_null();
5913 if (src_k != nullptr && src_k->is_array_klass()) {
5914 could_have_src = true;
5915 }
5916 }
5917 if (!dest_spec) {
5918 dest_k = dest_type->speculative_type_not_null();
5919 if (dest_k != nullptr && dest_k->is_array_klass()) {
5920 could_have_dest = true;
5921 }
5922 }
5923 if (could_have_src && could_have_dest) {
5924 // If we can have both exact types, emit the missing guards
5925 if (could_have_src && !src_spec) {
5926 src = maybe_cast_profiled_obj(src, src_k, true);
5927 }
5928 if (could_have_dest && !dest_spec) {
5929 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5930 }
5931 }
5932 }
5933 }
5934
5935 ciMethod* trap_method = method();
5936 int trap_bci = bci();
5937 if (saved_jvms_before_guards != nullptr) {
5938 trap_method = alloc->jvms()->method();
5939 trap_bci = alloc->jvms()->bci();
5940 }
5941
5942 bool negative_length_guard_generated = false;
5943
5944 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5945 can_emit_guards &&
5946 !src->is_top() && !dest->is_top()) {
5947 // validate arguments: enables transformation the ArrayCopyNode
5948 validated = true;
5949
5950 RegionNode* slow_region = new RegionNode(1);
5951 record_for_igvn(slow_region);
5952
5953 // (1) src and dest are arrays.
5954 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5955 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5956
5957 // (2) src and dest arrays must have elements of the same BasicType
5958 // done at macro expansion or at Ideal transformation time
5959
5960 // (4) src_offset must not be negative.
5961 generate_negative_guard(src_offset, slow_region);
5962
5963 // (5) dest_offset must not be negative.
5964 generate_negative_guard(dest_offset, slow_region);
5965
5966 // (7) src_offset + length must not exceed length of src.
5967 generate_limit_guard(src_offset, length,
5968 load_array_length(src),
5969 slow_region);
5970
5971 // (8) dest_offset + length must not exceed length of dest.
5972 generate_limit_guard(dest_offset, length,
5973 load_array_length(dest),
5974 slow_region);
5975
5976 // (6) length must not be negative.
5977 // This is also checked in generate_arraycopy() during macro expansion, but
5978 // we also have to check it here for the case where the ArrayCopyNode will
5979 // be eliminated by Escape Analysis.
5980 if (EliminateAllocations) {
5981 generate_negative_guard(length, slow_region);
5982 negative_length_guard_generated = true;
5983 }
5984
5985 // (9) each element of an oop array must be assignable
5986 Node* dest_klass = load_object_klass(dest);
5987 if (src != dest) {
5988 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5989
5990 if (not_subtype_ctrl != top()) {
5991 PreserveJVMState pjvms(this);
5992 set_control(not_subtype_ctrl);
5993 uncommon_trap(Deoptimization::Reason_intrinsic,
5994 Deoptimization::Action_make_not_entrant);
5995 assert(stopped(), "Should be stopped");
5996 }
5997 }
5998 {
5999 PreserveJVMState pjvms(this);
6000 set_control(_gvn.transform(slow_region));
6001 uncommon_trap(Deoptimization::Reason_intrinsic,
6002 Deoptimization::Action_make_not_entrant);
6003 assert(stopped(), "Should be stopped");
6004 }
6005
6006 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6007 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6008 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6009 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6010 }
6011
6012 if (stopped()) {
6013 return true;
6014 }
6015
6016 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6017 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6018 // so the compiler has a chance to eliminate them: during macro expansion,
6019 // we have to set their control (CastPP nodes are eliminated).
6020 load_object_klass(src), load_object_klass(dest),
6021 load_array_length(src), load_array_length(dest));
6022
6023 ac->set_arraycopy(validated);
6024
6025 Node* n = _gvn.transform(ac);
6026 if (n == ac) {
6027 ac->connect_outputs(this);
6028 } else {
6029 assert(validated, "shouldn't transform if all arguments not validated");
6030 set_all_memory(n);
6031 }
6032 clear_upper_avx();
6033
6034
6035 return true;
6036 }
6037
6038
6039 // Helper function which determines if an arraycopy immediately follows
6040 // an allocation, with no intervening tests or other escapes for the object.
6041 AllocateArrayNode*
6042 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6043 if (stopped()) return nullptr; // no fast path
6044 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6045
6046 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6047 if (alloc == nullptr) return nullptr;
6048
6049 Node* rawmem = memory(Compile::AliasIdxRaw);
6050 // Is the allocation's memory state untouched?
6051 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6052 // Bail out if there have been raw-memory effects since the allocation.
6053 // (Example: There might have been a call or safepoint.)
6054 return nullptr;
6055 }
6056 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6057 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6058 return nullptr;
6059 }
6060
6061 // There must be no unexpected observers of this allocation.
6062 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6063 Node* obs = ptr->fast_out(i);
6064 if (obs != this->map()) {
6065 return nullptr;
6066 }
6067 }
6068
6069 // This arraycopy must unconditionally follow the allocation of the ptr.
6070 Node* alloc_ctl = ptr->in(0);
6071 Node* ctl = control();
6072 while (ctl != alloc_ctl) {
6073 // There may be guards which feed into the slow_region.
6074 // Any other control flow means that we might not get a chance
6075 // to finish initializing the allocated object.
6076 // Various low-level checks bottom out in uncommon traps. These
6077 // are considered safe since we've already checked above that
6078 // there is no unexpected observer of this allocation.
6079 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6080 assert(ctl->in(0)->is_If(), "must be If");
6081 ctl = ctl->in(0)->in(0);
6082 } else {
6083 return nullptr;
6084 }
6085 }
6086
6087 // If we get this far, we have an allocation which immediately
6088 // precedes the arraycopy, and we can take over zeroing the new object.
6089 // The arraycopy will finish the initialization, and provide
6090 // a new control state to which we will anchor the destination pointer.
6091
6092 return alloc;
6093 }
6094
6095 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6096 if (node->is_IfProj()) {
6097 Node* other_proj = node->as_IfProj()->other_if_proj();
6098 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6099 Node* obs = other_proj->fast_out(j);
6100 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6101 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6102 return obs->as_CallStaticJava();
6103 }
6104 }
6105 }
6106 return nullptr;
6107 }
6108
6109 //-------------inline_encodeISOArray-----------------------------------
6110 // encode char[] to byte[] in ISO_8859_1 or ASCII
6111 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6112 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6113 // no receiver since it is static method
6114 Node *src = argument(0);
6115 Node *src_offset = argument(1);
6116 Node *dst = argument(2);
6117 Node *dst_offset = argument(3);
6118 Node *length = argument(4);
6119
6120 src = must_be_not_null(src, true);
6121 dst = must_be_not_null(dst, true);
6122
6123 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6124 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6125 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6126 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6127 // failed array check
6128 return false;
6129 }
6130
6131 // Figure out the size and type of the elements we will be copying.
6132 BasicType src_elem = src_type->elem()->array_element_basic_type();
6133 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6134 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6135 return false;
6136 }
6137
6138 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6139 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6140 // 'src_start' points to src array + scaled offset
6141 // 'dst_start' points to dst array + scaled offset
6142
6143 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6144 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6145 enc = _gvn.transform(enc);
6146 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6147 set_memory(res_mem, mtype);
6148 set_result(enc);
6149 clear_upper_avx();
6150
6151 return true;
6152 }
6153
6154 //-------------inline_multiplyToLen-----------------------------------
6155 bool LibraryCallKit::inline_multiplyToLen() {
6156 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6157
6158 address stubAddr = StubRoutines::multiplyToLen();
6159 if (stubAddr == nullptr) {
6160 return false; // Intrinsic's stub is not implemented on this platform
6161 }
6162 const char* stubName = "multiplyToLen";
6163
6164 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6165
6166 // no receiver because it is a static method
6167 Node* x = argument(0);
6168 Node* xlen = argument(1);
6169 Node* y = argument(2);
6170 Node* ylen = argument(3);
6171 Node* z = argument(4);
6172
6173 x = must_be_not_null(x, true);
6174 y = must_be_not_null(y, true);
6175
6176 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6177 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6178 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6179 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6180 // failed array check
6181 return false;
6182 }
6183
6184 BasicType x_elem = x_type->elem()->array_element_basic_type();
6185 BasicType y_elem = y_type->elem()->array_element_basic_type();
6186 if (x_elem != T_INT || y_elem != T_INT) {
6187 return false;
6188 }
6189
6190 Node* x_start = array_element_address(x, intcon(0), x_elem);
6191 Node* y_start = array_element_address(y, intcon(0), y_elem);
6192 // 'x_start' points to x array + scaled xlen
6193 // 'y_start' points to y array + scaled ylen
6194
6195 Node* z_start = array_element_address(z, intcon(0), T_INT);
6196
6197 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6198 OptoRuntime::multiplyToLen_Type(),
6199 stubAddr, stubName, TypePtr::BOTTOM,
6200 x_start, xlen, y_start, ylen, z_start);
6201
6202 C->set_has_split_ifs(true); // Has chance for split-if optimization
6203 set_result(z);
6204 return true;
6205 }
6206
6207 //-------------inline_squareToLen------------------------------------
6208 bool LibraryCallKit::inline_squareToLen() {
6209 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6210
6211 address stubAddr = StubRoutines::squareToLen();
6212 if (stubAddr == nullptr) {
6213 return false; // Intrinsic's stub is not implemented on this platform
6214 }
6215 const char* stubName = "squareToLen";
6216
6217 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6218
6219 Node* x = argument(0);
6220 Node* len = argument(1);
6221 Node* z = argument(2);
6222 Node* zlen = argument(3);
6223
6224 x = must_be_not_null(x, true);
6225 z = must_be_not_null(z, true);
6226
6227 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6228 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6229 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6230 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6231 // failed array check
6232 return false;
6233 }
6234
6235 BasicType x_elem = x_type->elem()->array_element_basic_type();
6236 BasicType z_elem = z_type->elem()->array_element_basic_type();
6237 if (x_elem != T_INT || z_elem != T_INT) {
6238 return false;
6239 }
6240
6241
6242 Node* x_start = array_element_address(x, intcon(0), x_elem);
6243 Node* z_start = array_element_address(z, intcon(0), z_elem);
6244
6245 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6246 OptoRuntime::squareToLen_Type(),
6247 stubAddr, stubName, TypePtr::BOTTOM,
6248 x_start, len, z_start, zlen);
6249
6250 set_result(z);
6251 return true;
6252 }
6253
6254 //-------------inline_mulAdd------------------------------------------
6255 bool LibraryCallKit::inline_mulAdd() {
6256 assert(UseMulAddIntrinsic, "not implemented on this platform");
6257
6258 address stubAddr = StubRoutines::mulAdd();
6259 if (stubAddr == nullptr) {
6260 return false; // Intrinsic's stub is not implemented on this platform
6261 }
6262 const char* stubName = "mulAdd";
6263
6264 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6265
6266 Node* out = argument(0);
6267 Node* in = argument(1);
6268 Node* offset = argument(2);
6269 Node* len = argument(3);
6270 Node* k = argument(4);
6271
6272 in = must_be_not_null(in, true);
6273 out = must_be_not_null(out, true);
6274
6275 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6276 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6277 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6278 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6279 // failed array check
6280 return false;
6281 }
6282
6283 BasicType out_elem = out_type->elem()->array_element_basic_type();
6284 BasicType in_elem = in_type->elem()->array_element_basic_type();
6285 if (out_elem != T_INT || in_elem != T_INT) {
6286 return false;
6287 }
6288
6289 Node* outlen = load_array_length(out);
6290 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6291 Node* out_start = array_element_address(out, intcon(0), out_elem);
6292 Node* in_start = array_element_address(in, intcon(0), in_elem);
6293
6294 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6295 OptoRuntime::mulAdd_Type(),
6296 stubAddr, stubName, TypePtr::BOTTOM,
6297 out_start,in_start, new_offset, len, k);
6298 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6299 set_result(result);
6300 return true;
6301 }
6302
6303 //-------------inline_montgomeryMultiply-----------------------------------
6304 bool LibraryCallKit::inline_montgomeryMultiply() {
