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/mountUnmountDisabler.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "utilities/macros.hpp"
63 #include "utilities/powerOfTwo.hpp"
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return nullptr;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
85 if (is_available && is_virtual) {
86 is_available = vmIntrinsics::does_virtual_dispatch(id);
87 }
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return nullptr;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 return kit.transfer_exceptions_into_jvms();
135 }
136
137 // The intrinsic bailed out
138 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
139 assert(jvms->map() == kit.map(), "Out of sync JVM state");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dsinh:
250 case vmIntrinsics::_dtanh:
251 case vmIntrinsics::_dcbrt:
252 case vmIntrinsics::_dabs:
253 case vmIntrinsics::_fabs:
254 case vmIntrinsics::_iabs:
255 case vmIntrinsics::_labs:
256 case vmIntrinsics::_datan2:
257 case vmIntrinsics::_dsqrt:
258 case vmIntrinsics::_dsqrt_strict:
259 case vmIntrinsics::_dexp:
260 case vmIntrinsics::_dlog:
261 case vmIntrinsics::_dlog10:
262 case vmIntrinsics::_dpow:
263 case vmIntrinsics::_dcopySign:
264 case vmIntrinsics::_fcopySign:
265 case vmIntrinsics::_dsignum:
266 case vmIntrinsics::_roundF:
267 case vmIntrinsics::_roundD:
268 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
269
270 case vmIntrinsics::_notify:
271 case vmIntrinsics::_notifyAll:
272 return inline_notify(intrinsic_id());
273
274 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
275 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
276 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
277 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
278 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
279 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
280 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
281 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
282 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
283 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
284 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
285 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
286 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
287 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
288
289 case vmIntrinsics::_arraycopy: return inline_arraycopy();
290
291 case vmIntrinsics::_arraySort: return inline_array_sort();
292 case vmIntrinsics::_arrayPartition: return inline_array_partition();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309
310 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
311
312 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
313 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
314 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
315 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
316
317 case vmIntrinsics::_compressStringC:
318 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
319 case vmIntrinsics::_inflateStringC:
320 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
321
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
331
332 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
341
342 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
343 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
344 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
345 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
346 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
347 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
348 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
349 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
350 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
351
352 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
373 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
374 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
375 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
376 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
377 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
378 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
379 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
380 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
381
382 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
383 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
384 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
385 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
386 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
387 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
388 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
389 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
390 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
391
392 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
393 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
394 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
395 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
396 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
397 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
398 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
399 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
400 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
401
402 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
417
418 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
438
439 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
454
455 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
459
460 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
465
466 case vmIntrinsics::_loadFence:
467 case vmIntrinsics::_storeFence:
468 case vmIntrinsics::_storeStoreFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
474 case vmIntrinsics::_currentThread: return inline_native_currentThread();
475 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
476
477 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
478 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
479
480 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
481 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
482
483 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
484 "endFirstTransition", true);
485 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
486 "startFinalTransition", true);
487 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
488 "startTransition", false);
489 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
490 "endTransition", false);
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
508 case vmIntrinsics::_getLength: return inline_native_getLength();
509 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
510 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
511 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
512 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
513 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
514 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
515 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
516
517 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
518 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
519
520 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
521
522 case vmIntrinsics::_isInstance:
523 case vmIntrinsics::_isHidden:
524 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
535 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
536 case vmIntrinsics::_floatIsFinite:
537 case vmIntrinsics::_floatIsInfinite:
538 case vmIntrinsics::_doubleIsFinite:
539 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
540
541 case vmIntrinsics::_numberOfLeadingZeros_i:
542 case vmIntrinsics::_numberOfLeadingZeros_l:
543 case vmIntrinsics::_numberOfTrailingZeros_i:
544 case vmIntrinsics::_numberOfTrailingZeros_l:
545 case vmIntrinsics::_bitCount_i:
546 case vmIntrinsics::_bitCount_l:
547 case vmIntrinsics::_reverse_i:
548 case vmIntrinsics::_reverse_l:
549 case vmIntrinsics::_reverseBytes_i:
550 case vmIntrinsics::_reverseBytes_l:
551 case vmIntrinsics::_reverseBytes_s:
552 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
553
554 case vmIntrinsics::_compress_i:
555 case vmIntrinsics::_compress_l:
556 case vmIntrinsics::_expand_i:
557 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
558
559 case vmIntrinsics::_compareUnsigned_i:
560 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
561
562 case vmIntrinsics::_divideUnsigned_i:
563 case vmIntrinsics::_divideUnsigned_l:
564 case vmIntrinsics::_remainderUnsigned_i:
565 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
566
567 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
568
569 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
570 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
571 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
572 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
573 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
574
575 case vmIntrinsics::_Class_cast: return inline_Class_cast();
576
577 case vmIntrinsics::_aescrypt_encryptBlock:
578 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
579
580 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
581 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
582 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
583
584 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
585 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
586 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_counterMode_AESCrypt:
589 return inline_counterMode_AESCrypt(intrinsic_id());
590
591 case vmIntrinsics::_galoisCounterMode_AESCrypt:
592 return inline_galoisCounterMode_AESCrypt();
593
594 case vmIntrinsics::_md5_implCompress:
595 case vmIntrinsics::_sha_implCompress:
596 case vmIntrinsics::_sha2_implCompress:
597 case vmIntrinsics::_sha5_implCompress:
598 case vmIntrinsics::_sha3_implCompress:
599 return inline_digestBase_implCompress(intrinsic_id());
600 case vmIntrinsics::_double_keccak:
601 return inline_double_keccak();
602
603 case vmIntrinsics::_digestBase_implCompressMB:
604 return inline_digestBase_implCompressMB(predicate);
605
606 case vmIntrinsics::_multiplyToLen:
607 return inline_multiplyToLen();
608
609 case vmIntrinsics::_squareToLen:
610 return inline_squareToLen();
611
612 case vmIntrinsics::_mulAdd:
613 return inline_mulAdd();
614
615 case vmIntrinsics::_montgomeryMultiply:
616 return inline_montgomeryMultiply();
617 case vmIntrinsics::_montgomerySquare:
618 return inline_montgomerySquare();
619
620 case vmIntrinsics::_bigIntegerRightShiftWorker:
621 return inline_bigIntegerShift(true);
622 case vmIntrinsics::_bigIntegerLeftShiftWorker:
623 return inline_bigIntegerShift(false);
624
625 case vmIntrinsics::_vectorizedMismatch:
626 return inline_vectorizedMismatch();
627
628 case vmIntrinsics::_ghash_processBlocks:
629 return inline_ghash_processBlocks();
630 case vmIntrinsics::_chacha20Block:
631 return inline_chacha20Block();
632 case vmIntrinsics::_kyberNtt:
633 return inline_kyberNtt();
634 case vmIntrinsics::_kyberInverseNtt:
635 return inline_kyberInverseNtt();
636 case vmIntrinsics::_kyberNttMult:
637 return inline_kyberNttMult();
638 case vmIntrinsics::_kyberAddPoly_2:
639 return inline_kyberAddPoly_2();
640 case vmIntrinsics::_kyberAddPoly_3:
641 return inline_kyberAddPoly_3();
642 case vmIntrinsics::_kyber12To16:
643 return inline_kyber12To16();
644 case vmIntrinsics::_kyberBarrettReduce:
645 return inline_kyberBarrettReduce();
646 case vmIntrinsics::_dilithiumAlmostNtt:
647 return inline_dilithiumAlmostNtt();
648 case vmIntrinsics::_dilithiumAlmostInverseNtt:
649 return inline_dilithiumAlmostInverseNtt();
650 case vmIntrinsics::_dilithiumNttMult:
651 return inline_dilithiumNttMult();
652 case vmIntrinsics::_dilithiumMontMulByConstant:
653 return inline_dilithiumMontMulByConstant();
654 case vmIntrinsics::_dilithiumDecomposePoly:
655 return inline_dilithiumDecomposePoly();
656 case vmIntrinsics::_base64_encodeBlock:
657 return inline_base64_encodeBlock();
658 case vmIntrinsics::_base64_decodeBlock:
659 return inline_base64_decodeBlock();
660 case vmIntrinsics::_poly1305_processBlocks:
661 return inline_poly1305_processBlocks();
662 case vmIntrinsics::_intpoly_montgomeryMult_P256:
663 return inline_intpoly_montgomeryMult_P256();
664 case vmIntrinsics::_intpoly_assign:
665 return inline_intpoly_assign();
666 case vmIntrinsics::_encodeISOArray:
667 case vmIntrinsics::_encodeByteISOArray:
668 return inline_encodeISOArray(false);
669 case vmIntrinsics::_encodeAsciiArray:
670 return inline_encodeISOArray(true);
671
672 case vmIntrinsics::_updateCRC32:
673 return inline_updateCRC32();
674 case vmIntrinsics::_updateBytesCRC32:
675 return inline_updateBytesCRC32();
676 case vmIntrinsics::_updateByteBufferCRC32:
677 return inline_updateByteBufferCRC32();
678
679 case vmIntrinsics::_updateBytesCRC32C:
680 return inline_updateBytesCRC32C();
681 case vmIntrinsics::_updateDirectByteBufferCRC32C:
682 return inline_updateDirectByteBufferCRC32C();
683
684 case vmIntrinsics::_updateBytesAdler32:
685 return inline_updateBytesAdler32();
686 case vmIntrinsics::_updateByteBufferAdler32:
687 return inline_updateByteBufferAdler32();
688
689 case vmIntrinsics::_profileBoolean:
690 return inline_profileBoolean();
691 case vmIntrinsics::_isCompileConstant:
692 return inline_isCompileConstant();
693
694 case vmIntrinsics::_countPositives:
695 return inline_countPositives();
696
697 case vmIntrinsics::_fmaD:
698 case vmIntrinsics::_fmaF:
699 return inline_fma(intrinsic_id());
700
701 case vmIntrinsics::_isDigit:
702 case vmIntrinsics::_isLowerCase:
703 case vmIntrinsics::_isUpperCase:
704 case vmIntrinsics::_isWhitespace:
705 return inline_character_compare(intrinsic_id());
706
707 case vmIntrinsics::_min:
708 case vmIntrinsics::_max:
709 case vmIntrinsics::_min_strict:
710 case vmIntrinsics::_max_strict:
711 case vmIntrinsics::_minL:
712 case vmIntrinsics::_maxL:
713 case vmIntrinsics::_minF:
714 case vmIntrinsics::_maxF:
715 case vmIntrinsics::_minD:
716 case vmIntrinsics::_maxD:
717 case vmIntrinsics::_minF_strict:
718 case vmIntrinsics::_maxF_strict:
719 case vmIntrinsics::_minD_strict:
720 case vmIntrinsics::_maxD_strict:
721 return inline_min_max(intrinsic_id());
722
723 case vmIntrinsics::_VectorUnaryOp:
724 return inline_vector_nary_operation(1);
725 case vmIntrinsics::_VectorBinaryOp:
726 return inline_vector_nary_operation(2);
727 case vmIntrinsics::_VectorUnaryLibOp:
728 return inline_vector_call(1);
729 case vmIntrinsics::_VectorBinaryLibOp:
730 return inline_vector_call(2);
731 case vmIntrinsics::_VectorTernaryOp:
732 return inline_vector_nary_operation(3);
733 case vmIntrinsics::_VectorFromBitsCoerced:
734 return inline_vector_frombits_coerced();
735 case vmIntrinsics::_VectorMaskOp:
736 return inline_vector_mask_operation();
737 case vmIntrinsics::_VectorLoadOp:
738 return inline_vector_mem_operation(/*is_store=*/false);
739 case vmIntrinsics::_VectorLoadMaskedOp:
740 return inline_vector_mem_masked_operation(/*is_store*/false);
741 case vmIntrinsics::_VectorStoreOp:
742 return inline_vector_mem_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorStoreMaskedOp:
744 return inline_vector_mem_masked_operation(/*is_store=*/true);
745 case vmIntrinsics::_VectorGatherOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ false);
747 case vmIntrinsics::_VectorScatterOp:
748 return inline_vector_gather_scatter(/*is_scatter*/ true);
749 case vmIntrinsics::_VectorReductionCoerced:
750 return inline_vector_reduction();
751 case vmIntrinsics::_VectorTest:
752 return inline_vector_test();
753 case vmIntrinsics::_VectorBlend:
754 return inline_vector_blend();
755 case vmIntrinsics::_VectorRearrange:
756 return inline_vector_rearrange();
757 case vmIntrinsics::_VectorSelectFrom:
758 return inline_vector_select_from();
759 case vmIntrinsics::_VectorCompare:
760 return inline_vector_compare();
761 case vmIntrinsics::_VectorBroadcastInt:
762 return inline_vector_broadcast_int();
763 case vmIntrinsics::_VectorConvert:
764 return inline_vector_convert();
765 case vmIntrinsics::_VectorInsert:
766 return inline_vector_insert();
767 case vmIntrinsics::_VectorExtract:
768 return inline_vector_extract();
769 case vmIntrinsics::_VectorCompressExpand:
770 return inline_vector_compress_expand();
771 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
772 return inline_vector_select_from_two_vectors();
773 case vmIntrinsics::_IndexVector:
774 return inline_index_vector();
775 case vmIntrinsics::_IndexPartiallyInUpperRange:
776 return inline_index_partially_in_upper_range();
777
778 case vmIntrinsics::_getObjectSize:
779 return inline_getObjectSize();
780
781 case vmIntrinsics::_blackhole:
782 return inline_blackhole();
783
784 default:
785 // If you get here, it may be that someone has added a new intrinsic
786 // to the list in vmIntrinsics.hpp without implementing it here.