6305 address stubAddr = StubRoutines::montgomeryMultiply();
6306 if (stubAddr == nullptr) {
6307 return false; // Intrinsic's stub is not implemented on this platform
6308 }
6309
6310 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6311 const char* stubName = "montgomery_multiply";
6312
6313 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6314
6315 Node* a = argument(0);
6316 Node* b = argument(1);
6317 Node* n = argument(2);
6318 Node* len = argument(3);
6319 Node* inv = argument(4);
6320 Node* m = argument(6);
6321
6322 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6323 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6324 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6325 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6326 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6327 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6328 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6329 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6330 // failed array check
6331 return false;
6332 }
6333
6334 BasicType a_elem = a_type->elem()->array_element_basic_type();
6335 BasicType b_elem = b_type->elem()->array_element_basic_type();
6336 BasicType n_elem = n_type->elem()->array_element_basic_type();
6337 BasicType m_elem = m_type->elem()->array_element_basic_type();
6338 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6339 return false;
6340 }
6341
6342 // Make the call
6343 {
6344 Node* a_start = array_element_address(a, intcon(0), a_elem);
6345 Node* b_start = array_element_address(b, intcon(0), b_elem);
6346 Node* n_start = array_element_address(n, intcon(0), n_elem);
6347 Node* m_start = array_element_address(m, intcon(0), m_elem);
6348
6349 Node* call = make_runtime_call(RC_LEAF,
6350 OptoRuntime::montgomeryMultiply_Type(),
6351 stubAddr, stubName, TypePtr::BOTTOM,
6352 a_start, b_start, n_start, len, inv, top(),
6353 m_start);
6354 set_result(m);
6355 }
6356
6357 return true;
6358 }
6359
6360 bool LibraryCallKit::inline_montgomerySquare() {
6361 address stubAddr = StubRoutines::montgomerySquare();
6362 if (stubAddr == nullptr) {
6363 return false; // Intrinsic's stub is not implemented on this platform
6364 }
6365
6366 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6367 const char* stubName = "montgomery_square";
6368
6369 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6370
6371 Node* a = argument(0);
6372 Node* n = argument(1);
6373 Node* len = argument(2);
6374 Node* inv = argument(3);
6375 Node* m = argument(5);
6376
6377 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6378 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6379 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6380 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6381 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6382 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6383 // failed array check
6384 return false;
6385 }
6386
6387 BasicType a_elem = a_type->elem()->array_element_basic_type();
6388 BasicType n_elem = n_type->elem()->array_element_basic_type();
6389 BasicType m_elem = m_type->elem()->array_element_basic_type();
6390 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6391 return false;
6392 }
6393
6394 // Make the call
6395 {
6396 Node* a_start = array_element_address(a, intcon(0), a_elem);
6397 Node* n_start = array_element_address(n, intcon(0), n_elem);
6398 Node* m_start = array_element_address(m, intcon(0), m_elem);
6399
6400 Node* call = make_runtime_call(RC_LEAF,
6401 OptoRuntime::montgomerySquare_Type(),
6402 stubAddr, stubName, TypePtr::BOTTOM,
6403 a_start, n_start, len, inv, top(),
6404 m_start);
6405 set_result(m);
6406 }
6407
6408 return true;
6409 }
6410
6411 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6412 address stubAddr = nullptr;
6413 const char* stubName = nullptr;
6414
6415 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6416 if (stubAddr == nullptr) {
6417 return false; // Intrinsic's stub is not implemented on this platform
6418 }
6419
6420 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6421
6422 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6423
6424 Node* newArr = argument(0);
6425 Node* oldArr = argument(1);
6426 Node* newIdx = argument(2);
6427 Node* shiftCount = argument(3);
6428 Node* numIter = argument(4);
6429
6430 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6431 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6432 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6433 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6434 return false;
6435 }
6436
6437 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6438 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6439 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6440 return false;
6441 }
6442
6443 // Make the call
6444 {
6445 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6446 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6447
6448 Node* call = make_runtime_call(RC_LEAF,
6449 OptoRuntime::bigIntegerShift_Type(),
6450 stubAddr,
6451 stubName,
6452 TypePtr::BOTTOM,
6453 newArr_start,
6454 oldArr_start,
6455 newIdx,
6456 shiftCount,
6457 numIter);
6458 }
6459
6460 return true;
6461 }
6462
6463 //-------------inline_vectorizedMismatch------------------------------
6464 bool LibraryCallKit::inline_vectorizedMismatch() {
6465 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6466
6467 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6468 Node* obja = argument(0); // Object
6469 Node* aoffset = argument(1); // long
6470 Node* objb = argument(3); // Object
6471 Node* boffset = argument(4); // long
6472 Node* length = argument(6); // int
6473 Node* scale = argument(7); // int
6474
6475 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6476 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6477 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6478 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6479 scale == top()) {
6480 return false; // failed input validation
6481 }
6482
6483 Node* obja_adr = make_unsafe_address(obja, aoffset);
6484 Node* objb_adr = make_unsafe_address(objb, boffset);
6485
6486 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6487 //
6488 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6489 // if (length <= inline_limit) {
6490 // inline_path:
6491 // vmask = VectorMaskGen length
6492 // vload1 = LoadVectorMasked obja, vmask
6493 // vload2 = LoadVectorMasked objb, vmask
6494 // result1 = VectorCmpMasked vload1, vload2, vmask
6495 // } else {
6496 // call_stub_path:
6497 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6498 // }
6499 // exit_block:
6500 // return Phi(result1, result2);
6501 //
6502 enum { inline_path = 1, // input is small enough to process it all at once
6503 stub_path = 2, // input is too large; call into the VM
6504 PATH_LIMIT = 3
6505 };
6506
6507 Node* exit_block = new RegionNode(PATH_LIMIT);
6508 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6509 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6510
6511 Node* call_stub_path = control();
6512
6513 BasicType elem_bt = T_ILLEGAL;
6514
6515 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6516 if (scale_t->is_con()) {
6517 switch (scale_t->get_con()) {
6518 case 0: elem_bt = T_BYTE; break;
6519 case 1: elem_bt = T_SHORT; break;
6520 case 2: elem_bt = T_INT; break;
6521 case 3: elem_bt = T_LONG; break;
6522
6523 default: elem_bt = T_ILLEGAL; break; // not supported
6524 }
6525 }
6526
6527 int inline_limit = 0;
6528 bool do_partial_inline = false;
6529
6530 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6531 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6532 do_partial_inline = inline_limit >= 16;
6533 }
6534
6535 if (do_partial_inline) {
6536 assert(elem_bt != T_ILLEGAL, "sanity");
6537
6538 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6539 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6540 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6541
6542 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6543 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6544 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6545
6546 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6547
6548 if (!stopped()) {
6549 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6550
6551 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6552 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6553 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6554 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6555
6556 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6557 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6558 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6559 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6560
6561 exit_block->init_req(inline_path, control());
6562 memory_phi->init_req(inline_path, map()->memory());
6563 result_phi->init_req(inline_path, result);
6564
6565 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6566 clear_upper_avx();
6567 }
6568 }
6569 }
6570
6571 if (call_stub_path != nullptr) {
6572 set_control(call_stub_path);
6573
6574 Node* call = make_runtime_call(RC_LEAF,
6575 OptoRuntime::vectorizedMismatch_Type(),
6576 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6577 obja_adr, objb_adr, length, scale);
6578
6579 exit_block->init_req(stub_path, control());
6580 memory_phi->init_req(stub_path, map()->memory());
6581 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6582 }
6583
6584 exit_block = _gvn.transform(exit_block);
6585 memory_phi = _gvn.transform(memory_phi);
6586 result_phi = _gvn.transform(result_phi);
6587
6588 record_for_igvn(exit_block);
6589 record_for_igvn(memory_phi);
6590 record_for_igvn(result_phi);
6591
6592 set_control(exit_block);
6593 set_all_memory(memory_phi);
6594 set_result(result_phi);
6595
6596 return true;
6597 }
6598
6599 //------------------------------inline_vectorizedHashcode----------------------------
6600 bool LibraryCallKit::inline_vectorizedHashCode() {
6601 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6602
6603 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6604 Node* array = argument(0);
6605 Node* offset = argument(1);
6606 Node* length = argument(2);
6607 Node* initialValue = argument(3);
6608 Node* basic_type = argument(4);
6609
6610 if (basic_type == top()) {
6611 return false; // failed input validation
6612 }
6613
6614 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6615 if (!basic_type_t->is_con()) {
6616 return false; // Only intrinsify if mode argument is constant
6617 }
6618
6619 array = must_be_not_null(array, true);
6620
6621 BasicType bt = (BasicType)basic_type_t->get_con();
6622
6623 // Resolve address of first element
6624 Node* array_start = array_element_address(array, offset, bt);
6625
6626 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6627 array_start, length, initialValue, basic_type)));
6628 clear_upper_avx();
6629
6630 return true;
6631 }
6632
6633 /**
6634 * Calculate CRC32 for byte.
6635 * int java.util.zip.CRC32.update(int crc, int b)
6636 */
6637 bool LibraryCallKit::inline_updateCRC32() {
6638 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6639 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6640 // no receiver since it is static method
6641 Node* crc = argument(0); // type: int
6642 Node* b = argument(1); // type: int
6643
6644 /*
6645 * int c = ~ crc;
6646 * b = timesXtoThe32[(b ^ c) & 0xFF];
6647 * b = b ^ (c >>> 8);
6648 * crc = ~b;
6649 */
6650
6651 Node* M1 = intcon(-1);
6652 crc = _gvn.transform(new XorINode(crc, M1));
6653 Node* result = _gvn.transform(new XorINode(crc, b));
6654 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6655
6656 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6657 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6658 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6659 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6660
6661 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6662 result = _gvn.transform(new XorINode(crc, result));
6663 result = _gvn.transform(new XorINode(result, M1));
6664 set_result(result);
6665 return true;
6666 }
6667
6668 /**
6669 * Calculate CRC32 for byte[] array.
6670 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6671 */
6672 bool LibraryCallKit::inline_updateBytesCRC32() {
6673 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6674 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6675 // no receiver since it is static method
6676 Node* crc = argument(0); // type: int
6677 Node* src = argument(1); // type: oop
6678 Node* offset = argument(2); // type: int
6679 Node* length = argument(3); // type: int
6680
6681 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6682 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6683 // failed array check
6684 return false;
6685 }
6686
6687 // Figure out the size and type of the elements we will be copying.
6688 BasicType src_elem = src_type->elem()->array_element_basic_type();
6689 if (src_elem != T_BYTE) {
6690 return false;
6691 }
6692
6693 // 'src_start' points to src array + scaled offset
6694 src = must_be_not_null(src, true);
6695 Node* src_start = array_element_address(src, offset, src_elem);
6696
6697 // We assume that range check is done by caller.
6698 // TODO: generate range check (offset+length < src.length) in debug VM.
6699
6700 // Call the stub.
6701 address stubAddr = StubRoutines::updateBytesCRC32();
6702 const char *stubName = "updateBytesCRC32";
6703
6704 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6705 stubAddr, stubName, TypePtr::BOTTOM,
6706 crc, src_start, length);
6707 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6708 set_result(result);
6709 return true;
6710 }
6711
6712 /**
6713 * Calculate CRC32 for ByteBuffer.