787 #ifndef PRODUCT
788 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
789 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
790 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
791 }
792 #endif
793 return false;
794 }
795 }
796
797 Node* LibraryCallKit::try_to_predicate(int predicate) {
798 if (!jvms()->has_method()) {
799 // Root JVMState has a null method.
800 assert(map()->memory()->Opcode() == Op_Parm, "");
801 // Insert the memory aliasing node
802 set_all_memory(reset_memory());
803 }
804 assert(merged_memory(), "");
805
806 switch (intrinsic_id()) {
807 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(false);
809 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt_predicate(true);
811 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(false);
813 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
814 return inline_electronicCodeBook_AESCrypt_predicate(true);
815 case vmIntrinsics::_counterMode_AESCrypt:
816 return inline_counterMode_AESCrypt_predicate();
817 case vmIntrinsics::_digestBase_implCompressMB:
818 return inline_digestBase_implCompressMB_predicate(predicate);
819 case vmIntrinsics::_galoisCounterMode_AESCrypt:
820 return inline_galoisCounterMode_AESCrypt_predicate();
821
822 default:
823 // If you get here, it may be that someone has added a new intrinsic
824 // to the list in vmIntrinsics.hpp without implementing it here.
825 #ifndef PRODUCT
826 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
827 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
828 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
829 }
830 #endif
831 Node* slow_ctl = control();
832 set_control(top()); // No fast path intrinsic
833 return slow_ctl;
834 }
835 }
836
837 //------------------------------set_result-------------------------------
838 // Helper function for finishing intrinsics.
839 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
840 record_for_igvn(region);
841 set_control(_gvn.transform(region));
842 set_result( _gvn.transform(value));
843 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
844 }
845
846 //------------------------------generate_guard---------------------------
847 // Helper function for generating guarded fast-slow graph structures.
848 // The given 'test', if true, guards a slow path. If the test fails
849 // then a fast path can be taken. (We generally hope it fails.)
850 // In all cases, GraphKit::control() is updated to the fast path.
851 // The returned value represents the control for the slow path.
852 // The return value is never 'top'; it is either a valid control
853 // or null if it is obvious that the slow path can never be taken.
854 // Also, if region and the slow control are not null, the slow edge
855 // is appended to the region.
856 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
857 if (stopped()) {
858 // Already short circuited.
859 return nullptr;
860 }
861
862 // Build an if node and its projections.
863 // If test is true we take the slow path, which we assume is uncommon.
864 if (_gvn.type(test) == TypeInt::ZERO) {
865 // The slow branch is never taken. No need to build this guard.
866 return nullptr;
867 }
868
869 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
870
871 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
872 if (if_slow == top()) {
873 // The slow branch is never taken. No need to build this guard.
874 return nullptr;
875 }
876
877 if (region != nullptr)
878 region->add_req(if_slow);
879
880 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
881 set_control(if_fast);
882
883 return if_slow;
884 }
885
886 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
887 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
888 }
889 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
890 return generate_guard(test, region, PROB_FAIR);
891 }
892
893 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
894 Node* *pos_index) {
895 if (stopped())
896 return nullptr; // already stopped
897 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
898 return nullptr; // index is already adequately typed
899 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
900 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
901 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
902 if (is_neg != nullptr && pos_index != nullptr) {
903 // Emulate effect of Parse::adjust_map_after_if.
904 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
905 (*pos_index) = _gvn.transform(ccast);
906 }
907 return is_neg;
908 }
909
910 // Make sure that 'position' is a valid limit index, in [0..length].
911 // There are two equivalent plans for checking this:
912 // A. (offset + copyLength) unsigned<= arrayLength
913 // B. offset <= (arrayLength - copyLength)
914 // We require that all of the values above, except for the sum and
915 // difference, are already known to be non-negative.
916 // Plan A is robust in the face of overflow, if offset and copyLength
917 // are both hugely positive.
918 //
919 // Plan B is less direct and intuitive, but it does not overflow at
920 // all, since the difference of two non-negatives is always
921 // representable. Whenever Java methods must perform the equivalent
922 // check they generally use Plan B instead of Plan A.
923 // For the moment we use Plan A.
924 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
925 Node* subseq_length,
926 Node* array_length,
927 RegionNode* region) {
928 if (stopped())
929 return nullptr; // already stopped
930 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
931 if (zero_offset && subseq_length->eqv_uncast(array_length))
932 return nullptr; // common case of whole-array copy
933 Node* last = subseq_length;
934 if (!zero_offset) // last += offset
935 last = _gvn.transform(new AddINode(last, offset));
936 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
937 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
938 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
939 return is_over;
940 }
941
942 // Emit range checks for the given String.value byte array
943 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
944 if (stopped()) {
945 return; // already stopped
946 }
947 RegionNode* bailout = new RegionNode(1);
948 record_for_igvn(bailout);
949 if (char_count) {
950 // Convert char count to byte count
951 count = _gvn.transform(new LShiftINode(count, intcon(1)));
952 }
953
954 // Offset and count must not be negative
955 generate_negative_guard(offset, bailout);
956 generate_negative_guard(count, bailout);
957 // Offset + count must not exceed length of array
958 generate_limit_guard(offset, count, load_array_length(array), bailout);
959
960 if (bailout->req() > 1) {
961 PreserveJVMState pjvms(this);
962 set_control(_gvn.transform(bailout));
963 uncommon_trap(Deoptimization::Reason_intrinsic,
964 Deoptimization::Action_maybe_recompile);
965 }
966 }
967
968 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
969 bool is_immutable) {
970 ciKlass* thread_klass = env()->Thread_klass();
971 const Type* thread_type
972 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
973
974 Node* thread = _gvn.transform(new ThreadLocalNode());
975 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
976 tls_output = thread;
977
978 Node* thread_obj_handle
979 = (is_immutable
980 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
981 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
982 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
983 thread_obj_handle = _gvn.transform(thread_obj_handle);
984
985 DecoratorSet decorators = IN_NATIVE;
986 if (is_immutable) {
987 decorators |= C2_IMMUTABLE_MEMORY;
988 }
989 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
990 }
991
992 //--------------------------generate_current_thread--------------------
993 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
994 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
995 /*is_immutable*/false);
996 }
997
998 //--------------------------generate_virtual_thread--------------------
999 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1000 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1001 !C->method()->changes_current_thread());
1002 }
1003
1004 //------------------------------make_string_method_node------------------------
1005 // Helper method for String intrinsic functions. This version is called with
1006 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1007 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1008 // containing the lengths of str1 and str2.
1009 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1010 Node* result = nullptr;
1011 switch (opcode) {
1012 case Op_StrIndexOf:
1013 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1014 str1_start, cnt1, str2_start, cnt2, ae);
1015 break;
1016 case Op_StrComp:
1017 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1018 str1_start, cnt1, str2_start, cnt2, ae);
1019 break;
1020 case Op_StrEquals:
1021 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1022 // Use the constant length if there is one because optimized match rule may exist.
1023 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1024 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1025 break;
1026 default:
1027 ShouldNotReachHere();
1028 return nullptr;
1029 }
1030
1031 // All these intrinsics have checks.
1032 C->set_has_split_ifs(true); // Has chance for split-if optimization
1033 clear_upper_avx();
1034
1035 return _gvn.transform(result);
1036 }
1037
1038 //------------------------------inline_string_compareTo------------------------
1039 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1040 Node* arg1 = argument(0);
1041 Node* arg2 = argument(1);
1042
1043 arg1 = must_be_not_null(arg1, true);
1044 arg2 = must_be_not_null(arg2, true);
1045
1046 // Get start addr and length of first argument
1047 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1048 Node* arg1_cnt = load_array_length(arg1);
1049
1050 // Get start addr and length of second argument
1051 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1052 Node* arg2_cnt = load_array_length(arg2);
1053
1054 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1055 set_result(result);
1056 return true;
1057 }
1058
1059 //------------------------------inline_string_equals------------------------
1060 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1061 Node* arg1 = argument(0);
1062 Node* arg2 = argument(1);
1063
1064 // paths (plus control) merge
1065 RegionNode* region = new RegionNode(3);
1066 Node* phi = new PhiNode(region, TypeInt::BOOL);
1067
1068 if (!stopped()) {
1069
1070 arg1 = must_be_not_null(arg1, true);
1071 arg2 = must_be_not_null(arg2, true);
1072
1073 // Get start addr and length of first argument
1074 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1075 Node* arg1_cnt = load_array_length(arg1);
1076
1077 // Get start addr and length of second argument
1078 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1079 Node* arg2_cnt = load_array_length(arg2);
1080
1081 // Check for arg1_cnt != arg2_cnt
1082 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1083 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1084 Node* if_ne = generate_slow_guard(bol, nullptr);
1085 if (if_ne != nullptr) {
1086 phi->init_req(2, intcon(0));
1087 region->init_req(2, if_ne);
1088 }
1089
1090 // Check for count == 0 is done by assembler code for StrEquals.