6714 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6715 */
6716 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6717 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6718 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6719 // no receiver since it is static method
6720 Node* crc = argument(0); // type: int
6721 Node* src = argument(1); // type: long
6722 Node* offset = argument(3); // type: int
6723 Node* length = argument(4); // type: int
6724
6725 src = ConvL2X(src); // adjust Java long to machine word
6726 Node* base = _gvn.transform(new CastX2PNode(src));
6727 offset = ConvI2X(offset);
6728
6729 // 'src_start' points to src array + scaled offset
6730 Node* src_start = basic_plus_adr(top(), base, offset);
6731
6732 // Call the stub.
6733 address stubAddr = StubRoutines::updateBytesCRC32();
6734 const char *stubName = "updateBytesCRC32";
6735
6736 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6737 stubAddr, stubName, TypePtr::BOTTOM,
6738 crc, src_start, length);
6739 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6740 set_result(result);
6741 return true;
6742 }
6743
6744 //------------------------------get_table_from_crc32c_class-----------------------
6745 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6746 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6747 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6748
6749 return table;
6750 }
6751
6752 //------------------------------inline_updateBytesCRC32C-----------------------
6753 //
6754 // Calculate CRC32C for byte[] array.
6755 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6756 //
6757 bool LibraryCallKit::inline_updateBytesCRC32C() {
6758 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6759 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6760 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6761 // no receiver since it is a static method
6762 Node* crc = argument(0); // type: int
6763 Node* src = argument(1); // type: oop
6764 Node* offset = argument(2); // type: int
6765 Node* end = argument(3); // type: int
6766
6767 Node* length = _gvn.transform(new SubINode(end, offset));
6768
6769 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6770 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6771 // failed array check
6772 return false;
6773 }
6774
6775 // Figure out the size and type of the elements we will be copying.
6776 BasicType src_elem = src_type->elem()->array_element_basic_type();
6777 if (src_elem != T_BYTE) {
6778 return false;
6779 }
6780
6781 // 'src_start' points to src array + scaled offset
6782 src = must_be_not_null(src, true);
6783 Node* src_start = array_element_address(src, offset, src_elem);
6784
6785 // static final int[] byteTable in class CRC32C
6786 Node* table = get_table_from_crc32c_class(callee()->holder());
6787 table = must_be_not_null(table, true);
6788 Node* table_start = array_element_address(table, intcon(0), T_INT);
6789
6790 // We assume that range check is done by caller.
6791 // TODO: generate range check (offset+length < src.length) in debug VM.
6792
6793 // Call the stub.
6794 address stubAddr = StubRoutines::updateBytesCRC32C();
6795 const char *stubName = "updateBytesCRC32C";
6796
6797 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6798 stubAddr, stubName, TypePtr::BOTTOM,
6799 crc, src_start, length, table_start);
6800 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6801 set_result(result);
6802 return true;
6803 }
6804
6805 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6806 //
6807 // Calculate CRC32C for DirectByteBuffer.
6808 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6809 //
6810 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6811 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6812 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6813 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6814 // no receiver since it is a static method
6815 Node* crc = argument(0); // type: int
6816 Node* src = argument(1); // type: long
6817 Node* offset = argument(3); // type: int
6818 Node* end = argument(4); // type: int
6819
6820 Node* length = _gvn.transform(new SubINode(end, offset));
6821
6822 src = ConvL2X(src); // adjust Java long to machine word
6823 Node* base = _gvn.transform(new CastX2PNode(src));
6824 offset = ConvI2X(offset);
6825
6826 // 'src_start' points to src array + scaled offset
6827 Node* src_start = basic_plus_adr(top(), base, offset);
6828
6829 // static final int[] byteTable in class CRC32C
6830 Node* table = get_table_from_crc32c_class(callee()->holder());
6831 table = must_be_not_null(table, true);
6832 Node* table_start = array_element_address(table, intcon(0), T_INT);
6833
6834 // Call the stub.
6835 address stubAddr = StubRoutines::updateBytesCRC32C();
6836 const char *stubName = "updateBytesCRC32C";
6837
6838 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6839 stubAddr, stubName, TypePtr::BOTTOM,
6840 crc, src_start, length, table_start);
6841 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6842 set_result(result);
6843 return true;
6844 }
6845
6846 //------------------------------inline_updateBytesAdler32----------------------
6847 //
6848 // Calculate Adler32 checksum for byte[] array.
6849 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6850 //
6851 bool LibraryCallKit::inline_updateBytesAdler32() {
6852 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6853 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6854 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6855 // no receiver since it is static method
6856 Node* crc = argument(0); // type: int
6857 Node* src = argument(1); // type: oop
6858 Node* offset = argument(2); // type: int
6859 Node* length = argument(3); // type: int
6860
6861 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6862 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6863 // failed array check
6864 return false;
6865 }
6866
6867 // Figure out the size and type of the elements we will be copying.
6868 BasicType src_elem = src_type->elem()->array_element_basic_type();
6869 if (src_elem != T_BYTE) {
6870 return false;
6871 }
6872
6873 // 'src_start' points to src array + scaled offset
6874 Node* src_start = array_element_address(src, offset, src_elem);
6875
6876 // We assume that range check is done by caller.
6877 // TODO: generate range check (offset+length < src.length) in debug VM.
6878
6879 // Call the stub.
6880 address stubAddr = StubRoutines::updateBytesAdler32();
6881 const char *stubName = "updateBytesAdler32";
6882
6883 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6884 stubAddr, stubName, TypePtr::BOTTOM,
6885 crc, src_start, length);
6886 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6887 set_result(result);
6888 return true;
6889 }
6890
6891 //------------------------------inline_updateByteBufferAdler32---------------
6892 //
6893 // Calculate Adler32 checksum for DirectByteBuffer.
6894 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6895 //
6896 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6897 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6898 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6899 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6900 // no receiver since it is static method
6901 Node* crc = argument(0); // type: int
6902 Node* src = argument(1); // type: long
6903 Node* offset = argument(3); // type: int
6904 Node* length = argument(4); // type: int
6905
6906 src = ConvL2X(src); // adjust Java long to machine word
6907 Node* base = _gvn.transform(new CastX2PNode(src));
6908 offset = ConvI2X(offset);
6909
6910 // 'src_start' points to src array + scaled offset
6911 Node* src_start = basic_plus_adr(top(), base, offset);
6912
6913 // Call the stub.
6914 address stubAddr = StubRoutines::updateBytesAdler32();
6915 const char *stubName = "updateBytesAdler32";
6916
6917 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6918 stubAddr, stubName, TypePtr::BOTTOM,
6919 crc, src_start, length);
6920
6921 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6922 set_result(result);
6923 return true;
6924 }
6925
6926 //----------------------------inline_reference_get0----------------------------
6927 // public T java.lang.ref.Reference.get();
6928 bool LibraryCallKit::inline_reference_get0() {
6929 const int referent_offset = java_lang_ref_Reference::referent_offset();
6930
6931 // Get the argument:
6932 Node* reference_obj = null_check_receiver();
6933 if (stopped()) return true;
6934
6935 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6936 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6937 decorators, /*is_static*/ false, nullptr);
6938 if (result == nullptr) return false;
6939
6940 // Add memory barrier to prevent commoning reads from this field
6941 // across safepoint since GC can change its value.
6942 insert_mem_bar(Op_MemBarCPUOrder);
6943
6944 set_result(result);
6945 return true;
6946 }
6947
6948 //----------------------------inline_reference_refersTo0----------------------------
6949 // bool java.lang.ref.Reference.refersTo0();
6950 // bool java.lang.ref.PhantomReference.refersTo0();
6951 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6952 // Get arguments:
6953 Node* reference_obj = null_check_receiver();
6954 Node* other_obj = argument(1);
6955 if (stopped()) return true;
6956
6957 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6958 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6959 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6960 decorators, /*is_static*/ false, nullptr);
6961 if (referent == nullptr) return false;
6962
6963 // Add memory barrier to prevent commoning reads from this field
6964 // across safepoint since GC can change its value.
6965 insert_mem_bar(Op_MemBarCPUOrder);
6966
6967 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6968 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6969 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6970
6971 RegionNode* region = new RegionNode(3);
6972 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6973
6974 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6975 region->init_req(1, if_true);
6976 phi->init_req(1, intcon(1));
6977
6978 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6979 region->init_req(2, if_false);
6980 phi->init_req(2, intcon(0));
6981
6982 set_control(_gvn.transform(region));
6983 record_for_igvn(region);
6984 set_result(_gvn.transform(phi));
6985 return true;
6986 }
6987
6988 //----------------------------inline_reference_clear0----------------------------
6989 // void java.lang.ref.Reference.clear0();
6990 // void java.lang.ref.PhantomReference.clear0();
6991 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6992 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6993
6994 // Get arguments
6995 Node* reference_obj = null_check_receiver();
6996 if (stopped()) return true;
6997
6998 // Common access parameters
6999 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7000 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7001 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7002 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7003 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7004
7005 Node* referent = access_load_at(reference_obj,
7006 referent_field_addr,
7007 referent_field_addr_type,
7008 val_type,
7009 T_OBJECT,
7010 decorators);
7011
7012 IdealKit ideal(this);
7013 #define __ ideal.
7014 __ if_then(referent, BoolTest::ne, null());
7015 sync_kit(ideal);
7016 access_store_at(reference_obj,
7017 referent_field_addr,
7018 referent_field_addr_type,
7019 null(),
7020 val_type,
7021 T_OBJECT,
7022 decorators);
7023 __ sync_kit(this);
7024 __ end_if();
7025 final_sync(ideal);
7026 #undef __
7027
7028 return true;
7029 }
7030
7031 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7032 DecoratorSet decorators, bool is_static,
7033 ciInstanceKlass* fromKls) {
7034 if (fromKls == nullptr) {
7035 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7036 assert(tinst != nullptr, "obj is null");
7037 assert(tinst->is_loaded(), "obj is not loaded");
7038 fromKls = tinst->instance_klass();
7039 } else {
7040 assert(is_static, "only for static field access");
7041 }
7042 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7043 ciSymbol::make(fieldTypeString),
7044 is_static);
7045
7046 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7047 if (field == nullptr) return (Node *) nullptr;
7048
7049 if (is_static) {
7050 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7051 fromObj = makecon(tip);
7052 }
7053
7054 // Next code copied from Parse::do_get_xxx():
7055
7056 // Compute address and memory type.
7057 int offset = field->offset_in_bytes();
7058 bool is_vol = field->is_volatile();
7059 ciType* field_klass = field->type();
7060 assert(field_klass->is_loaded(), "should be loaded");
7061 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7062 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7063 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7064 "slice of address and input slice don't match");
7065 BasicType bt = field->layout_type();
7066
7067 // Build the resultant type of the load
7068 const Type *type;
7069 if (bt == T_OBJECT) {
7070 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7071 } else {
7072 type = Type::get_const_basic_type(bt);
7073 }
7074
7075 if (is_vol) {
7076 decorators |= MO_SEQ_CST;
7077 }
7078
7079 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7080 }
7081
7082 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7083 bool is_exact /* true */, bool is_static /* false */,
7084 ciInstanceKlass * fromKls /* nullptr */) {
7085 if (fromKls == nullptr) {
7086 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7087 assert(tinst != nullptr, "obj is null");
7088 assert(tinst->is_loaded(), "obj is not loaded");
7089 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7090 fromKls = tinst->instance_klass();
7091 }
7092 else {
7093 assert(is_static, "only for static field access");
7094 }
7095 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7096 ciSymbol::make(fieldTypeString),
7097 is_static);
7098
7099 assert(field != nullptr, "undefined field");
7100 assert(!field->is_volatile(), "not defined for volatile fields");
7101
7102 if (is_static) {
7103 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7104 fromObj = makecon(tip);
7105 }
7106
7107 // Next code copied from Parse::do_get_xxx():
7108
7109 // Compute address and memory type.