1091
1092 if (!stopped()) {
1093 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1094 phi->init_req(1, equals);
1095 region->init_req(1, control());
1096 }
1097 }
1098
1099 // post merge
1100 set_control(_gvn.transform(region));
1101 record_for_igvn(region);
1102
1103 set_result(_gvn.transform(phi));
1104 return true;
1105 }
1106
1107 //------------------------------inline_array_equals----------------------------
1108 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1109 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1110 Node* arg1 = argument(0);
1111 Node* arg2 = argument(1);
1112
1113 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1114 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1115 clear_upper_avx();
1116
1117 return true;
1118 }
1119
1120
1121 //------------------------------inline_countPositives------------------------------
1122 bool LibraryCallKit::inline_countPositives() {
1123 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1124 return false;
1125 }
1126
1127 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1128 // no receiver since it is static method
1129 Node* ba = argument(0);
1130 Node* offset = argument(1);
1131 Node* len = argument(2);
1132
1133 ba = must_be_not_null(ba, true);
1134
1135 // Range checks
1136 generate_string_range_check(ba, offset, len, false);
1137 if (stopped()) {
1138 return true;
1139 }
1140 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1141 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1142 set_result(_gvn.transform(result));
1143 clear_upper_avx();
1144 return true;
1145 }
1146
1147 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1148 Node* index = argument(0);
1149 Node* length = bt == T_INT ? argument(1) : argument(2);
1150 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1151 return false;
1152 }
1153
1154 // check that length is positive
1155 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1156 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1157
1158 {
1159 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1160 uncommon_trap(Deoptimization::Reason_intrinsic,
1161 Deoptimization::Action_make_not_entrant);
1162 }
1163
1164 if (stopped()) {
1165 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1166 return true;
1167 }
1168
1169 // length is now known positive, add a cast node to make this explicit
1170 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1171 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1172 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1173 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1174 casted_length = _gvn.transform(casted_length);
1175 replace_in_map(length, casted_length);
1176 length = casted_length;
1177
1178 // Use an unsigned comparison for the range check itself
1179 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1180 BoolTest::mask btest = BoolTest::lt;
1181 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1182 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1183 _gvn.set_type(rc, rc->Value(&_gvn));
1184 if (!rc_bool->is_Con()) {
1185 record_for_igvn(rc);
1186 }
1187 set_control(_gvn.transform(new IfTrueNode(rc)));
1188 {
1189 PreserveJVMState pjvms(this);
1190 set_control(_gvn.transform(new IfFalseNode(rc)));
1191 uncommon_trap(Deoptimization::Reason_range_check,
1192 Deoptimization::Action_make_not_entrant);
1193 }
1194
1195 if (stopped()) {
1196 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1197 return true;
1198 }
1199
1200 // index is now known to be >= 0 and < length, cast it
1201 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1202 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1203 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1204 result = _gvn.transform(result);
1205 set_result(result);
1206 replace_in_map(index, result);
1207 return true;
1208 }
1209
1210 //------------------------------inline_string_indexOf------------------------
1211 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1212 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1213 return false;
1214 }
1215 Node* src = argument(0);
1216 Node* tgt = argument(1);
1217
1218 // Make the merge point
1219 RegionNode* result_rgn = new RegionNode(4);
1220 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1221
1222 src = must_be_not_null(src, true);
1223 tgt = must_be_not_null(tgt, true);
1224
1225 // Get start addr and length of source string
1226 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1227 Node* src_count = load_array_length(src);
1228
1229 // Get start addr and length of substring
1230 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1231 Node* tgt_count = load_array_length(tgt);
1232
1233 Node* result = nullptr;
1234 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1235
1236 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1237 // Divide src size by 2 if String is UTF16 encoded
1238 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1239 }
1240 if (ae == StrIntrinsicNode::UU) {
1241 // Divide substring size by 2 if String is UTF16 encoded
1242 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1243 }
1244
1245 if (call_opt_stub) {
1246 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1247 StubRoutines::_string_indexof_array[ae],
1248 "stringIndexOf", TypePtr::BOTTOM, src_start,
1249 src_count, tgt_start, tgt_count);
1250 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1251 } else {
1252 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1253 result_rgn, result_phi, ae);
1254 }
1255 if (result != nullptr) {
1256 result_phi->init_req(3, result);
1257 result_rgn->init_req(3, control());
1258 }
1259 set_control(_gvn.transform(result_rgn));
1260 record_for_igvn(result_rgn);
1261 set_result(_gvn.transform(result_phi));
1262
1263 return true;
1264 }
1265
1266 //-----------------------------inline_string_indexOfI-----------------------
1267 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1268 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1269 return false;
1270 }
1271 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1272 return false;
1273 }
1274
1275 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1276 Node* src = argument(0); // byte[]
1277 Node* src_count = argument(1); // char count
1278 Node* tgt = argument(2); // byte[]
1279 Node* tgt_count = argument(3); // char count
1280 Node* from_index = argument(4); // char index
1281
1282 src = must_be_not_null(src, true);
1283 tgt = must_be_not_null(tgt, true);
1284
1285 // Multiply byte array index by 2 if String is UTF16 encoded
1286 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1287 src_count = _gvn.transform(new SubINode(src_count, from_index));
1288 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1289 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1290
1291 // Range checks
1292 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1293 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1294 if (stopped()) {
1295 return true;
1296 }
1297
1298 RegionNode* region = new RegionNode(5);
1299 Node* phi = new PhiNode(region, TypeInt::INT);
1300 Node* result = nullptr;
1301
1302 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1303
1304 if (call_opt_stub) {
1305 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1306 StubRoutines::_string_indexof_array[ae],
1307 "stringIndexOf", TypePtr::BOTTOM, src_start,
1308 src_count, tgt_start, tgt_count);
1309 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1310 } else {
1311 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1312 region, phi, ae);
1313 }
1314 if (result != nullptr) {
1315 // The result is index relative to from_index if substring was found, -1 otherwise.
1316 // Generate code which will fold into cmove.
1317 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1318 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1319
1320 Node* if_lt = generate_slow_guard(bol, nullptr);
1321 if (if_lt != nullptr) {
1322 // result == -1
1323 phi->init_req(3, result);
1324 region->init_req(3, if_lt);
1325 }
1326 if (!stopped()) {
1327 result = _gvn.transform(new AddINode(result, from_index));
1328 phi->init_req(4, result);
1329 region->init_req(4, control());
1330 }
1331 }
1332
1333 set_control(_gvn.transform(region));
1334 record_for_igvn(region);
1335 set_result(_gvn.transform(phi));
1336 clear_upper_avx();
1337
1338 return true;
1339 }
1340
1341 // Create StrIndexOfNode with fast path checks
1342 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1343 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1344 // Check for substr count > string count
1345 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1346 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1347 Node* if_gt = generate_slow_guard(bol, nullptr);
1348 if (if_gt != nullptr) {
1349 phi->init_req(1, intcon(-1));
1350 region->init_req(1, if_gt);
1351 }
1352 if (!stopped()) {
1353 // Check for substr count == 0
1354 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1355 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1356 Node* if_zero = generate_slow_guard(bol, nullptr);
1357 if (if_zero != nullptr) {
1358 phi->init_req(2, intcon(0));
1359 region->init_req(2, if_zero);
1360 }
1361 }
1362 if (!stopped()) {
1363 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1364 }
1365 return nullptr;
1366 }
1367
1368 //-----------------------------inline_string_indexOfChar-----------------------
1369 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1370 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1371 return false;
1372 }
1373 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1374 return false;
1375 }
1376 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1377 Node* src = argument(0); // byte[]
1378 Node* int_ch = argument(1);
1379 Node* from_index = argument(2);
1380 Node* max = argument(3);
1381
1382 src = must_be_not_null(src, true);
1383
1384 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1385 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1386 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1387
1388 // Range checks
1389 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1390
1391 // Check for int_ch >= 0
1392 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1393 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1394 {
1395 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1396 uncommon_trap(Deoptimization::Reason_intrinsic,
1397 Deoptimization::Action_maybe_recompile);
1398 }
1399 if (stopped()) {
1400 return true;
1401 }
1402
1403 RegionNode* region = new RegionNode(3);
1404 Node* phi = new PhiNode(region, TypeInt::INT);
1405
1406 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1407 C->set_has_split_ifs(true); // Has chance for split-if optimization
1408 _gvn.transform(result);
1409
1410 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1411 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1412
1413 Node* if_lt = generate_slow_guard(bol, nullptr);
1414 if (if_lt != nullptr) {
1415 // result == -1
1416 phi->init_req(2, result);
1417 region->init_req(2, if_lt);
1418 }
1419 if (!stopped()) {
1420 result = _gvn.transform(new AddINode(result, from_index));
1421 phi->init_req(1, result);
1422 region->init_req(1, control());
1423 }
1424 set_control(_gvn.transform(region));
1425 record_for_igvn(region);
1426 set_result(_gvn.transform(phi));
1427 clear_upper_avx();
1428
1429 return true;
1430 }
1431 //---------------------------inline_string_copy---------------------
1432 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1433 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1434 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1435 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1436 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1437 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1438 bool LibraryCallKit::inline_string_copy(bool compress) {
1439 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1440 return false;
1441 }
1442 int nargs = 5; // 2 oops, 3 ints
1443 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1444
1445 Node* src = argument(0);
1446 Node* src_offset = argument(1);
1447 Node* dst = argument(2);
1448 Node* dst_offset = argument(3);
1449 Node* length = argument(4);
1450
1451 // Check for allocation before we add nodes that would confuse
1452 // tightly_coupled_allocation()
1453 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1454
1455 // Figure out the size and type of the elements we will be copying.
1456 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1457 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1458 if (src_type == nullptr || dst_type == nullptr) {
1459 return false;
1460 }
1461 BasicType src_elem = src_type->elem()->array_element_basic_type();
1462 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1463 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1464 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1465 "Unsupported array types for inline_string_copy");
1466
1467 src = must_be_not_null(src, true);
1468 dst = must_be_not_null(dst, true);
1469
1470 // Convert char[] offsets to byte[] offsets
1471 bool convert_src = (compress && src_elem == T_BYTE);
1472 bool convert_dst = (!compress && dst_elem == T_BYTE);
1473 if (convert_src) {
1474 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1475 } else if (convert_dst) {
1476 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1477 }
1478
1479 // Range checks
1480 generate_string_range_check(src, src_offset, length, convert_src);
1481 generate_string_range_check(dst, dst_offset, length, convert_dst);
1482 if (stopped()) {
1483 return true;
1484 }
1485
1486 Node* src_start = array_element_address(src, src_offset, src_elem);
1487 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1488 // 'src_start' points to src array + scaled offset
1489 // 'dst_start' points to dst array + scaled offset
1490 Node* count = nullptr;
1491 if (compress) {
1492 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1493 } else {
1494 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1495 }
1496
1497 if (alloc != nullptr) {
1498 if (alloc->maybe_set_complete(&_gvn)) {
1499 // "You break it, you buy it."
1500 InitializeNode* init = alloc->initialization();
1501 assert(init->is_complete(), "we just did this");
1502 init->set_complete_with_arraycopy();
1503 assert(dst->is_CheckCastPP(), "sanity");
1504 assert(dst->in(0)->in(0) == init, "dest pinned");
1505 }
1506 // Do not let stores that initialize this object be reordered with
1507 // a subsequent store that would make this object accessible by
1508 // other threads.
1509 // Record what AllocateNode this StoreStore protects so that
1510 // escape analysis can go from the MemBarStoreStoreNode to the
1511 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1512 // based on the escape status of the AllocateNode.
1513 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1514 }
1515 if (compress) {
1516 set_result(_gvn.transform(count));
1517 }
1518 clear_upper_avx();
1519
1520 return true;
1521 }
1522
1523 #ifdef _LP64
1524 #define XTOP ,top() /*additional argument*/
1525 #else //_LP64
1526 #define XTOP /*no additional argument*/
1527 #endif //_LP64
1528
1529 //------------------------inline_string_toBytesU--------------------------
1530 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1531 bool LibraryCallKit::inline_string_toBytesU() {
1532 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1533 return false;
1534 }
1535 // Get the arguments.
1536 Node* value = argument(0);
1537 Node* offset = argument(1);
1538 Node* length = argument(2);
1539
1540 Node* newcopy = nullptr;
1541
1542 // Set the original stack and the reexecute bit for the interpreter to reexecute
1543 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1544 { PreserveReexecuteState preexecs(this);
1545 jvms()->set_should_reexecute(true);
1546
1547 // Check if a null path was taken unconditionally.
1548 value = null_check(value);
1549
1550 RegionNode* bailout = new RegionNode(1);
1551 record_for_igvn(bailout);
1552
1553 // Range checks
1554 generate_negative_guard(offset, bailout);
1555 generate_negative_guard(length, bailout);
1556 generate_limit_guard(offset, length, load_array_length(value), bailout);
1557 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1558 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1559
1560 if (bailout->req() > 1) {
1561 PreserveJVMState pjvms(this);
1562 set_control(_gvn.transform(bailout));
1563 uncommon_trap(Deoptimization::Reason_intrinsic,
1564 Deoptimization::Action_maybe_recompile);
1565 }
1566 if (stopped()) {
1567 return true;
1568 }
1569
1570 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1571 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1572 newcopy = new_array(klass_node, size, 0); // no arguments to push
1573 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1574 guarantee(alloc != nullptr, "created above");
1575
1576 // Calculate starting addresses.
1577 Node* src_start = array_element_address(value, offset, T_CHAR);
1578 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1579
1580 // Check if dst array address is aligned to HeapWordSize
1581 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1582 // If true, then check if src array address is aligned to HeapWordSize
1583 if (aligned) {
1584 const TypeInt* toffset = gvn().type(offset)->is_int();
1585 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1586 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1587 }
1588
1589 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1590 const char* copyfunc_name = "arraycopy";
1591 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1592 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1593 OptoRuntime::fast_arraycopy_Type(),
1594 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1595 src_start, dst_start, ConvI2X(length) XTOP);
1596 // Do not let reads from the cloned object float above the arraycopy.
1597 if (alloc->maybe_set_complete(&_gvn)) {
1598 // "You break it, you buy it."
1599 InitializeNode* init = alloc->initialization();
1600 assert(init->is_complete(), "we just did this");
1601 init->set_complete_with_arraycopy();
1602 assert(newcopy->is_CheckCastPP(), "sanity");
1603 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1604 }
1605 // Do not let stores that initialize this object be reordered with
1606 // a subsequent store that would make this object accessible by
1607 // other threads.
1608 // Record what AllocateNode this StoreStore protects so that
1609 // escape analysis can go from the MemBarStoreStoreNode to the
1610 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1611 // based on the escape status of the AllocateNode.
1612 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1613 } // original reexecute is set back here
1614
1615 C->set_has_split_ifs(true); // Has chance for split-if optimization
1616 if (!stopped()) {
1617 set_result(newcopy);
1618 }
1619 clear_upper_avx();
1620
1621 return true;
1622 }
1623
1624 //------------------------inline_string_getCharsU--------------------------
1625 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1626 bool LibraryCallKit::inline_string_getCharsU() {
1627 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1628 return false;
1629 }
1630
1631 // Get the arguments.
1632 Node* src = argument(0);
1633 Node* src_begin = argument(1);
1634 Node* src_end = argument(2); // exclusive offset (i < src_end)
1635 Node* dst = argument(3);
1636 Node* dst_begin = argument(4);
1637
1638 // Check for allocation before we add nodes that would confuse
1639 // tightly_coupled_allocation()
1640 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1641
1642 // Check if a null path was taken unconditionally.
1643 src = null_check(src);
1644 dst = null_check(dst);
1645 if (stopped()) {
1646 return true;
1647 }
1648
1649 // Get length and convert char[] offset to byte[] offset
1650 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1651 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1652
1653 // Range checks
1654 generate_string_range_check(src, src_begin, length, true);
1655 generate_string_range_check(dst, dst_begin, length, false);
1656 if (stopped()) {
1657 return true;
1658 }
1659
1660 if (!stopped()) {
1661 // Calculate starting addresses.
1662 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1663 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1664
1665 // Check if array addresses are aligned to HeapWordSize
1666 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1667 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1668 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1669 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1670
1671 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1672 const char* copyfunc_name = "arraycopy";
1673 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1674 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1675 OptoRuntime::fast_arraycopy_Type(),
1676 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1677 src_start, dst_start, ConvI2X(length) XTOP);
1678 // Do not let reads from the cloned object float above the arraycopy.
1679 if (alloc != nullptr) {
1680 if (alloc->maybe_set_complete(&_gvn)) {
1681 // "You break it, you buy it."
1682 InitializeNode* init = alloc->initialization();
1683 assert(init->is_complete(), "we just did this");
1684 init->set_complete_with_arraycopy();
1685 assert(dst->is_CheckCastPP(), "sanity");
1686 assert(dst->in(0)->in(0) == init, "dest pinned");
1687 }
1688 // Do not let stores that initialize this object be reordered with
1689 // a subsequent store that would make this object accessible by
1690 // other threads.
1691 // Record what AllocateNode this StoreStore protects so that
1692 // escape analysis can go from the MemBarStoreStoreNode to the
1693 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1694 // based on the escape status of the AllocateNode.
1695 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1696 } else {
1697 insert_mem_bar(Op_MemBarCPUOrder);
1698 }
1699 }
1700
1701 C->set_has_split_ifs(true); // Has chance for split-if optimization
1702 return true;
1703 }
1704
1705 //----------------------inline_string_char_access----------------------------
1706 // Store/Load char to/from byte[] array.
1707 // static void StringUTF16.putChar(byte[] val, int index, int c)
1708 // static char StringUTF16.getChar(byte[] val, int index)
1709 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1710 Node* value = argument(0);
1711 Node* index = argument(1);
1712 Node* ch = is_store ? argument(2) : nullptr;
1713
1714 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1715 // correctly requires matched array shapes.
1716 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1717 "sanity: byte[] and char[] bases agree");
1718 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1719 "sanity: byte[] and char[] scales agree");
1720
1721 // Bail when getChar over constants is requested: constant folding would
1722 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1723 // Java method would constant fold nicely instead.
1724 if (!is_store && value->is_Con() && index->is_Con()) {
1725 return false;
1726 }
1727
1728 // Save state and restore on bailout
1729 SavedState old_state(this);
1730
1731 value = must_be_not_null(value, true);
1732
1733 Node* adr = array_element_address(value, index, T_CHAR);
1734 if (adr->is_top()) {
1735 return false;
1736 }
1737 old_state.discard();
1738 if (is_store) {
1739 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1740 } else {
1741 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);
1742 set_result(ch);
1743 }
1744 return true;
1745 }
1746
1747
1748 //------------------------------inline_math-----------------------------------
1749 // public static double Math.abs(double)
1750 // public static double Math.sqrt(double)
1751 // public static double Math.log(double)
1752 // public static double Math.log10(double)
1753 // public static double Math.round(double)
1754 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1755 Node* arg = argument(0);
1756 Node* n = nullptr;
1757 switch (id) {
1758 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1759 case vmIntrinsics::_dsqrt:
1760 case vmIntrinsics::_dsqrt_strict:
1761 n = new SqrtDNode(C, control(), arg); break;
1762 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1763 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1764 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1765 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1766 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1767 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1768 default: fatal_unexpected_iid(id); break;
1769 }
1770 set_result(_gvn.transform(n));
1771 return true;
1772 }
1773
1774 //------------------------------inline_math-----------------------------------
1775 // public static float Math.abs(float)
1776 // public static int Math.abs(int)
1777 // public static long Math.abs(long)
1778 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1779 Node* arg = argument(0);
1780 Node* n = nullptr;
1781 switch (id) {
1782 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1783 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1784 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1785 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1786 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1787 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1788 default: fatal_unexpected_iid(id); break;
1789 }
1790 set_result(_gvn.transform(n));
1791 return true;
1792 }
1793
1794 //------------------------------runtime_math-----------------------------
1795 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1796 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1797 "must be (DD)D or (D)D type");
1798
1799 // Inputs
1800 Node* a = argument(0);
1801 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1802
1803 const TypePtr* no_memory_effects = nullptr;
1804 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1805 no_memory_effects,
1806 a, top(), b, b ? top() : nullptr);
1807 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1808 #ifdef ASSERT
1809 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1810 assert(value_top == top(), "second value must be top");
1811 #endif
1812
1813 set_result(value);
1814 return true;
1815 }
1816
1817 //------------------------------inline_math_pow-----------------------------
1818 bool LibraryCallKit::inline_math_pow() {
1819 Node* exp = argument(2);
1820 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1821 if (d != nullptr) {
1822 if (d->getd() == 2.0) {
1823 // Special case: pow(x, 2.0) => x * x
1824 Node* base = argument(0);
1825 set_result(_gvn.transform(new MulDNode(base, base)));
1826 return true;
1827 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1828 // Special case: pow(x, 0.5) => sqrt(x)
1829 Node* base = argument(0);
1830 Node* zero = _gvn.zerocon(T_DOUBLE);
1831
1832 RegionNode* region = new RegionNode(3);
1833 Node* phi = new PhiNode(region, Type::DOUBLE);
1834
1835 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1836 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1837 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1838 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1839 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1840
1841 Node* if_pow = generate_slow_guard(test, nullptr);
1842 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1843 phi->init_req(1, value_sqrt);
1844 region->init_req(1, control());
1845
1846 if (if_pow != nullptr) {
1847 set_control(if_pow);
1848 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1849 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1850 const TypePtr* no_memory_effects = nullptr;
1851 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1852 no_memory_effects, base, top(), exp, top());
1853 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1854 #ifdef ASSERT
1855 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1856 assert(value_top == top(), "second value must be top");
1857 #endif
1858 phi->init_req(2, value_pow);
1859 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1860 }
1861
1862 C->set_has_split_ifs(true); // Has chance for split-if optimization
1863 set_control(_gvn.transform(region));
1864 record_for_igvn(region);
1865 set_result(_gvn.transform(phi));
1866
1867 return true;
1868 }
1869 }
1870
1871 return StubRoutines::dpow() != nullptr ?