7110 int offset = field->offset_in_bytes();
7111 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7112
7113 return adr;
7114 }
7115
7116 //------------------------------inline_aescrypt_Block-----------------------
7117 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7118 address stubAddr = nullptr;
7119 const char *stubName;
7120 assert(UseAES, "need AES instruction support");
7121
7122 switch(id) {
7123 case vmIntrinsics::_aescrypt_encryptBlock:
7124 stubAddr = StubRoutines::aescrypt_encryptBlock();
7125 stubName = "aescrypt_encryptBlock";
7126 break;
7127 case vmIntrinsics::_aescrypt_decryptBlock:
7128 stubAddr = StubRoutines::aescrypt_decryptBlock();
7129 stubName = "aescrypt_decryptBlock";
7130 break;
7131 default:
7132 break;
7133 }
7134 if (stubAddr == nullptr) return false;
7135
7136 Node* aescrypt_object = argument(0);
7137 Node* src = argument(1);
7138 Node* src_offset = argument(2);
7139 Node* dest = argument(3);
7140 Node* dest_offset = argument(4);
7141
7142 src = must_be_not_null(src, true);
7143 dest = must_be_not_null(dest, true);
7144
7145 // (1) src and dest are arrays.
7146 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7147 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7148 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7149 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7150
7151 // for the quick and dirty code we will skip all the checks.
7152 // we are just trying to get the call to be generated.
7153 Node* src_start = src;
7154 Node* dest_start = dest;
7155 if (src_offset != nullptr || dest_offset != nullptr) {
7156 assert(src_offset != nullptr && dest_offset != nullptr, "");
7157 src_start = array_element_address(src, src_offset, T_BYTE);
7158 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7159 }
7160
7161 // now need to get the start of its expanded key array
7162 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7163 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7164 if (k_start == nullptr) return false;
7165
7166 // Call the stub.
7167 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7168 stubAddr, stubName, TypePtr::BOTTOM,
7169 src_start, dest_start, k_start);
7170
7171 return true;
7172 }
7173
7174 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7175 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7176 address stubAddr = nullptr;
7177 const char *stubName = nullptr;
7178
7179 assert(UseAES, "need AES instruction support");
7180
7181 switch(id) {
7182 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7183 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7184 stubName = "cipherBlockChaining_encryptAESCrypt";
7185 break;
7186 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7187 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7188 stubName = "cipherBlockChaining_decryptAESCrypt";
7189 break;
7190 default:
7191 break;
7192 }
7193 if (stubAddr == nullptr) return false;
7194
7195 Node* cipherBlockChaining_object = argument(0);
7196 Node* src = argument(1);
7197 Node* src_offset = argument(2);
7198 Node* len = argument(3);
7199 Node* dest = argument(4);
7200 Node* dest_offset = argument(5);
7201
7202 src = must_be_not_null(src, false);
7203 dest = must_be_not_null(dest, false);
7204
7205 // (1) src and dest are arrays.
7206 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7207 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7208 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7209 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7210
7211 // checks are the responsibility of the caller
7212 Node* src_start = src;
7213 Node* dest_start = dest;
7214 if (src_offset != nullptr || dest_offset != nullptr) {
7215 assert(src_offset != nullptr && dest_offset != nullptr, "");
7216 src_start = array_element_address(src, src_offset, T_BYTE);
7217 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7218 }
7219
7220 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7221 // (because of the predicated logic executed earlier).
7222 // so we cast it here safely.
7223 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7224
7225 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7226 if (embeddedCipherObj == nullptr) return false;
7227
7228 // cast it to what we know it will be at runtime
7229 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7230 assert(tinst != nullptr, "CBC obj is null");
7231 assert(tinst->is_loaded(), "CBC obj is not loaded");
7232 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7233 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7234
7235 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7236 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7237 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7238 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7239 aescrypt_object = _gvn.transform(aescrypt_object);
7240
7241 // we need to get the start of the aescrypt_object's expanded key array
7242 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7243 if (k_start == nullptr) return false;
7244
7245 // similarly, get the start address of the r vector
7246 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7247 if (objRvec == nullptr) return false;
7248 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7249
7250 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7251 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7252 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7253 stubAddr, stubName, TypePtr::BOTTOM,
7254 src_start, dest_start, k_start, r_start, len);
7255
7256 // return cipher length (int)
7257 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7258 set_result(retvalue);
7259 return true;
7260 }
7261
7262 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7263 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7264 address stubAddr = nullptr;
7265 const char *stubName = nullptr;
7266
7267 assert(UseAES, "need AES instruction support");
7268
7269 switch (id) {
7270 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7271 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7272 stubName = "electronicCodeBook_encryptAESCrypt";
7273 break;
7274 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7275 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7276 stubName = "electronicCodeBook_decryptAESCrypt";
7277 break;
7278 default:
7279 break;
7280 }
7281
7282 if (stubAddr == nullptr) return false;
7283
7284 Node* electronicCodeBook_object = argument(0);
7285 Node* src = argument(1);
7286 Node* src_offset = argument(2);
7287 Node* len = argument(3);
7288 Node* dest = argument(4);
7289 Node* dest_offset = argument(5);
7290
7291 // (1) src and dest are arrays.
7292 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7293 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7294 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7295 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7296
7297 // checks are the responsibility of the caller
7298 Node* src_start = src;
7299 Node* dest_start = dest;
7300 if (src_offset != nullptr || dest_offset != nullptr) {
7301 assert(src_offset != nullptr && dest_offset != nullptr, "");
7302 src_start = array_element_address(src, src_offset, T_BYTE);
7303 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7304 }
7305
7306 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7307 // (because of the predicated logic executed earlier).
7308 // so we cast it here safely.
7309 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7310
7311 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7312 if (embeddedCipherObj == nullptr) return false;
7313
7314 // cast it to what we know it will be at runtime
7315 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7316 assert(tinst != nullptr, "ECB obj is null");
7317 assert(tinst->is_loaded(), "ECB obj is not loaded");
7318 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7319 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7320
7321 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7322 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7323 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7324 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7325 aescrypt_object = _gvn.transform(aescrypt_object);
7326
7327 // we need to get the start of the aescrypt_object's expanded key array
7328 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7329 if (k_start == nullptr) return false;
7330
7331 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7332 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7333 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7334 stubAddr, stubName, TypePtr::BOTTOM,
7335 src_start, dest_start, k_start, len);
7336
7337 // return cipher length (int)
7338 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7339 set_result(retvalue);
7340 return true;
7341 }
7342
7343 //------------------------------inline_counterMode_AESCrypt-----------------------
7344 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7345 assert(UseAES, "need AES instruction support");
7346 if (!UseAESCTRIntrinsics) return false;
7347
7348 address stubAddr = nullptr;
7349 const char *stubName = nullptr;
7350 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7351 stubAddr = StubRoutines::counterMode_AESCrypt();
7352 stubName = "counterMode_AESCrypt";
7353 }
7354 if (stubAddr == nullptr) return false;
7355
7356 Node* counterMode_object = argument(0);
7357 Node* src = argument(1);
7358 Node* src_offset = argument(2);
7359 Node* len = argument(3);
7360 Node* dest = argument(4);
7361 Node* dest_offset = argument(5);
7362
7363 // (1) src and dest are arrays.
7364 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7365 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7366 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7367 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7368
7369 // checks are the responsibility of the caller
7370 Node* src_start = src;
7371 Node* dest_start = dest;
7372 if (src_offset != nullptr || dest_offset != nullptr) {
7373 assert(src_offset != nullptr && dest_offset != nullptr, "");
7374 src_start = array_element_address(src, src_offset, T_BYTE);
7375 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7376 }
7377
7378 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7379 // (because of the predicated logic executed earlier).
7380 // so we cast it here safely.
7381 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7382 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7383 if (embeddedCipherObj == nullptr) return false;
7384 // cast it to what we know it will be at runtime
7385 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7386 assert(tinst != nullptr, "CTR obj is null");
7387 assert(tinst->is_loaded(), "CTR obj is not loaded");
7388 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7389 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7390 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7391 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7392 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7393 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7394 aescrypt_object = _gvn.transform(aescrypt_object);
7395 // we need to get the start of the aescrypt_object's expanded key array
7396 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7397 if (k_start == nullptr) return false;
7398 // similarly, get the start address of the r vector
7399 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7400 if (obj_counter == nullptr) return false;
7401 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7402
7403 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7404 if (saved_encCounter == nullptr) return false;
7405 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7406 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7407
7408 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7409 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7410 OptoRuntime::counterMode_aescrypt_Type(),
7411 stubAddr, stubName, TypePtr::BOTTOM,
7412 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7413
7414 // return cipher length (int)
7415 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7416 set_result(retvalue);
7417 return true;
7418 }
7419
7420 //------------------------------get_key_start_from_aescrypt_object-----------------------
7421 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7422 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7423 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7424 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7425 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7426 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7427 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7428 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7429 if (objSessionK == nullptr) {
7430 return (Node *) nullptr;
7431 }
7432 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7433 #else
7434 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7435 #endif // PPC64
7436 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7437 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7438
7439 // now have the array, need to get the start address of the K array
7440 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7441 return k_start;
7442 }
7443
7444 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7445 // Return node representing slow path of predicate check.
7446 // the pseudo code we want to emulate with this predicate is:
7447 // for encryption:
7448 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7449 // for decryption:
7450 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7451 // note cipher==plain is more conservative than the original java code but that's OK
7452 //
7453 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7454 // The receiver was checked for null already.
7455 Node* objCBC = argument(0);
7456
7457 Node* src = argument(1);
7458 Node* dest = argument(4);
7459
7460 // Load embeddedCipher field of CipherBlockChaining object.
7461 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7462
7463 // get AESCrypt klass for instanceOf check
7464 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7465 // will have same classloader as CipherBlockChaining object
7466 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7467 assert(tinst != nullptr, "CBCobj is null");
7468 assert(tinst->is_loaded(), "CBCobj is not loaded");
7469
7470 // we want to do an instanceof comparison against the AESCrypt class
7471 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7472 if (!klass_AESCrypt->is_loaded()) {
7473 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7474 Node* ctrl = control();
7475 set_control(top()); // no regular fast path
7476 return ctrl;
7477 }
7478
7479 src = must_be_not_null(src, true);
7480 dest = must_be_not_null(dest, true);
7481
7482 // Resolve oops to stable for CmpP below.
7483 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7484
7485 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7486 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7487 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7488
7489 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7490
7491 // for encryption, we are done
7492 if (!decrypting)
7493 return instof_false; // even if it is null
7494
7495 // for decryption, we need to add a further check to avoid
7496 // taking the intrinsic path when cipher and plain are the same
7497 // see the original java code for why.
7498 RegionNode* region = new RegionNode(3);
7499 region->init_req(1, instof_false);
7500
7501 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7502 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7503 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7504 region->init_req(2, src_dest_conjoint);
7505
7506 record_for_igvn(region);
7507 return _gvn.transform(region);
7508 }
7509
7510 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7511 // Return node representing slow path of predicate check.
7512 // the pseudo code we want to emulate with this predicate is:
7513 // for encryption:
7514 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7515 // for decryption:
7516 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7517 // note cipher==plain is more conservative than the original java code but that's OK
7518 //
7519 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7520 // The receiver was checked for null already.
7521 Node* objECB = argument(0);
7522
7523 // Load embeddedCipher field of ElectronicCodeBook object.