1872 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1873 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1874 }
1875
1876 //------------------------------inline_math_native-----------------------------
1877 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1878 switch (id) {
1879 case vmIntrinsics::_dsin:
1880 return StubRoutines::dsin() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1883 case vmIntrinsics::_dcos:
1884 return StubRoutines::dcos() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1887 case vmIntrinsics::_dtan:
1888 return StubRoutines::dtan() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1891 case vmIntrinsics::_dsinh:
1892 return StubRoutines::dsinh() != nullptr ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1894 case vmIntrinsics::_dtanh:
1895 return StubRoutines::dtanh() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1897 case vmIntrinsics::_dcbrt:
1898 return StubRoutines::dcbrt() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1900 case vmIntrinsics::_dexp:
1901 return StubRoutines::dexp() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1904 case vmIntrinsics::_dlog:
1905 return StubRoutines::dlog() != nullptr ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1907 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1908 case vmIntrinsics::_dlog10:
1909 return StubRoutines::dlog10() != nullptr ?
1910 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1911 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1912
1913 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_ceil:
1915 case vmIntrinsics::_floor:
1916 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1917
1918 case vmIntrinsics::_dsqrt:
1919 case vmIntrinsics::_dsqrt_strict:
1920 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1921 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1922 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1923 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1924 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1925
1926 case vmIntrinsics::_dpow: return inline_math_pow();
1927 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1928 case vmIntrinsics::_fcopySign: return inline_math(id);
1929 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1930 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1931 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1932
1933 // These intrinsics are not yet correctly implemented
1934 case vmIntrinsics::_datan2:
1935 return false;
1936
1937 default:
1938 fatal_unexpected_iid(id);
1939 return false;
1940 }
1941 }
1942
1943 //----------------------------inline_notify-----------------------------------*
1944 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1945 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1946 address func;
1947 if (id == vmIntrinsics::_notify) {
1948 func = OptoRuntime::monitor_notify_Java();
1949 } else {
1950 func = OptoRuntime::monitor_notifyAll_Java();
1951 }
1952 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1953 make_slow_call_ex(call, env()->Throwable_klass(), false);
1954 return true;
1955 }
1956
1957
1958 //----------------------------inline_min_max-----------------------------------
1959 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1960 Node* a = nullptr;
1961 Node* b = nullptr;
1962 Node* n = nullptr;
1963 switch (id) {
1964 case vmIntrinsics::_min:
1965 case vmIntrinsics::_max:
1966 case vmIntrinsics::_minF:
1967 case vmIntrinsics::_maxF:
1968 case vmIntrinsics::_minF_strict:
1969 case vmIntrinsics::_maxF_strict:
1970 case vmIntrinsics::_min_strict:
1971 case vmIntrinsics::_max_strict:
1972 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1973 a = argument(0);
1974 b = argument(1);
1975 break;
1976 case vmIntrinsics::_minD:
1977 case vmIntrinsics::_maxD:
1978 case vmIntrinsics::_minD_strict:
1979 case vmIntrinsics::_maxD_strict:
1980 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1981 a = argument(0);
1982 b = argument(2);
1983 break;
1984 case vmIntrinsics::_minL:
1985 case vmIntrinsics::_maxL:
1986 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1987 a = argument(0);
1988 b = argument(2);
1989 break;
1990 default:
1991 fatal_unexpected_iid(id);
1992 break;
1993 }
1994
1995 switch (id) {
1996 case vmIntrinsics::_min:
1997 case vmIntrinsics::_min_strict:
1998 n = new MinINode(a, b);
1999 break;
2000 case vmIntrinsics::_max:
2001 case vmIntrinsics::_max_strict:
2002 n = new MaxINode(a, b);
2003 break;
2004 case vmIntrinsics::_minF:
2005 case vmIntrinsics::_minF_strict:
2006 n = new MinFNode(a, b);
2007 break;
2008 case vmIntrinsics::_maxF:
2009 case vmIntrinsics::_maxF_strict:
2010 n = new MaxFNode(a, b);
2011 break;
2012 case vmIntrinsics::_minD:
2013 case vmIntrinsics::_minD_strict:
2014 n = new MinDNode(a, b);
2015 break;
2016 case vmIntrinsics::_maxD:
2017 case vmIntrinsics::_maxD_strict:
2018 n = new MaxDNode(a, b);
2019 break;
2020 case vmIntrinsics::_minL:
2021 n = new MinLNode(_gvn.C, a, b);
2022 break;
2023 case vmIntrinsics::_maxL:
2024 n = new MaxLNode(_gvn.C, a, b);
2025 break;
2026 default:
2027 fatal_unexpected_iid(id);
2028 break;
2029 }
2030
2031 set_result(_gvn.transform(n));
2032 return true;
2033 }
2034
2035 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2036 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2037 env()->ArithmeticException_instance())) {
2038 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2039 // so let's bail out intrinsic rather than risking deopting again.
2040 return false;
2041 }
2042
2043 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2044 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2045 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2046 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2047
2048 {
2049 PreserveJVMState pjvms(this);
2050 PreserveReexecuteState preexecs(this);
2051 jvms()->set_should_reexecute(true);
2052
2053 set_control(slow_path);
2054 set_i_o(i_o());
2055
2056 builtin_throw(Deoptimization::Reason_intrinsic,
2057 env()->ArithmeticException_instance(),
2058 /*allow_too_many_traps*/ false);
2059 }
2060
2061 set_control(fast_path);
2062 set_result(math);
2063 return true;
2064 }
2065
2066 template <typename OverflowOp>
2067 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2068 typedef typename OverflowOp::MathOp MathOp;
2069
2070 MathOp* mathOp = new MathOp(arg1, arg2);
2071 Node* operation = _gvn.transform( mathOp );
2072 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2073 return inline_math_mathExact(operation, ofcheck);
2074 }
2075
2076 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2077 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2078 }
2079
2080 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2081 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2082 }
2083
2084 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2085 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2086 }
2087
2088 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2089 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2090 }
2091
2092 bool LibraryCallKit::inline_math_negateExactI() {
2093 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2094 }
2095
2096 bool LibraryCallKit::inline_math_negateExactL() {
2097 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2098 }
2099
2100 bool LibraryCallKit::inline_math_multiplyExactI() {
2101 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2102 }
2103
2104 bool LibraryCallKit::inline_math_multiplyExactL() {
2105 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2106 }
2107
2108 bool LibraryCallKit::inline_math_multiplyHigh() {
2109 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2110 return true;
2111 }
2112
2113 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2114 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2115 return true;
2116 }
2117
2118 inline int
2119 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2120 const TypePtr* base_type = TypePtr::NULL_PTR;
2121 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2122 if (base_type == nullptr) {
2123 // Unknown type.
2124 return Type::AnyPtr;
2125 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2126 // Since this is a null+long form, we have to switch to a rawptr.
2127 base = _gvn.transform(new CastX2PNode(offset));
2128 offset = MakeConX(0);
2129 return Type::RawPtr;
2130 } else if (base_type->base() == Type::RawPtr) {
2131 return Type::RawPtr;
2132 } else if (base_type->isa_oopptr()) {
2133 // Base is never null => always a heap address.
2134 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2135 return Type::OopPtr;
2136 }
2137 // Offset is small => always a heap address.
2138 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2139 if (offset_type != nullptr &&
2140 base_type->offset() == 0 && // (should always be?)
2141 offset_type->_lo >= 0 &&
2142 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2143 return Type::OopPtr;
2144 } else if (type == T_OBJECT) {
2145 // off heap access to an oop doesn't make any sense. Has to be on
2146 // heap.
2147 return Type::OopPtr;
2148 }
2149 // Otherwise, it might either be oop+off or null+addr.
2150 return Type::AnyPtr;
2151 } else {
2152 // No information:
2153 return Type::AnyPtr;
2154 }
2155 }
2156
2157 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2158 Node* uncasted_base = base;
2159 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2160 if (kind == Type::RawPtr) {
2161 return basic_plus_adr(top(), uncasted_base, offset);
2162 } else if (kind == Type::AnyPtr) {
2163 assert(base == uncasted_base, "unexpected base change");
2164 if (can_cast) {
2165 if (!_gvn.type(base)->speculative_maybe_null() &&
2166 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2167 // According to profiling, this access is always on
2168 // heap. Casting the base to not null and thus avoiding membars
2169 // around the access should allow better optimizations
2170 Node* null_ctl = top();
2171 base = null_check_oop(base, &null_ctl, true, true, true);
2172 assert(null_ctl->is_top(), "no null control here");
2173 return basic_plus_adr(base, offset);
2174 } else if (_gvn.type(base)->speculative_always_null() &&
2175 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2176 // According to profiling, this access is always off
2177 // heap.
2178 base = null_assert(base);
2179 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2180 offset = MakeConX(0);
2181 return basic_plus_adr(top(), raw_base, offset);
2182 }
2183 }
2184 // We don't know if it's an on heap or off heap access. Fall back
2185 // to raw memory access.
2186 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2187 return basic_plus_adr(top(), raw, offset);
2188 } else {
2189 assert(base == uncasted_base, "unexpected base change");
2190 // We know it's an on heap access so base can't be null
2191 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2192 base = must_be_not_null(base, true);
2193 }
2194 return basic_plus_adr(base, offset);
2195 }
2196 }
2197
2198 //--------------------------inline_number_methods-----------------------------
2199 // inline int Integer.numberOfLeadingZeros(int)
2200 // inline int Long.numberOfLeadingZeros(long)
2201 //
2202 // inline int Integer.numberOfTrailingZeros(int)
2203 // inline int Long.numberOfTrailingZeros(long)
2204 //
2205 // inline int Integer.bitCount(int)
2206 // inline int Long.bitCount(long)
2207 //
2208 // inline char Character.reverseBytes(char)
2209 // inline short Short.reverseBytes(short)
2210 // inline int Integer.reverseBytes(int)
2211 // inline long Long.reverseBytes(long)
2212 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2213 Node* arg = argument(0);
2214 Node* n = nullptr;
2215 switch (id) {
2216 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2217 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2218 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2219 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2220 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2221 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2222 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2223 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2224 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2225 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2226 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2227 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2228 default: fatal_unexpected_iid(id); break;
2229 }
2230 set_result(_gvn.transform(n));
2231 return true;
2232 }
2233
2234 //--------------------------inline_bitshuffle_methods-----------------------------
2235 // inline int Integer.compress(int, int)
2236 // inline int Integer.expand(int, int)
2237 // inline long Long.compress(long, long)
2238 // inline long Long.expand(long, long)
2239 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2240 Node* n = nullptr;
2241 switch (id) {
2242 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2243 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2244 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2245 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2246 default: fatal_unexpected_iid(id); break;
2247 }
2248 set_result(_gvn.transform(n));
2249 return true;
2250 }
2251
2252 //--------------------------inline_number_methods-----------------------------
2253 // inline int Integer.compareUnsigned(int, int)
2254 // inline int Long.compareUnsigned(long, long)
2255 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2256 Node* arg1 = argument(0);
2257 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2258 Node* n = nullptr;
2259 switch (id) {
2260 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2261 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2262 default: fatal_unexpected_iid(id); break;
2263 }
2264 set_result(_gvn.transform(n));
2265 return true;
2266 }
2267
2268 //--------------------------inline_unsigned_divmod_methods-----------------------------
2269 // inline int Integer.divideUnsigned(int, int)
2270 // inline int Integer.remainderUnsigned(int, int)
2271 // inline long Long.divideUnsigned(long, long)
2272 // inline long Long.remainderUnsigned(long, long)
2273 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2274 Node* n = nullptr;
2275 switch (id) {
2276 case vmIntrinsics::_divideUnsigned_i: {
2277 zero_check_int(argument(1));
2278 // Compile-time detect of null-exception
2279 if (stopped()) {
2280 return true; // keep the graph constructed so far
2281 }
2282 n = new UDivINode(control(), argument(0), argument(1));
2283 break;
2284 }
2285 case vmIntrinsics::_divideUnsigned_l: {
2286 zero_check_long(argument(2));
2287 // Compile-time detect of null-exception
2288 if (stopped()) {
2289 return true; // keep the graph constructed so far
2290 }
2291 n = new UDivLNode(control(), argument(0), argument(2));
2292 break;
2293 }
2294 case vmIntrinsics::_remainderUnsigned_i: {
2295 zero_check_int(argument(1));
2296 // Compile-time detect of null-exception
2297 if (stopped()) {
2298 return true; // keep the graph constructed so far
2299 }
2300 n = new UModINode(control(), argument(0), argument(1));
2301 break;
2302 }
2303 case vmIntrinsics::_remainderUnsigned_l: {
2304 zero_check_long(argument(2));
2305 // Compile-time detect of null-exception
2306 if (stopped()) {
2307 return true; // keep the graph constructed so far
2308 }
2309 n = new UModLNode(control(), argument(0), argument(2));
2310 break;
2311 }
2312 default: fatal_unexpected_iid(id); break;
2313 }
2314 set_result(_gvn.transform(n));
2315 return true;
2316 }
2317
2318 //----------------------------inline_unsafe_access----------------------------
2319
2320 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2321 // Attempt to infer a sharper value type from the offset and base type.