7524 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7525
7526 // get AESCrypt klass for instanceOf check
7527 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7528 // will have same classloader as ElectronicCodeBook object
7529 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7530 assert(tinst != nullptr, "ECBobj is null");
7531 assert(tinst->is_loaded(), "ECBobj is not loaded");
7532
7533 // we want to do an instanceof comparison against the AESCrypt class
7534 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7535 if (!klass_AESCrypt->is_loaded()) {
7536 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7537 Node* ctrl = control();
7538 set_control(top()); // no regular fast path
7539 return ctrl;
7540 }
7541 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7542
7543 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7544 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7545 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7546
7547 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7548
7549 // for encryption, we are done
7550 if (!decrypting)
7551 return instof_false; // even if it is null
7552
7553 // for decryption, we need to add a further check to avoid
7554 // taking the intrinsic path when cipher and plain are the same
7555 // see the original java code for why.
7556 RegionNode* region = new RegionNode(3);
7557 region->init_req(1, instof_false);
7558 Node* src = argument(1);
7559 Node* dest = argument(4);
7560 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7561 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7562 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7563 region->init_req(2, src_dest_conjoint);
7564
7565 record_for_igvn(region);
7566 return _gvn.transform(region);
7567 }
7568
7569 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7570 // Return node representing slow path of predicate check.
7571 // the pseudo code we want to emulate with this predicate is:
7572 // for encryption:
7573 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7574 // for decryption:
7575 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7576 // note cipher==plain is more conservative than the original java code but that's OK
7577 //
7578
7579 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7580 // The receiver was checked for null already.
7581 Node* objCTR = argument(0);
7582
7583 // Load embeddedCipher field of CipherBlockChaining object.
7584 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7585
7586 // get AESCrypt klass for instanceOf check
7587 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7588 // will have same classloader as CipherBlockChaining object
7589 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7590 assert(tinst != nullptr, "CTRobj is null");
7591 assert(tinst->is_loaded(), "CTRobj is not loaded");
7592
7593 // we want to do an instanceof comparison against the AESCrypt class
7594 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7595 if (!klass_AESCrypt->is_loaded()) {
7596 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7597 Node* ctrl = control();
7598 set_control(top()); // no regular fast path
7599 return ctrl;
7600 }
7601
7602 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7603 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7604 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7605 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7606 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7607
7608 return instof_false; // even if it is null
7609 }
7610
7611 //------------------------------inline_ghash_processBlocks
7612 bool LibraryCallKit::inline_ghash_processBlocks() {
7613 address stubAddr;
7614 const char *stubName;
7615 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7616
7617 stubAddr = StubRoutines::ghash_processBlocks();
7618 stubName = "ghash_processBlocks";
7619
7620 Node* data = argument(0);
7621 Node* offset = argument(1);
7622 Node* len = argument(2);
7623 Node* state = argument(3);
7624 Node* subkeyH = argument(4);
7625
7626 state = must_be_not_null(state, true);
7627 subkeyH = must_be_not_null(subkeyH, true);
7628 data = must_be_not_null(data, true);
7629
7630 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7631 assert(state_start, "state is null");
7632 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7633 assert(subkeyH_start, "subkeyH is null");
7634 Node* data_start = array_element_address(data, offset, T_BYTE);
7635 assert(data_start, "data is null");
7636
7637 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7638 OptoRuntime::ghash_processBlocks_Type(),
7639 stubAddr, stubName, TypePtr::BOTTOM,
7640 state_start, subkeyH_start, data_start, len);
7641 return true;
7642 }
7643
7644 //------------------------------inline_chacha20Block
7645 bool LibraryCallKit::inline_chacha20Block() {
7646 address stubAddr;
7647 const char *stubName;
7648 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7649
7650 stubAddr = StubRoutines::chacha20Block();
7651 stubName = "chacha20Block";
7652
7653 Node* state = argument(0);
7654 Node* result = argument(1);
7655
7656 state = must_be_not_null(state, true);
7657 result = must_be_not_null(result, true);
7658
7659 Node* state_start = array_element_address(state, intcon(0), T_INT);
7660 assert(state_start, "state is null");
7661 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7662 assert(result_start, "result is null");
7663
7664 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7665 OptoRuntime::chacha20Block_Type(),
7666 stubAddr, stubName, TypePtr::BOTTOM,
7667 state_start, result_start);
7668 // return key stream length (int)
7669 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7670 set_result(retvalue);
7671 return true;
7672 }
7673
7674 //------------------------------inline_kyberNtt
7675 bool LibraryCallKit::inline_kyberNtt() {
7676 address stubAddr;
7677 const char *stubName;
7678 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7679 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7680
7681 stubAddr = StubRoutines::kyberNtt();
7682 stubName = "kyberNtt";
7683 if (!stubAddr) return false;
7684
7685 Node* coeffs = argument(0);
7686 Node* ntt_zetas = argument(1);
7687
7688 coeffs = must_be_not_null(coeffs, true);
7689 ntt_zetas = must_be_not_null(ntt_zetas, true);
7690
7691 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7692 assert(coeffs_start, "coeffs is null");
7693 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7694 assert(ntt_zetas_start, "ntt_zetas is null");
7695 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7696 OptoRuntime::kyberNtt_Type(),
7697 stubAddr, stubName, TypePtr::BOTTOM,
7698 coeffs_start, ntt_zetas_start);
7699 // return an int
7700 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7701 set_result(retvalue);
7702 return true;
7703 }
7704
7705 //------------------------------inline_kyberInverseNtt
7706 bool LibraryCallKit::inline_kyberInverseNtt() {
7707 address stubAddr;
7708 const char *stubName;
7709 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7710 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7711
7712 stubAddr = StubRoutines::kyberInverseNtt();
7713 stubName = "kyberInverseNtt";
7714 if (!stubAddr) return false;
7715
7716 Node* coeffs = argument(0);
7717 Node* zetas = argument(1);
7718
7719 coeffs = must_be_not_null(coeffs, true);
7720 zetas = must_be_not_null(zetas, true);
7721
7722 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7723 assert(coeffs_start, "coeffs is null");
7724 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7725 assert(zetas_start, "inverseNtt_zetas is null");
7726 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7727 OptoRuntime::kyberInverseNtt_Type(),
7728 stubAddr, stubName, TypePtr::BOTTOM,
7729 coeffs_start, zetas_start);
7730
7731 // return an int
7732 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7733 set_result(retvalue);
7734 return true;
7735 }
7736
7737 //------------------------------inline_kyberNttMult
7738 bool LibraryCallKit::inline_kyberNttMult() {
7739 address stubAddr;
7740 const char *stubName;
7741 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7742 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7743
7744 stubAddr = StubRoutines::kyberNttMult();
7745 stubName = "kyberNttMult";
7746 if (!stubAddr) return false;
7747
7748 Node* result = argument(0);
7749 Node* ntta = argument(1);
7750 Node* nttb = argument(2);
7751 Node* zetas = argument(3);
7752
7753 result = must_be_not_null(result, true);
7754 ntta = must_be_not_null(ntta, true);
7755 nttb = must_be_not_null(nttb, true);
7756 zetas = must_be_not_null(zetas, true);
7757
7758 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7759 assert(result_start, "result is null");
7760 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7761 assert(ntta_start, "ntta is null");
7762 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7763 assert(nttb_start, "nttb is null");
7764 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7765 assert(zetas_start, "nttMult_zetas is null");
7766 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7767 OptoRuntime::kyberNttMult_Type(),
7768 stubAddr, stubName, TypePtr::BOTTOM,
7769 result_start, ntta_start, nttb_start,
7770 zetas_start);
7771
7772 // return an int
7773 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7774 set_result(retvalue);
7775
7776 return true;
7777 }
7778
7779 //------------------------------inline_kyberAddPoly_2
7780 bool LibraryCallKit::inline_kyberAddPoly_2() {
7781 address stubAddr;
7782 const char *stubName;
7783 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7784 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7785
7786 stubAddr = StubRoutines::kyberAddPoly_2();
7787 stubName = "kyberAddPoly_2";
7788 if (!stubAddr) return false;
7789
7790 Node* result = argument(0);
7791 Node* a = argument(1);
7792 Node* b = argument(2);
7793
7794 result = must_be_not_null(result, true);
7795 a = must_be_not_null(a, true);
7796 b = must_be_not_null(b, true);
7797
7798 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7799 assert(result_start, "result is null");
7800 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7801 assert(a_start, "a is null");
7802 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7803 assert(b_start, "b is null");
7804 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7805 OptoRuntime::kyberAddPoly_2_Type(),
7806 stubAddr, stubName, TypePtr::BOTTOM,
7807 result_start, a_start, b_start);
7808 // return an int
7809 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7810 set_result(retvalue);
7811 return true;
7812 }
7813
7814 //------------------------------inline_kyberAddPoly_3
7815 bool LibraryCallKit::inline_kyberAddPoly_3() {
7816 address stubAddr;
7817 const char *stubName;
7818 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7819 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7820
7821 stubAddr = StubRoutines::kyberAddPoly_3();
7822 stubName = "kyberAddPoly_3";
7823 if (!stubAddr) return false;
7824
7825 Node* result = argument(0);
7826 Node* a = argument(1);
7827 Node* b = argument(2);
7828 Node* c = argument(3);
7829
7830 result = must_be_not_null(result, true);
7831 a = must_be_not_null(a, true);
7832 b = must_be_not_null(b, true);
7833 c = must_be_not_null(c, true);
7834
7835 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7836 assert(result_start, "result is null");
7837 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7838 assert(a_start, "a is null");
7839 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7840 assert(b_start, "b is null");
7841 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7842 assert(c_start, "c is null");
7843 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7844 OptoRuntime::kyberAddPoly_3_Type(),
7845 stubAddr, stubName, TypePtr::BOTTOM,
7846 result_start, a_start, b_start, c_start);
7847 // return an int
7848 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7849 set_result(retvalue);
7850 return true;
7851 }
7852
7853 //------------------------------inline_kyber12To16
7854 bool LibraryCallKit::inline_kyber12To16() {
7855 address stubAddr;
7856 const char *stubName;
7857 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7858 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7859
7860 stubAddr = StubRoutines::kyber12To16();
7861 stubName = "kyber12To16";
7862 if (!stubAddr) return false;
7863
7864 Node* condensed = argument(0);
7865 Node* condensedOffs = argument(1);
7866 Node* parsed = argument(2);
7867 Node* parsedLength = argument(3);
7868
7869 condensed = must_be_not_null(condensed, true);
7870 parsed = must_be_not_null(parsed, true);
7871
7872 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7873 assert(condensed_start, "condensed is null");
7874 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7875 assert(parsed_start, "parsed is null");
7876 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7877 OptoRuntime::kyber12To16_Type(),
7878 stubAddr, stubName, TypePtr::BOTTOM,
7879 condensed_start, condensedOffs, parsed_start, parsedLength);
7880 // return an int
7881 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7882 set_result(retvalue);
7883 return true;
7884
7885 }
7886
7887 //------------------------------inline_kyberBarrettReduce
7888 bool LibraryCallKit::inline_kyberBarrettReduce() {
7889 address stubAddr;
7890 const char *stubName;
7891 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7892 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7893
7894 stubAddr = StubRoutines::kyberBarrettReduce();
7895 stubName = "kyberBarrettReduce";
7896 if (!stubAddr) return false;
7897
7898 Node* coeffs = argument(0);
7899
7900 coeffs = must_be_not_null(coeffs, true);
7901
7902 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7903 assert(coeffs_start, "coeffs is null");
7904 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7905 OptoRuntime::kyberBarrettReduce_Type(),
7906 stubAddr, stubName, TypePtr::BOTTOM,
7907 coeffs_start);
7908 // return an int
7909 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7910 set_result(retvalue);
7911 return true;
7912 }
7913
7914 //------------------------------inline_dilithiumAlmostNtt
7915 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7916 address stubAddr;
7917 const char *stubName;
7918 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7919 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7920
7921 stubAddr = StubRoutines::dilithiumAlmostNtt();
7922 stubName = "dilithiumAlmostNtt";
7923 if (!stubAddr) return false;
7924
7925 Node* coeffs = argument(0);
7926 Node* ntt_zetas = argument(1);
7927
7928 coeffs = must_be_not_null(coeffs, true);
7929 ntt_zetas = must_be_not_null(ntt_zetas, true);
7930
7931 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7932 assert(coeffs_start, "coeffs is null");
7933 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7934 assert(ntt_zetas_start, "ntt_zetas is null");
7935 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7936 OptoRuntime::dilithiumAlmostNtt_Type(),
7937 stubAddr, stubName, TypePtr::BOTTOM,
7938 coeffs_start, ntt_zetas_start);
7939 // return an int
7940 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7941 set_result(retvalue);
7942 return true;
7943 }
7944
7945 //------------------------------inline_dilithiumAlmostInverseNtt
7946 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7947 address stubAddr;
7948 const char *stubName;
7949 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7950 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7951
7952 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7953 stubName = "dilithiumAlmostInverseNtt";
7954 if (!stubAddr) return false;
7955
7956 Node* coeffs = argument(0);
7957 Node* zetas = argument(1);
7958
7959 coeffs = must_be_not_null(coeffs, true);
7960 zetas = must_be_not_null(zetas, true);
7961
7962 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7963 assert(coeffs_start, "coeffs is null");
7964 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
7965 assert(zetas_start, "inverseNtt_zetas is null");
7966 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7967 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
7968 stubAddr, stubName, TypePtr::BOTTOM,
7969 coeffs_start, zetas_start);
7970 // return an int
7971 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
7972 set_result(retvalue);
7973 return true;
7974 }
7975
7976 //------------------------------inline_dilithiumNttMult
7977 bool LibraryCallKit::inline_dilithiumNttMult() {
7978 address stubAddr;
7979 const char *stubName;
7980 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7981 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
7982
7983 stubAddr = StubRoutines::dilithiumNttMult();
7984 stubName = "dilithiumNttMult";
7985 if (!stubAddr) return false;
7986
7987 Node* result = argument(0);
7988 Node* ntta = argument(1);
7989 Node* nttb = argument(2);
7990 Node* zetas = argument(3);
7991
7992 result = must_be_not_null(result, true);
7993 ntta = must_be_not_null(ntta, true);
7994 nttb = must_be_not_null(nttb, true);
7995 zetas = must_be_not_null(zetas, true);
7996
7997 Node* result_start = array_element_address(result, intcon(0), T_INT);
7998 assert(result_start, "result is null");
7999 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8000 assert(ntta_start, "ntta is null");
8001 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8002 assert(nttb_start, "nttb is null");
8003 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8004 OptoRuntime::dilithiumNttMult_Type(),
8005 stubAddr, stubName, TypePtr::BOTTOM,
8006 result_start, ntta_start, nttb_start);
8007
8008 // return an int
8009 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8010 set_result(retvalue);
8011
8012 return true;
8013 }
8014
8015 //------------------------------inline_dilithiumMontMulByConstant
8016 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8017 address stubAddr;
8018 const char *stubName;
8019 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8020 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8021
8022 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8023 stubName = "dilithiumMontMulByConstant";
8024 if (!stubAddr) return false;
8025
8026 Node* coeffs = argument(0);
8027 Node* constant = argument(1);
8028
8029 coeffs = must_be_not_null(coeffs, true);
8030
8031 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8032 assert(coeffs_start, "coeffs is null");
8033 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8034 OptoRuntime::dilithiumMontMulByConstant_Type(),
8035 stubAddr, stubName, TypePtr::BOTTOM,
8036 coeffs_start, constant);
8037
8038 // return an int
8039 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8040 set_result(retvalue);
8041 return true;
8042 }
8043
8044
8045 //------------------------------inline_dilithiumDecomposePoly
8046 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8047 address stubAddr;
8048 const char *stubName;
8049 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8050 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8051
8052 stubAddr = StubRoutines::dilithiumDecomposePoly();
8053 stubName = "dilithiumDecomposePoly";
8054 if (!stubAddr) return false;
8055
8056 Node* input = argument(0);
8057 Node* lowPart = argument(1);
8058 Node* highPart = argument(2);
8059 Node* twoGamma2 = argument(3);
8060 Node* multiplier = argument(4);
8061
8062 input = must_be_not_null(input, true);
8063 lowPart = must_be_not_null(lowPart, true);
8064 highPart = must_be_not_null(highPart, true);
8065
8066 Node* input_start = array_element_address(input, intcon(0), T_INT);
8067 assert(input_start, "input is null");
8068 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8069 assert(lowPart_start, "lowPart is null");
8070 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8071 assert(highPart_start, "highPart is null");
8072
8073 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8074 OptoRuntime::dilithiumDecomposePoly_Type(),
8075 stubAddr, stubName, TypePtr::BOTTOM,
8076 input_start, lowPart_start, highPart_start,
8077 twoGamma2, multiplier);
8078
8079 // return an int
8080 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8081 set_result(retvalue);
8082 return true;
8083 }
8084
8085 bool LibraryCallKit::inline_base64_encodeBlock() {
8086 address stubAddr;
8087 const char *stubName;
8088 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8089 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8090 stubAddr = StubRoutines::base64_encodeBlock();
8091 stubName = "encodeBlock";
8092
8093 if (!stubAddr) return false;
8094 Node* base64obj = argument(0);
8095 Node* src = argument(1);
8096 Node* offset = argument(2);
8097 Node* len = argument(3);
8098 Node* dest = argument(4);
8099 Node* dp = argument(5);
8100 Node* isURL = argument(6);
8101
8102 src = must_be_not_null(src, true);
8103 dest = must_be_not_null(dest, true);
8104
8105 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8106 assert(src_start, "source array is null");
8107 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8108 assert(dest_start, "destination array is null");
8109
8110 Node* base64 = make_runtime_call(RC_LEAF,
8111 OptoRuntime::base64_encodeBlock_Type(),
8112 stubAddr, stubName, TypePtr::BOTTOM,
8113 src_start, offset, len, dest_start, dp, isURL);
8114 return true;
8115 }
8116
8117 bool LibraryCallKit::inline_base64_decodeBlock() {
8118 address stubAddr;
8119 const char *stubName;
8120 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8121 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8122 stubAddr = StubRoutines::base64_decodeBlock();
8123 stubName = "decodeBlock";
8124
8125 if (!stubAddr) return false;
8126 Node* base64obj = argument(0);
8127 Node* src = argument(1);
8128 Node* src_offset = argument(2);
8129 Node* len = argument(3);
8130 Node* dest = argument(4);
8131 Node* dest_offset = argument(5);
8132 Node* isURL = argument(6);
8133 Node* isMIME = argument(7);
8134
8135 src = must_be_not_null(src, true);
8136 dest = must_be_not_null(dest, true);
8137
8138 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8139 assert(src_start, "source array is null");
8140 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8141 assert(dest_start, "destination array is null");
8142
8143 Node* call = make_runtime_call(RC_LEAF,
8144 OptoRuntime::base64_decodeBlock_Type(),
8145 stubAddr, stubName, TypePtr::BOTTOM,
8146 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8147 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8148 set_result(result);
8149 return true;
8150 }
8151
8152 bool LibraryCallKit::inline_poly1305_processBlocks() {
8153 address stubAddr;
8154 const char *stubName;
8155 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8156 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8157 stubAddr = StubRoutines::poly1305_processBlocks();
8158 stubName = "poly1305_processBlocks";
8159
8160 if (!stubAddr) return false;
8161 null_check_receiver(); // null-check receiver
8162 if (stopped()) return true;
8163
8164 Node* input = argument(1);
8165 Node* input_offset = argument(2);
8166 Node* len = argument(3);
8167 Node* alimbs = argument(4);
8168 Node* rlimbs = argument(5);
8169
8170 input = must_be_not_null(input, true);
8171 alimbs = must_be_not_null(alimbs, true);
8172 rlimbs = must_be_not_null(rlimbs, true);
8173
8174 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8175 assert(input_start, "input array is null");
8176 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8177 assert(acc_start, "acc array is null");
8178 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8179 assert(r_start, "r array is null");
8180
8181 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8182 OptoRuntime::poly1305_processBlocks_Type(),
8183 stubAddr, stubName, TypePtr::BOTTOM,
8184 input_start, len, acc_start, r_start);
8185 return true;
8186 }
8187
8188 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8189 address stubAddr;
8190 const char *stubName;
8191 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8192 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8193 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8194 stubName = "intpoly_montgomeryMult_P256";
8195
8196 if (!stubAddr) return false;
8197 null_check_receiver(); // null-check receiver
8198 if (stopped()) return true;
8199
8200 Node* a = argument(1);
8201 Node* b = argument(2);
8202 Node* r = argument(3);
8203
8204 a = must_be_not_null(a, true);
8205 b = must_be_not_null(b, true);
8206 r = must_be_not_null(r, true);
8207
8208 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8209 assert(a_start, "a array is null");
8210 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8211 assert(b_start, "b array is null");
8212 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8213 assert(r_start, "r array is null");
8214
8215 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8216 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8217 stubAddr, stubName, TypePtr::BOTTOM,
8218 a_start, b_start, r_start);
8219 return true;
8220 }
8221
8222 bool LibraryCallKit::inline_intpoly_assign() {
8223 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8224 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8225 const char *stubName = "intpoly_assign";
8226 address stubAddr = StubRoutines::intpoly_assign();
8227 if (!stubAddr) return false;
8228
8229 Node* set = argument(0);
8230 Node* a = argument(1);
8231 Node* b = argument(2);
8232 Node* arr_length = load_array_length(a);
8233
8234 a = must_be_not_null(a, true);
8235 b = must_be_not_null(b, true);
8236
8237 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8238 assert(a_start, "a array is null");
8239 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8240 assert(b_start, "b array is null");
8241
8242 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8243 OptoRuntime::intpoly_assign_Type(),
8244 stubAddr, stubName, TypePtr::BOTTOM,
8245 set, a_start, b_start, arr_length);
8246 return true;
8247 }
8248
8249 //------------------------------inline_digestBase_implCompress-----------------------
8250 //
8251 // Calculate MD5 for single-block byte[] array.
8252 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8253 //
8254 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8255 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8256 //
8257 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8258 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8259 //
8260 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8261 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8262 //
8263 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8264 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8265 //
8266 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8267 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8268
8269 Node* digestBase_obj = argument(0);
8270 Node* src = argument(1); // type oop
8271 Node* ofs = argument(2); // type int
8272
8273 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8274 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8275 // failed array check
8276 return false;
8277 }
8278 // Figure out the size and type of the elements we will be copying.
8279 BasicType src_elem = src_type->elem()->array_element_basic_type();
8280 if (src_elem != T_BYTE) {
8281 return false;
8282 }
8283 // 'src_start' points to src array + offset
8284 src = must_be_not_null(src, true);
8285 Node* src_start = array_element_address(src, ofs, src_elem);
8286 Node* state = nullptr;
8287 Node* block_size = nullptr;
8288 address stubAddr;
8289 const char *stubName;
8290
8291 switch(id) {
8292 case vmIntrinsics::_md5_implCompress:
8293 assert(UseMD5Intrinsics, "need MD5 instruction support");
8294 state = get_state_from_digest_object(digestBase_obj, T_INT);
8295 stubAddr = StubRoutines::md5_implCompress();
8296 stubName = "md5_implCompress";
8297 break;
8298 case vmIntrinsics::_sha_implCompress:
8299 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8300 state = get_state_from_digest_object(digestBase_obj, T_INT);
8301 stubAddr = StubRoutines::sha1_implCompress();
8302 stubName = "sha1_implCompress";
8303 break;
8304 case vmIntrinsics::_sha2_implCompress:
8305 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8306 state = get_state_from_digest_object(digestBase_obj, T_INT);
8307 stubAddr = StubRoutines::sha256_implCompress();
8308 stubName = "sha256_implCompress";
8309 break;
8310 case vmIntrinsics::_sha5_implCompress:
8311 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8312 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8313 stubAddr = StubRoutines::sha512_implCompress();
8314 stubName = "sha512_implCompress";
8315 break;
8316 case vmIntrinsics::_sha3_implCompress:
8317 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8318 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8319 stubAddr = StubRoutines::sha3_implCompress();
8320 stubName = "sha3_implCompress";
8321 block_size = get_block_size_from_digest_object(digestBase_obj);
8322 if (block_size == nullptr) return false;
8323 break;
8324 default:
8325 fatal_unexpected_iid(id);
8326 return false;
8327 }
8328 if (state == nullptr) return false;
8329
8330 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8331 if (stubAddr == nullptr) return false;
8332
8333 // Call the stub.