2322 ciKlass* sharpened_klass = nullptr;
2323
2324 // See if it is an instance field, with an object type.
2325 if (alias_type->field() != nullptr) {
2326 if (alias_type->field()->type()->is_klass()) {
2327 sharpened_klass = alias_type->field()->type()->as_klass();
2328 }
2329 }
2330
2331 const TypeOopPtr* result = nullptr;
2332 // See if it is a narrow oop array.
2333 if (adr_type->isa_aryptr()) {
2334 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2335 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2336 if (elem_type != nullptr && elem_type->is_loaded()) {
2337 // Sharpen the value type.
2338 result = elem_type;
2339 }
2340 }
2341 }
2342
2343 // The sharpened class might be unloaded if there is no class loader
2344 // contraint in place.
2345 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2346 // Sharpen the value type.
2347 result = TypeOopPtr::make_from_klass(sharpened_klass);
2348 }
2349 if (result != nullptr) {
2350 #ifndef PRODUCT
2351 if (C->print_intrinsics() || C->print_inlining()) {
2352 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2353 tty->print(" sharpened value: "); result->dump(); tty->cr();
2354 }
2355 #endif
2356 }
2357 return result;
2358 }
2359
2360 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2361 switch (kind) {
2362 case Relaxed:
2363 return MO_UNORDERED;
2364 case Opaque:
2365 return MO_RELAXED;
2366 case Acquire:
2367 return MO_ACQUIRE;
2368 case Release:
2369 return MO_RELEASE;
2370 case Volatile:
2371 return MO_SEQ_CST;
2372 default:
2373 ShouldNotReachHere();
2374 return 0;
2375 }
2376 }
2377
2378 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2379 _kit(kit),
2380 _sp(kit->sp()),
2381 _jvms(kit->jvms()),
2382 _map(kit->clone_map()),
2383 _discarded(false)
2384 {
2385 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2386 Node* out = kit->control()->fast_out(i);
2387 if (out->is_CFG()) {
2388 _ctrl_succ.push(out);
2389 }
2390 }
2391 }
2392
2393 LibraryCallKit::SavedState::~SavedState() {
2394 if (_discarded) {
2395 _kit->destruct_map_clone(_map);
2396 return;
2397 }
2398 _kit->jvms()->set_map(_map);
2399 _kit->jvms()->set_sp(_sp);
2400 _map->set_jvms(_kit->jvms());
2401 _kit->set_map(_map);
2402 _kit->set_sp(_sp);
2403 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2404 Node* out = _kit->control()->fast_out(i);
2405 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2406 _kit->_gvn.hash_delete(out);
2407 out->set_req(0, _kit->C->top());
2408 _kit->C->record_for_igvn(out);
2409 --i; --imax;
2410 _kit->_gvn.hash_find_insert(out);
2411 }
2412 }
2413 }
2414
2415 void LibraryCallKit::SavedState::discard() {
2416 _discarded = true;
2417 }
2418
2419 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2420 if (callee()->is_static()) return false; // caller must have the capability!
2421 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2422 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2423 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2424 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2425
2426 if (is_reference_type(type)) {
2427 decorators |= ON_UNKNOWN_OOP_REF;
2428 }
2429
2430 if (unaligned) {
2431 decorators |= C2_UNALIGNED;
2432 }
2433
2434 #ifndef PRODUCT
2435 {
2436 ResourceMark rm;
2437 // Check the signatures.
2438 ciSignature* sig = callee()->signature();
2439 #ifdef ASSERT
2440 if (!is_store) {
2441 // Object getReference(Object base, int/long offset), etc.
2442 BasicType rtype = sig->return_type()->basic_type();
2443 assert(rtype == type, "getter must return the expected value");
2444 assert(sig->count() == 2, "oop getter has 2 arguments");
2445 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2446 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2447 } else {
2448 // void putReference(Object base, int/long offset, Object x), etc.
2449 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2450 assert(sig->count() == 3, "oop putter has 3 arguments");
2451 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2452 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2453 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2454 assert(vtype == type, "putter must accept the expected value");
2455 }
2456 #endif // ASSERT
2457 }
2458 #endif //PRODUCT
2459
2460 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2461
2462 Node* receiver = argument(0); // type: oop
2463
2464 // Build address expression.
2465 Node* heap_base_oop = top();
2466
2467 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2468 Node* base = argument(1); // type: oop
2469 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2470 Node* offset = argument(2); // type: long
2471 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2472 // to be plain byte offsets, which are also the same as those accepted
2473 // by oopDesc::field_addr.
2474 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2475 "fieldOffset must be byte-scaled");
2476 // 32-bit machines ignore the high half!
2477 offset = ConvL2X(offset);
2478
2479 // Save state and restore on bailout
2480 SavedState old_state(this);
2481
2482 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2483 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2484
2485 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2486 if (type != T_OBJECT) {
2487 decorators |= IN_NATIVE; // off-heap primitive access
2488 } else {
2489 return false; // off-heap oop accesses are not supported
2490 }
2491 } else {
2492 heap_base_oop = base; // on-heap or mixed access
2493 }
2494
2495 // Can base be null? Otherwise, always on-heap access.
2496 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2497
2498 if (!can_access_non_heap) {
2499 decorators |= IN_HEAP;
2500 }
2501
2502 Node* val = is_store ? argument(4) : nullptr;
2503
2504 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2505 if (adr_type == TypePtr::NULL_PTR) {
2506 return false; // off-heap access with zero address
2507 }
2508
2509 // Try to categorize the address.
2510 Compile::AliasType* alias_type = C->alias_type(adr_type);
2511 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2512
2513 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2514 alias_type->adr_type() == TypeAryPtr::RANGE) {
2515 return false; // not supported
2516 }
2517
2518 bool mismatched = false;
2519 BasicType bt = alias_type->basic_type();
2520 if (bt != T_ILLEGAL) {
2521 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2522 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2523 // Alias type doesn't differentiate between byte[] and boolean[]).
2524 // Use address type to get the element type.
2525 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2526 }
2527 if (is_reference_type(bt, true)) {
2528 // accessing an array field with getReference is not a mismatch
2529 bt = T_OBJECT;
2530 }
2531 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2532 // Don't intrinsify mismatched object accesses
2533 return false;
2534 }
2535 mismatched = (bt != type);
2536 } else if (alias_type->adr_type()->isa_oopptr()) {
2537 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2538 }
2539
2540 old_state.discard();
2541 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2542
2543 if (mismatched) {
2544 decorators |= C2_MISMATCHED;
2545 }
2546
2547 // First guess at the value type.
2548 const Type *value_type = Type::get_const_basic_type(type);
2549
2550 // Figure out the memory ordering.
2551 decorators |= mo_decorator_for_access_kind(kind);
2552
2553 if (!is_store && type == T_OBJECT) {
2554 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2555 if (tjp != nullptr) {
2556 value_type = tjp;
2557 }
2558 }
2559
2560 receiver = null_check(receiver);
2561 if (stopped()) {
2562 return true;
2563 }
2564 // Heap pointers get a null-check from the interpreter,
2565 // as a courtesy. However, this is not guaranteed by Unsafe,
2566 // and it is not possible to fully distinguish unintended nulls
2567 // from intended ones in this API.
2568
2569 if (!is_store) {
2570 Node* p = nullptr;
2571 // Try to constant fold a load from a constant field
2572 ciField* field = alias_type->field();
2573 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2574 // final or stable field
2575 p = make_constant_from_field(field, heap_base_oop);
2576 }
2577
2578 if (p == nullptr) { // Could not constant fold the load
2579 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2580 // Normalize the value returned by getBoolean in the following cases
2581 if (type == T_BOOLEAN &&
2582 (mismatched ||
2583 heap_base_oop == top() || // - heap_base_oop is null or
2584 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2585 // and the unsafe access is made to large offset
2586 // (i.e., larger than the maximum offset necessary for any
2587 // field access)
2588 ) {
2589 IdealKit ideal = IdealKit(this);
2590 #define __ ideal.
2591 IdealVariable normalized_result(ideal);
2592 __ declarations_done();
2593 __ set(normalized_result, p);
2594 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2595 __ set(normalized_result, ideal.ConI(1));
2596 ideal.end_if();
2597 final_sync(ideal);
2598 p = __ value(normalized_result);
2599 #undef __
2600 }
2601 }
2602 if (type == T_ADDRESS) {
2603 p = gvn().transform(new CastP2XNode(nullptr, p));
2604 p = ConvX2UL(p);
2605 }
2606 // The load node has the control of the preceding MemBarCPUOrder. All
2607 // following nodes will have the control of the MemBarCPUOrder inserted at
2608 // the end of this method. So, pushing the load onto the stack at a later
2609 // point is fine.
2610 set_result(p);
2611 } else {
2612 if (bt == T_ADDRESS) {
2613 // Repackage the long as a pointer.
2614 val = ConvL2X(val);
2615 val = gvn().transform(new CastX2PNode(val));
2616 }
2617 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2618 }
2619
2620 return true;
2621 }
2622
2623 //----------------------------inline_unsafe_load_store----------------------------
2624 // This method serves a couple of different customers (depending on LoadStoreKind):
2625 //
2626 // LS_cmp_swap:
2627 //
2628 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2629 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2630 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2631 //
2632 // LS_cmp_swap_weak:
2633 //
2634 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2635 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2636 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2637 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2638 //
2639 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2640 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2641 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2642 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2643 //
2644 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2645 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2646 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2647 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2648 //
2649 // LS_cmp_exchange:
2650 //
2651 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2652 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2653 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2654 //
2655 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2656 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2657 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2658 //
2659 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2660 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2661 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2662 //
2663 // LS_get_add:
2664 //
2665 // int getAndAddInt( Object o, long offset, int delta)
2666 // long getAndAddLong(Object o, long offset, long delta)
2667 //
2668 // LS_get_set:
2669 //
2670 // int getAndSet(Object o, long offset, int newValue)
2671 // long getAndSet(Object o, long offset, long newValue)
2672 // Object getAndSet(Object o, long offset, Object newValue)
2673 //
2674 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2675 // This basic scheme here is the same as inline_unsafe_access, but
2676 // differs in enough details that combining them would make the code
2677 // overly confusing. (This is a true fact! I originally combined
2678 // them, but even I was confused by it!) As much code/comments as
2679 // possible are retained from inline_unsafe_access though to make
2680 // the correspondences clearer. - dl
2681
2682 if (callee()->is_static()) return false; // caller must have the capability!