8334 Node* call;
8335 if (block_size == nullptr) {
8336 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8337 stubAddr, stubName, TypePtr::BOTTOM,
8338 src_start, state);
8339 } else {
8340 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8341 stubAddr, stubName, TypePtr::BOTTOM,
8342 src_start, state, block_size);
8343 }
8344
8345 return true;
8346 }
8347
8348 //------------------------------inline_double_keccak
8349 bool LibraryCallKit::inline_double_keccak() {
8350 address stubAddr;
8351 const char *stubName;
8352 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8353 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8354
8355 stubAddr = StubRoutines::double_keccak();
8356 stubName = "double_keccak";
8357 if (!stubAddr) return false;
8358
8359 Node* status0 = argument(0);
8360 Node* status1 = argument(1);
8361
8362 status0 = must_be_not_null(status0, true);
8363 status1 = must_be_not_null(status1, true);
8364
8365 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8366 assert(status0_start, "status0 is null");
8367 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8368 assert(status1_start, "status1 is null");
8369 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8370 OptoRuntime::double_keccak_Type(),
8371 stubAddr, stubName, TypePtr::BOTTOM,
8372 status0_start, status1_start);
8373 // return an int
8374 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8375 set_result(retvalue);
8376 return true;
8377 }
8378
8379
8380 //------------------------------inline_digestBase_implCompressMB-----------------------
8381 //
8382 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8383 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8384 //
8385 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8386 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8387 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8388 assert((uint)predicate < 5, "sanity");
8389 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8390
8391 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8392 Node* src = argument(1); // byte[] array
8393 Node* ofs = argument(2); // type int
8394 Node* limit = argument(3); // type int
8395
8396 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8397 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8398 // failed array check
8399 return false;
8400 }
8401 // Figure out the size and type of the elements we will be copying.
8402 BasicType src_elem = src_type->elem()->array_element_basic_type();
8403 if (src_elem != T_BYTE) {
8404 return false;
8405 }
8406 // 'src_start' points to src array + offset
8407 src = must_be_not_null(src, false);
8408 Node* src_start = array_element_address(src, ofs, src_elem);
8409
8410 const char* klass_digestBase_name = nullptr;
8411 const char* stub_name = nullptr;
8412 address stub_addr = nullptr;
8413 BasicType elem_type = T_INT;
8414
8415 switch (predicate) {
8416 case 0:
8417 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8418 klass_digestBase_name = "sun/security/provider/MD5";
8419 stub_name = "md5_implCompressMB";
8420 stub_addr = StubRoutines::md5_implCompressMB();
8421 }
8422 break;
8423 case 1:
8424 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8425 klass_digestBase_name = "sun/security/provider/SHA";
8426 stub_name = "sha1_implCompressMB";
8427 stub_addr = StubRoutines::sha1_implCompressMB();
8428 }
8429 break;
8430 case 2:
8431 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8432 klass_digestBase_name = "sun/security/provider/SHA2";
8433 stub_name = "sha256_implCompressMB";
8434 stub_addr = StubRoutines::sha256_implCompressMB();
8435 }
8436 break;
8437 case 3:
8438 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8439 klass_digestBase_name = "sun/security/provider/SHA5";
8440 stub_name = "sha512_implCompressMB";
8441 stub_addr = StubRoutines::sha512_implCompressMB();
8442 elem_type = T_LONG;
8443 }
8444 break;
8445 case 4:
8446 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8447 klass_digestBase_name = "sun/security/provider/SHA3";
8448 stub_name = "sha3_implCompressMB";
8449 stub_addr = StubRoutines::sha3_implCompressMB();
8450 elem_type = T_LONG;
8451 }
8452 break;
8453 default:
8454 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8455 }
8456 if (klass_digestBase_name != nullptr) {
8457 assert(stub_addr != nullptr, "Stub is generated");
8458 if (stub_addr == nullptr) return false;
8459
8460 // get DigestBase klass to lookup for SHA klass
8461 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8462 assert(tinst != nullptr, "digestBase_obj is not instance???");
8463 assert(tinst->is_loaded(), "DigestBase is not loaded");
8464
8465 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8466 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8467 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8468 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8469 }
8470 return false;
8471 }
8472
8473 //------------------------------inline_digestBase_implCompressMB-----------------------
8474 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8475 BasicType elem_type, address stubAddr, const char *stubName,
8476 Node* src_start, Node* ofs, Node* limit) {
8477 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8478 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8479 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8480 digest_obj = _gvn.transform(digest_obj);
8481
8482 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8483 if (state == nullptr) return false;
8484
8485 Node* block_size = nullptr;
8486 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8487 block_size = get_block_size_from_digest_object(digest_obj);
8488 if (block_size == nullptr) return false;
8489 }
8490
8491 // Call the stub.
8492 Node* call;
8493 if (block_size == nullptr) {
8494 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8495 OptoRuntime::digestBase_implCompressMB_Type(false),
8496 stubAddr, stubName, TypePtr::BOTTOM,
8497 src_start, state, ofs, limit);
8498 } else {
8499 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8500 OptoRuntime::digestBase_implCompressMB_Type(true),
8501 stubAddr, stubName, TypePtr::BOTTOM,
8502 src_start, state, block_size, ofs, limit);
8503 }
8504
8505 // return ofs (int)
8506 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8507 set_result(result);
8508
8509 return true;
8510 }
8511
8512 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8513 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8514 assert(UseAES, "need AES instruction support");
8515 address stubAddr = nullptr;
8516 const char *stubName = nullptr;
8517 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8518 stubName = "galoisCounterMode_AESCrypt";
8519
8520 if (stubAddr == nullptr) return false;
8521
8522 Node* in = argument(0);
8523 Node* inOfs = argument(1);
8524 Node* len = argument(2);
8525 Node* ct = argument(3);
8526 Node* ctOfs = argument(4);
8527 Node* out = argument(5);
8528 Node* outOfs = argument(6);
8529 Node* gctr_object = argument(7);
8530 Node* ghash_object = argument(8);
8531
8532 // (1) in, ct and out are arrays.
8533 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8534 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8535 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8536 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8537 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8538 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8539
8540 // checks are the responsibility of the caller
8541 Node* in_start = in;
8542 Node* ct_start = ct;
8543 Node* out_start = out;
8544 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8545 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8546 in_start = array_element_address(in, inOfs, T_BYTE);
8547 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8548 out_start = array_element_address(out, outOfs, T_BYTE);
8549 }
8550
8551 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8552 // (because of the predicated logic executed earlier).
8553 // so we cast it here safely.
8554 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8555 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8556 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8557 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8558 Node* state = load_field_from_object(ghash_object, "state", "[J");
8559
8560 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8561 return false;
8562 }
8563 // cast it to what we know it will be at runtime
8564 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8565 assert(tinst != nullptr, "GCTR obj is null");
8566 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8567 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8568 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8569 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8570 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8571 const TypeOopPtr* xtype = aklass->as_instance_type();
8572 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8573 aescrypt_object = _gvn.transform(aescrypt_object);
8574 // we need to get the start of the aescrypt_object's expanded key array
8575 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8576 if (k_start == nullptr) return false;
8577 // similarly, get the start address of the r vector
8578 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8579 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8580 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8581
8582
8583 // Call the stub, passing params
8584 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8585 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8586 stubAddr, stubName, TypePtr::BOTTOM,
8587 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8588
8589 // return cipher length (int)
8590 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8591 set_result(retvalue);
8592
8593 return true;
8594 }
8595
8596 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8597 // Return node representing slow path of predicate check.
8598 // the pseudo code we want to emulate with this predicate is:
8599 // for encryption:
8600 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8601 // for decryption:
8602 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8603 // note cipher==plain is more conservative than the original java code but that's OK
8604 //
8605
8606 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8607 // The receiver was checked for null already.
8608 Node* objGCTR = argument(7);
8609 // Load embeddedCipher field of GCTR object.
8610 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8611 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8612
8613 // get AESCrypt klass for instanceOf check
8614 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8615 // will have same classloader as CipherBlockChaining object
8616 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8617 assert(tinst != nullptr, "GCTR obj is null");
8618 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8619
8620 // we want to do an instanceof comparison against the AESCrypt class
8621 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8622 if (!klass_AESCrypt->is_loaded()) {
8623 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8624 Node* ctrl = control();
8625 set_control(top()); // no regular fast path
8626 return ctrl;
8627 }
8628
8629 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8630 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8631 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8632 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8633 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8634
8635 return instof_false; // even if it is null
8636 }
8637
8638 //------------------------------get_state_from_digest_object-----------------------
8639 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8640 const char* state_type;
8641 switch (elem_type) {
8642 case T_BYTE: state_type = "[B"; break;
8643 case T_INT: state_type = "[I"; break;
8644 case T_LONG: state_type = "[J"; break;
8645 default: ShouldNotReachHere();
8646 }
8647 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8648 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8649 if (digest_state == nullptr) return (Node *) nullptr;
8650
8651 // now have the array, need to get the start address of the state array
8652 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8653 return state;
8654 }
8655
8656 //------------------------------get_block_size_from_sha3_object----------------------------------
8657 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8658 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8659 assert (block_size != nullptr, "sanity");
8660 return block_size;
8661 }
8662
8663 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8664 // Return node representing slow path of predicate check.
8665 // the pseudo code we want to emulate with this predicate is:
8666 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8667 //
8668 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8669 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8670 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8671 assert((uint)predicate < 5, "sanity");
8672
8673 // The receiver was checked for null already.