2683
2684 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2685 decorators |= mo_decorator_for_access_kind(access_kind);
2686
2687 #ifndef PRODUCT
2688 BasicType rtype;
2689 {
2690 ResourceMark rm;
2691 // Check the signatures.
2692 ciSignature* sig = callee()->signature();
2693 rtype = sig->return_type()->basic_type();
2694 switch(kind) {
2695 case LS_get_add:
2696 case LS_get_set: {
2697 // Check the signatures.
2698 #ifdef ASSERT
2699 assert(rtype == type, "get and set must return the expected type");
2700 assert(sig->count() == 3, "get and set has 3 arguments");
2701 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2702 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2703 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2704 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2705 #endif // ASSERT
2706 break;
2707 }
2708 case LS_cmp_swap:
2709 case LS_cmp_swap_weak: {
2710 // Check the signatures.
2711 #ifdef ASSERT
2712 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2713 assert(sig->count() == 4, "CAS has 4 arguments");
2714 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2715 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2716 #endif // ASSERT
2717 break;
2718 }
2719 case LS_cmp_exchange: {
2720 // Check the signatures.
2721 #ifdef ASSERT
2722 assert(rtype == type, "CAS must return the expected type");
2723 assert(sig->count() == 4, "CAS has 4 arguments");
2724 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2725 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2726 #endif // ASSERT
2727 break;
2728 }
2729 default:
2730 ShouldNotReachHere();
2731 }
2732 }
2733 #endif //PRODUCT
2734
2735 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2736
2737 // Get arguments:
2738 Node* receiver = nullptr;
2739 Node* base = nullptr;
2740 Node* offset = nullptr;
2741 Node* oldval = nullptr;
2742 Node* newval = nullptr;
2743 switch(kind) {
2744 case LS_cmp_swap:
2745 case LS_cmp_swap_weak:
2746 case LS_cmp_exchange: {
2747 const bool two_slot_type = type2size[type] == 2;
2748 receiver = argument(0); // type: oop
2749 base = argument(1); // type: oop
2750 offset = argument(2); // type: long
2751 oldval = argument(4); // type: oop, int, or long
2752 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2753 break;
2754 }
2755 case LS_get_add:
2756 case LS_get_set: {
2757 receiver = argument(0); // type: oop
2758 base = argument(1); // type: oop
2759 offset = argument(2); // type: long
2760 oldval = nullptr;
2761 newval = argument(4); // type: oop, int, or long
2762 break;
2763 }
2764 default:
2765 ShouldNotReachHere();
2766 }
2767
2768 // Build field offset expression.
2769 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2770 // to be plain byte offsets, which are also the same as those accepted
2771 // by oopDesc::field_addr.
2772 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2773 // 32-bit machines ignore the high half of long offsets
2774 offset = ConvL2X(offset);
2775 // Save state and restore on bailout
2776 SavedState old_state(this);
2777 Node* adr = make_unsafe_address(base, offset,type, false);
2778 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2779
2780 Compile::AliasType* alias_type = C->alias_type(adr_type);
2781 BasicType bt = alias_type->basic_type();
2782 if (bt != T_ILLEGAL &&
2783 (is_reference_type(bt) != (type == T_OBJECT))) {
2784 // Don't intrinsify mismatched object accesses.
2785 return false;
2786 }
2787
2788 old_state.discard();
2789
2790 // For CAS, unlike inline_unsafe_access, there seems no point in
2791 // trying to refine types. Just use the coarse types here.
2792 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2793 const Type *value_type = Type::get_const_basic_type(type);
2794
2795 switch (kind) {
2796 case LS_get_set:
2797 case LS_cmp_exchange: {
2798 if (type == T_OBJECT) {
2799 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2800 if (tjp != nullptr) {
2801 value_type = tjp;
2802 }
2803 }
2804 break;
2805 }
2806 case LS_cmp_swap:
2807 case LS_cmp_swap_weak:
2808 case LS_get_add:
2809 break;
2810 default:
2811 ShouldNotReachHere();
2812 }
2813
2814 // Null check receiver.
2815 receiver = null_check(receiver);
2816 if (stopped()) {
2817 return true;
2818 }
2819
2820 int alias_idx = C->get_alias_index(adr_type);
2821
2822 if (is_reference_type(type)) {
2823 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2824
2825 // Transformation of a value which could be null pointer (CastPP #null)
2826 // could be delayed during Parse (for example, in adjust_map_after_if()).
2827 // Execute transformation here to avoid barrier generation in such case.
2828 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2829 newval = _gvn.makecon(TypePtr::NULL_PTR);
2830
2831 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2832 // Refine the value to a null constant, when it is known to be null
2833 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2834 }
2835 }
2836
2837 Node* result = nullptr;
2838 switch (kind) {
2839 case LS_cmp_exchange: {
2840 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2841 oldval, newval, value_type, type, decorators);
2842 break;
2843 }
2844 case LS_cmp_swap_weak:
2845 decorators |= C2_WEAK_CMPXCHG;
2846 case LS_cmp_swap: {
2847 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2848 oldval, newval, value_type, type, decorators);
2849 break;
2850 }
2851 case LS_get_set: {
2852 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2853 newval, value_type, type, decorators);
2854 break;
2855 }
2856 case LS_get_add: {
2857 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2858 newval, value_type, type, decorators);
2859 break;
2860 }
2861 default:
2862 ShouldNotReachHere();
2863 }
2864
2865 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2866 set_result(result);
2867 return true;
2868 }
2869
2870 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2871 // Regardless of form, don't allow previous ld/st to move down,
2872 // then issue acquire, release, or volatile mem_bar.
2873 insert_mem_bar(Op_MemBarCPUOrder);
2874 switch(id) {
2875 case vmIntrinsics::_loadFence:
2876 insert_mem_bar(Op_LoadFence);
2877 return true;
2878 case vmIntrinsics::_storeFence:
2879 insert_mem_bar(Op_StoreFence);
2880 return true;
2881 case vmIntrinsics::_storeStoreFence:
2882 insert_mem_bar(Op_StoreStoreFence);
2883 return true;
2884 case vmIntrinsics::_fullFence:
2885 insert_mem_bar(Op_MemBarVolatile);
2886 return true;
2887 default:
2888 fatal_unexpected_iid(id);
2889 return false;
2890 }
2891 }
2892
2893 bool LibraryCallKit::inline_onspinwait() {
2894 insert_mem_bar(Op_OnSpinWait);
2895 return true;
2896 }
2897
2898 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2899 if (!kls->is_Con()) {
2900 return true;
2901 }
2902 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2903 if (klsptr == nullptr) {
2904 return true;
2905 }
2906 ciInstanceKlass* ik = klsptr->instance_klass();
2907 // don't need a guard for a klass that is already initialized
2908 return !ik->is_initialized();
2909 }
2910
2911 //----------------------------inline_unsafe_writeback0-------------------------
2912 // public native void Unsafe.writeback0(long address)
2913 bool LibraryCallKit::inline_unsafe_writeback0() {
2914 if (!Matcher::has_match_rule(Op_CacheWB)) {
2915 return false;
2916 }
2917 #ifndef PRODUCT
2918 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2919 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2920 ciSignature* sig = callee()->signature();
2921 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2922 #endif
2923 null_check_receiver(); // null-check, then ignore
2924 Node *addr = argument(1);
2925 addr = new CastX2PNode(addr);
2926 addr = _gvn.transform(addr);
2927 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2928 flush = _gvn.transform(flush);
2929 set_memory(flush, TypeRawPtr::BOTTOM);
2930 return true;
2931 }
2932
2933 //----------------------------inline_unsafe_writeback0-------------------------
2934 // public native void Unsafe.writeback0(long address)
2935 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2936 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2937 return false;
2938 }
2939 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2940 return false;
2941 }
2942 #ifndef PRODUCT
2943 assert(Matcher::has_match_rule(Op_CacheWB),
2944 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2945 : "found match rule for CacheWBPostSync but not CacheWB"));
2946
2947 #endif
2948 null_check_receiver(); // null-check, then ignore
2949 Node *sync;
2950 if (is_pre) {
2951 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2952 } else {
2953 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2954 }
2955 sync = _gvn.transform(sync);
2956 set_memory(sync, TypeRawPtr::BOTTOM);
2957 return true;
2958 }
2959
2960 //----------------------------inline_unsafe_allocate---------------------------
2961 // public native Object Unsafe.allocateInstance(Class<?> cls);
2962 bool LibraryCallKit::inline_unsafe_allocate() {
2963
2964 #if INCLUDE_JVMTI
2965 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2966 return false;
2967 }
2968 #endif //INCLUDE_JVMTI
2969
2970 if (callee()->is_static()) return false; // caller must have the capability!
2971
2972 null_check_receiver(); // null-check, then ignore
2973 Node* cls = null_check(argument(1));
2974 if (stopped()) return true;
2975
2976 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2977 kls = null_check(kls);
2978 if (stopped()) return true; // argument was like int.class
2979
2980 #if INCLUDE_JVMTI
2981 // Don't try to access new allocated obj in the intrinsic.
2982 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2983 // Deoptimize and allocate in interpreter instead.
2984 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2985 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2986 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2987 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2988 {
2989 BuildCutout unless(this, tst, PROB_MAX);
2990 uncommon_trap(Deoptimization::Reason_intrinsic,
2991 Deoptimization::Action_make_not_entrant);
2992 }
2993 if (stopped()) {
2994 return true;
2995 }
2996 #endif //INCLUDE_JVMTI
2997
2998 Node* test = nullptr;
2999 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3000 // Note: The argument might still be an illegal value like
3001 // Serializable.class or Object[].class. The runtime will handle it.
3002 // But we must make an explicit check for initialization.
3003 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3004 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3005 // can generate code to load it as unsigned byte.
3006 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3007 Node* bits = intcon(InstanceKlass::fully_initialized);
3008 test = _gvn.transform(new SubINode(inst, bits));
3009 // The 'test' is non-zero if we need to take a slow path.