8674 Node* digestBaseObj = argument(0);
8675
8676 // get DigestBase klass for instanceOf check
8677 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8678 assert(tinst != nullptr, "digestBaseObj is null");
8679 assert(tinst->is_loaded(), "DigestBase is not loaded");
8680
8681 const char* klass_name = nullptr;
8682 switch (predicate) {
8683 case 0:
8684 if (UseMD5Intrinsics) {
8685 // we want to do an instanceof comparison against the MD5 class
8686 klass_name = "sun/security/provider/MD5";
8687 }
8688 break;
8689 case 1:
8690 if (UseSHA1Intrinsics) {
8691 // we want to do an instanceof comparison against the SHA class
8692 klass_name = "sun/security/provider/SHA";
8693 }
8694 break;
8695 case 2:
8696 if (UseSHA256Intrinsics) {
8697 // we want to do an instanceof comparison against the SHA2 class
8698 klass_name = "sun/security/provider/SHA2";
8699 }
8700 break;
8701 case 3:
8702 if (UseSHA512Intrinsics) {
8703 // we want to do an instanceof comparison against the SHA5 class
8704 klass_name = "sun/security/provider/SHA5";
8705 }
8706 break;
8707 case 4:
8708 if (UseSHA3Intrinsics) {
8709 // we want to do an instanceof comparison against the SHA3 class
8710 klass_name = "sun/security/provider/SHA3";
8711 }
8712 break;
8713 default:
8714 fatal("unknown SHA intrinsic predicate: %d", predicate);
8715 }
8716
8717 ciKlass* klass = nullptr;
8718 if (klass_name != nullptr) {
8719 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8720 }
8721 if ((klass == nullptr) || !klass->is_loaded()) {
8722 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8723 Node* ctrl = control();
8724 set_control(top()); // no intrinsic path
8725 return ctrl;
8726 }
8727 ciInstanceKlass* instklass = klass->as_instance_klass();
8728
8729 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8730 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8731 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8732 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8733
8734 return instof_false; // even if it is null
8735 }
8736
8737 //-------------inline_fma-----------------------------------
8738 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8739 Node *a = nullptr;
8740 Node *b = nullptr;
8741 Node *c = nullptr;
8742 Node* result = nullptr;
8743 switch (id) {
8744 case vmIntrinsics::_fmaD:
8745 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8746 // no receiver since it is static method
8747 a = argument(0);
8748 b = argument(2);
8749 c = argument(4);
8750 result = _gvn.transform(new FmaDNode(a, b, c));
8751 break;
8752 case vmIntrinsics::_fmaF:
8753 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8754 a = argument(0);
8755 b = argument(1);
8756 c = argument(2);
8757 result = _gvn.transform(new FmaFNode(a, b, c));
8758 break;
8759 default:
8760 fatal_unexpected_iid(id); break;
8761 }
8762 set_result(result);
8763 return true;
8764 }
8765
8766 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8767 // argument(0) is receiver
8768 Node* codePoint = argument(1);
8769 Node* n = nullptr;
8770
8771 switch (id) {
8772 case vmIntrinsics::_isDigit :
8773 n = new DigitNode(control(), codePoint);
8774 break;
8775 case vmIntrinsics::_isLowerCase :
8776 n = new LowerCaseNode(control(), codePoint);
8777 break;
8778 case vmIntrinsics::_isUpperCase :
8779 n = new UpperCaseNode(control(), codePoint);
8780 break;
8781 case vmIntrinsics::_isWhitespace :
8782 n = new WhitespaceNode(control(), codePoint);
8783 break;
8784 default:
8785 fatal_unexpected_iid(id);
8786 }
8787
8788 set_result(_gvn.transform(n));
8789 return true;
8790 }
8791
8792 bool LibraryCallKit::inline_profileBoolean() {
8793 Node* counts = argument(1);
8794 const TypeAryPtr* ary = nullptr;
8795 ciArray* aobj = nullptr;
8796 if (counts->is_Con()
8797 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8798 && (aobj = ary->const_oop()->as_array()) != nullptr
8799 && (aobj->length() == 2)) {
8800 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8801 jint false_cnt = aobj->element_value(0).as_int();
8802 jint true_cnt = aobj->element_value(1).as_int();
8803
8804 if (C->log() != nullptr) {
8805 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8806 false_cnt, true_cnt);
8807 }
8808
8809 if (false_cnt + true_cnt == 0) {
8810 // According to profile, never executed.
8811 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8812 Deoptimization::Action_reinterpret);
8813 return true;
8814 }
8815
8816 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8817 // is a number of each value occurrences.
8818 Node* result = argument(0);
8819 if (false_cnt == 0 || true_cnt == 0) {
8820 // According to profile, one value has been never seen.
8821 int expected_val = (false_cnt == 0) ? 1 : 0;
8822
8823 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8824 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8825
8826 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8827 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8828 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8829
8830 { // Slow path: uncommon trap for never seen value and then reexecute
8831 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8832 // the value has been seen at least once.
8833 PreserveJVMState pjvms(this);
8834 PreserveReexecuteState preexecs(this);
8835 jvms()->set_should_reexecute(true);
8836
8837 set_control(slow_path);
8838 set_i_o(i_o());
8839
8840 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8841 Deoptimization::Action_reinterpret);
8842 }
8843 // The guard for never seen value enables sharpening of the result and
8844 // returning a constant. It allows to eliminate branches on the same value
8845 // later on.
8846 set_control(fast_path);
8847 result = intcon(expected_val);
8848 }
8849 // Stop profiling.
8850 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8851 // By replacing method body with profile data (represented as ProfileBooleanNode
8852 // on IR level) we effectively disable profiling.
8853 // It enables full speed execution once optimized code is generated.
8854 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8855 C->record_for_igvn(profile);
8856 set_result(profile);
8857 return true;
8858 } else {
8859 // Continue profiling.
8860 // Profile data isn't available at the moment. So, execute method's bytecode version.
8861 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8862 // is compiled and counters aren't available since corresponding MethodHandle
8863 // isn't a compile-time constant.
8864 return false;
8865 }
8866 }
8867
8868 bool LibraryCallKit::inline_isCompileConstant() {
8869 Node* n = argument(0);
8870 set_result(n->is_Con() ? intcon(1) : intcon(0));
8871 return true;
8872 }
8873
8874 //------------------------------- inline_getObjectSize --------------------------------------
8875 //
8876 // Calculate the runtime size of the object/array.
8877 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8878 //
8879 bool LibraryCallKit::inline_getObjectSize() {
8880 Node* obj = argument(3);
8881 Node* klass_node = load_object_klass(obj);
8882
8883 jint layout_con = Klass::_lh_neutral_value;
8884 Node* layout_val = get_layout_helper(klass_node, layout_con);
8885 int layout_is_con = (layout_val == nullptr);
8886
8887 if (layout_is_con) {
8888 // Layout helper is constant, can figure out things at compile time.
8889
8890 if (Klass::layout_helper_is_instance(layout_con)) {
8891 // Instance case: layout_con contains the size itself.
8892 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8893 set_result(size);
8894 } else {
8895 // Array case: size is round(header + element_size*arraylength).
8896 // Since arraylength is different for every array instance, we have to
8897 // compute the whole thing at runtime.
8898
8899 Node* arr_length = load_array_length(obj);
8900
8901 int round_mask = MinObjAlignmentInBytes - 1;
8902 int hsize = Klass::layout_helper_header_size(layout_con);
8903 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8904
8905 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8906 round_mask = 0; // strength-reduce it if it goes away completely
8907 }
8908 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8909 Node* header_size = intcon(hsize + round_mask);
8910
8911 Node* lengthx = ConvI2X(arr_length);
8912 Node* headerx = ConvI2X(header_size);
8913
8914 Node* abody = lengthx;
8915 if (eshift != 0) {
8916 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8917 }
8918 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8919 if (round_mask != 0) {
8920 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8921 }
8922 size = ConvX2L(size);
8923 set_result(size);
8924 }
8925 } else {
8926 // Layout helper is not constant, need to test for array-ness at runtime.
8927
8928 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8929 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8930 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8931 record_for_igvn(result_reg);
8932
8933 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8934 if (array_ctl != nullptr) {
8935 // Array case: size is round(header + element_size*arraylength).
8936 // Since arraylength is different for every array instance, we have to
8937 // compute the whole thing at runtime.
8938
8939 PreserveJVMState pjvms(this);
8940 set_control(array_ctl);
8941 Node* arr_length = load_array_length(obj);
8942
8943 int round_mask = MinObjAlignmentInBytes - 1;
8944 Node* mask = intcon(round_mask);
8945
8946 Node* hss = intcon(Klass::_lh_header_size_shift);
8947 Node* hsm = intcon(Klass::_lh_header_size_mask);
8948 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8949 header_size = _gvn.transform(new AndINode(header_size, hsm));
8950 header_size = _gvn.transform(new AddINode(header_size, mask));
8951
8952 // There is no need to mask or shift this value.
8953 // The semantics of LShiftINode include an implicit mask to 0x1F.
8954 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8955 Node* elem_shift = layout_val;
8956
8957 Node* lengthx = ConvI2X(arr_length);
8958 Node* headerx = ConvI2X(header_size);
8959
8960 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8961 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8962 if (round_mask != 0) {
8963 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8964 }
8965 size = ConvX2L(size);
8966
8967 result_reg->init_req(_array_path, control());
8968 result_val->init_req(_array_path, size);
8969 }
8970
8971 if (!stopped()) {
8972 // Instance case: the layout helper gives us instance size almost directly,
8973 // but we need to mask out the _lh_instance_slow_path_bit.
8974 Node* size = ConvI2X(layout_val);
8975 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8976 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8977 size = _gvn.transform(new AndXNode(size, mask));
8978 size = ConvX2L(size);
8979
8980 result_reg->init_req(_instance_path, control());
8981 result_val->init_req(_instance_path, size);
8982 }
8983
8984 set_result(result_reg, result_val);
8985 }
8986
8987 return true;
8988 }
8989
8990 //------------------------------- inline_blackhole --------------------------------------
8991 //
8992 // Make sure all arguments to this node are alive.
8993 // This matches methods that were requested to be blackholed through compile commands.
8994 //
8995 bool LibraryCallKit::inline_blackhole() {
8996 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8997 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8998 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8999
9000 // Blackhole node pinches only the control, not memory. This allows
9001 // the blackhole to be pinned in the loop that computes blackholed
9002 // values, but have no other side effects, like breaking the optimizations
9003 // across the blackhole.
9004
9005 Node* bh = _gvn.transform(new BlackholeNode(control()));
9006 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9007
9008 // Bind call arguments as blackhole arguments to keep them alive
9009 uint nargs = callee()->arg_size();
9010 for (uint i = 0; i < nargs; i++) {
9011 bh->add_req(argument(i));
9012 }
9013
9014 return true;
9015 }
9016
9017 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9018 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9019 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9020 return nullptr; // box klass is not Float16
9021 }
9022
9023 // Null check; get notnull casted pointer
9024 Node* null_ctl = top();
9025 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9026 // If not_null_box is dead, only null-path is taken
9027 if (stopped()) {
9028 set_control(null_ctl);
9029 return nullptr;
9030 }
9031 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9032 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9033 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9034 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9035 }
9036
9037 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9038 PreserveReexecuteState preexecs(this);
9039 jvms()->set_should_reexecute(true);
9040
9041 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9042 Node* klass_node = makecon(klass_type);
9043 Node* box = new_instance(klass_node);
9044
9045 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9046 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9047
9048 Node* field_store = _gvn.transform(access_store_at(box,
9049 value_field,
9050 value_adr_type,
9051 value,
9052 TypeInt::SHORT,
9053 T_SHORT,
9054 IN_HEAP));
9055 set_memory(field_store, value_adr_type);
9056 return box;
9057 }
9058
9059 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9060 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9061 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9062 return false;
9063 }
9064
9065 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9066 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9067 return false;
9068 }
9069
9070 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9071 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9072 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9073 ciSymbols::short_signature(),
9074 false);
9075 assert(field != nullptr, "");
9076
9077 // Transformed nodes
9078 Node* fld1 = nullptr;
9079 Node* fld2 = nullptr;
9080 Node* fld3 = nullptr;
9081 switch(num_args) {
9082 case 3:
9083 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9084 if (fld3 == nullptr) {
9085 return false;
9086 }
9087 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9088 // fall-through
9089 case 2:
9090 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9091 if (fld2 == nullptr) {
9092 return false;
9093 }
9094 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9095 // fall-through
9096 case 1:
9097 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9098 if (fld1 == nullptr) {
9099 return false;
9100 }
9101 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9102 break;
9103 default: fatal("Unsupported number of arguments %d", num_args);
9104 }
9105
9106 Node* result = nullptr;
9107 switch (id) {
9108 // Unary operations
9109 case vmIntrinsics::_sqrt_float16:
9110 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9111 break;
9112 // Ternary operations
9113 case vmIntrinsics::_fma_float16:
9114 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9115 break;
9116 default:
9117 fatal_unexpected_iid(id);
9118 break;
9119 }
9120 result = _gvn.transform(new ReinterpretHF2SNode(result));
9121 set_result(box_fp16_value(float16_box_type, field, result));
9122 return true;
9123 }
9124