3010 }
3011
3012 Node* obj = new_instance(kls, test);
3013 set_result(obj);
3014 return true;
3015 }
3016
3017 //------------------------inline_native_time_funcs--------------
3018 // inline code for System.currentTimeMillis() and System.nanoTime()
3019 // these have the same type and signature
3020 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3021 const TypeFunc* tf = OptoRuntime::void_long_Type();
3022 const TypePtr* no_memory_effects = nullptr;
3023 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3024 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3025 #ifdef ASSERT
3026 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3027 assert(value_top == top(), "second value must be top");
3028 #endif
3029 set_result(value);
3030 return true;
3031 }
3032
3033 //--------------------inline_native_vthread_start_transition--------------------
3034 // inline void startTransition(boolean is_mount);
3035 // inline void startFinalTransition();
3036 // Pseudocode of implementation:
3037 //
3038 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3039 // carrier->set_is_in_vthread_transition(true);
3040 // OrderAccess::storeload();
3041 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3042 // + global_vthread_transition_disable_count();
3043 // if (disable_requests > 0) {
3044 // slow path: runtime call
3045 // }
3046 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3047 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3048 IdealKit ideal(this);
3049
3050 Node* thread = ideal.thread();
3051 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3052 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3053 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3054 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3055 insert_mem_bar(Op_MemBarVolatile);
3056 ideal.sync_kit(this);
3057
3058 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3059 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3060 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3061 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3062 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3063
3064 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3065 sync_kit(ideal);
3066 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3067 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3068 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3069 ideal.sync_kit(this);
3070 }
3071 ideal.end_if();
3072
3073 final_sync(ideal);
3074 return true;
3075 }
3076
3077 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3078 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3079 IdealKit ideal(this);
3080
3081 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3082 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3083
3084 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3085 sync_kit(ideal);
3086 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3087 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3088 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3089 ideal.sync_kit(this);
3090 } ideal.else_(); {
3091 Node* thread = ideal.thread();
3092 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3093 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3094
3095 sync_kit(ideal);
3096 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3097 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3098 ideal.sync_kit(this);
3099 } ideal.end_if();
3100
3101 final_sync(ideal);
3102 return true;
3103 }
3104
3105 #if INCLUDE_JVMTI
3106
3107 // Always update the is_disable_suspend bit.
3108 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3109 if (!DoJVMTIVirtualThreadTransitions) {
3110 return true;
3111 }
3112 IdealKit ideal(this);
3113
3114 {
3115 // unconditionally update the is_disable_suspend bit in current JavaThread
3116 Node* thread = ideal.thread();
3117 Node* arg = _gvn.transform(argument(0)); // argument for notification
3118 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3119 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3120
3121 sync_kit(ideal);
3122 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3123 ideal.sync_kit(this);
3124 }
3125 final_sync(ideal);
3126
3127 return true;
3128 }
3129
3130 #endif // INCLUDE_JVMTI
3131
3132 #ifdef JFR_HAVE_INTRINSICS
3133
3134 /**
3135 * if oop->klass != null
3136 * // normal class
3137 * epoch = _epoch_state ? 2 : 1
3138 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3139 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3140 * }
3141 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3142 * else
3143 * // primitive class
3144 * if oop->array_klass != null
3145 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3146 * else
3147 * id = LAST_TYPE_ID + 1 // void class path
3148 * if (!signaled)
3149 * signaled = true
3150 */
3151 bool LibraryCallKit::inline_native_classID() {
3152 Node* cls = argument(0);
3153
3154 IdealKit ideal(this);
3155 #define __ ideal.
3156 IdealVariable result(ideal); __ declarations_done();
3157 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3158 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3159 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3160
3161
3162 __ if_then(kls, BoolTest::ne, null()); {
3163 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3164 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3165
3166 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3167 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3168 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3169 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3170 mask = _gvn.transform(new OrLNode(mask, epoch));
3171 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3172
3173 float unlikely = PROB_UNLIKELY(0.999);
3174 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3175 sync_kit(ideal);
3176 make_runtime_call(RC_LEAF,
3177 OptoRuntime::class_id_load_barrier_Type(),
3178 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3179 "class id load barrier",
3180 TypePtr::BOTTOM,
3181 kls);
3182 ideal.sync_kit(this);
3183 } __ end_if();
3184
3185 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3186 } __ else_(); {
3187 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3188 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3189 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3190 __ if_then(array_kls, BoolTest::ne, null()); {
3191 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3192 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3193 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3194 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3195 } __ else_(); {
3196 // void class case
3197 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3198 } __ end_if();
3199
3200 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3201 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3202 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3203 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3204 } __ end_if();
3205 } __ end_if();
3206
3207 final_sync(ideal);
3208 set_result(ideal.value(result));
3209 #undef __
3210 return true;
3211 }
3212
3213 //------------------------inline_native_jvm_commit------------------
3214 bool LibraryCallKit::inline_native_jvm_commit() {
3215 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3216
3217 // Save input memory and i_o state.
3218 Node* input_memory_state = reset_memory();
3219 set_all_memory(input_memory_state);
3220 Node* input_io_state = i_o();
3221
3222 // TLS.
3223 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3224 // Jfr java buffer.
3225 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3226 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3227 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3228
3229 // Load the current value of the notified field in the JfrThreadLocal.
3230 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3231 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3232
3233 // Test for notification.
3234 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3235 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3236 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3237
3238 // True branch, is notified.
3239 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3240 set_control(is_notified);
3241
3242 // Reset notified state.
3243 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3244 Node* notified_reset_memory = reset_memory();
3245
3246 // 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.
3247 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3248 // Convert the machine-word to a long.
3249 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3250
3251 // False branch, not notified.
3252 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3253 set_control(not_notified);
3254 set_all_memory(input_memory_state);
3255
3256 // Arg is the next position as a long.
3257 Node* arg = argument(0);
3258 // Convert long to machine-word.
3259 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3260
3261 // Store the next_position to the underlying jfr java buffer.
3262 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3263
3264 Node* commit_memory = reset_memory();
3265 set_all_memory(commit_memory);
3266
3267 // 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.
3268 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3269 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3270 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3271
3272 // And flags with lease constant.
3273 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3274
3275 // Branch on lease to conditionalize returning the leased java buffer.
3276 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3277 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3278 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3279
3280 // False branch, not a lease.
3281 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3282
3283 // True branch, is lease.
3284 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3285 set_control(is_lease);
3286
3287 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3288 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3289 OptoRuntime::void_void_Type(),
3290 SharedRuntime::jfr_return_lease(),
3291 "return_lease", TypePtr::BOTTOM);
3292 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3293
3294 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3295 record_for_igvn(lease_compare_rgn);
3296 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3297 record_for_igvn(lease_compare_mem);
3298 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3299 record_for_igvn(lease_compare_io);
3300 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3301 record_for_igvn(lease_result_value);
3302
3303 // Update control and phi nodes.
3304 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3305 lease_compare_rgn->init_req(_false_path, not_lease);
3306
3307 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3308 lease_compare_mem->init_req(_false_path, commit_memory);
3309
3310 lease_compare_io->init_req(_true_path, i_o());
3311 lease_compare_io->init_req(_false_path, input_io_state);
3312
3313 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3314 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3315
3316 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3317 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3318 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3319 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3320
3321 // Update control and phi nodes.
3322 result_rgn->init_req(_true_path, is_notified);
3323 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3324
3325 result_mem->init_req(_true_path, notified_reset_memory);
3326 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3327
3328 result_io->init_req(_true_path, input_io_state);
3329 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3330
3331 result_value->init_req(_true_path, current_pos);
3332 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3333
3334 // Set output state.
3335 set_control(_gvn.transform(result_rgn));
3336 set_all_memory(_gvn.transform(result_mem));
3337 set_i_o(_gvn.transform(result_io));
3338 set_result(result_rgn, result_value);
3339 return true;
3340 }
3341
3342 /*
3343 * The intrinsic is a model of this pseudo-code:
3344 *
3345 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3346 * jobject h_event_writer = tl->java_event_writer();
3347 * if (h_event_writer == nullptr) {
3348 * return nullptr;
3349 * }
3350 * oop threadObj = Thread::threadObj();
3351 * oop vthread = java_lang_Thread::vthread(threadObj);
3352 * traceid tid;
3353 * bool pinVirtualThread;
3354 * bool excluded;
3355 * if (vthread != threadObj) { // i.e. current thread is virtual
3356 * tid = java_lang_Thread::tid(vthread);
3357 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3358 * pinVirtualThread = VMContinuations;
3359 * excluded = vthread_epoch_raw & excluded_mask;
3360 * if (!excluded) {
3361 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3362 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3363 * if (vthread_epoch != current_epoch) {
3364 * write_checkpoint();
3365 * }
3366 * }
3367 * } else {
3368 * tid = java_lang_Thread::tid(threadObj);
3369 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3370 * pinVirtualThread = false;
3371 * excluded = thread_epoch_raw & excluded_mask;
3372 * }
3373 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3374 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3375 * if (tid_in_event_writer != tid) {
3376 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3377 * setField(event_writer, "excluded", excluded);
3378 * setField(event_writer, "threadID", tid);
3379 * }
3380 * return event_writer
3381 */
3382 bool LibraryCallKit::inline_native_getEventWriter() {
3383 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3384
3385 // Save input memory and i_o state.
3386 Node* input_memory_state = reset_memory();
3387 set_all_memory(input_memory_state);
3388 Node* input_io_state = i_o();
3389
3390 // The most significant bit of the u2 is used to denote thread exclusion
3391 Node* excluded_shift = _gvn.intcon(15);
3392 Node* excluded_mask = _gvn.intcon(1 << 15);
3393 // The epoch generation is the range [1-32767]
3394 Node* epoch_mask = _gvn.intcon(32767);
3395
3396 // TLS
3397 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3398
3399 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3400 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3401
3402 // Load the eventwriter jobject handle.
3403 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3404
3405 // Null check the jobject handle.
3406 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3407 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3408 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3409
3410 // False path, jobj is null.
3411 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3412
3413 // True path, jobj is not null.
3414 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3415
3416 set_control(jobj_is_not_null);
3417
3418 // Load the threadObj for the CarrierThread.
3419 Node* threadObj = generate_current_thread(tls_ptr);
3420
3421 // Load the vthread.
3422 Node* vthread = generate_virtual_thread(tls_ptr);
3423
3424 // If vthread != threadObj, this is a virtual thread.
3425 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3426 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3427 IfNode* iff_vthread_not_equal_threadObj =
3428 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3429
3430 // False branch, fallback to threadObj.
3431 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3432 set_control(vthread_equal_threadObj);
3433
3434 // Load the tid field from the vthread object.
3435 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3436
3437 // Load the raw epoch value from the threadObj.
3438 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3439 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3440 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3441 TypeInt::CHAR, T_CHAR,
3442 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3443
3444 // Mask off the excluded information from the epoch.
3445 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3446
3447 // True branch, this is a virtual thread.
3448 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3449 set_control(vthread_not_equal_threadObj);
3450
3451 // Load the tid field from the vthread object.
3452 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3453
3454 // Continuation support determines if a virtual thread should be pinned.
3455 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3456 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3457
3458 // Load the raw epoch value from the vthread.
3459 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3460 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3461 TypeInt::CHAR, T_CHAR,
3462 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3463
3464 // Mask off the excluded information from the epoch.
3465 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3466
3467 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3468 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3469 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3470 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3471
3472 // False branch, vthread is excluded, no need to write epoch info.
3473 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3474
3475 // True branch, vthread is included, update epoch info.
3476 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3477 set_control(included);
3478
3479 // Get epoch value.
3480 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3481
3482 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3483 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3484 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3485
3486 // Compare the epoch in the vthread to the current epoch generation.
3487 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3488 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3489 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3490
3491 // False path, epoch is equal, checkpoint information is valid.
3492 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3493
3494 // True path, epoch is not equal, write a checkpoint for the vthread.
3495 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3496
3497 set_control(epoch_is_not_equal);
3498
3499 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3500 // The call also updates the native thread local thread id and the vthread with the current epoch.
3501 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3502 OptoRuntime::jfr_write_checkpoint_Type(),
3503 SharedRuntime::jfr_write_checkpoint(),
3504 "write_checkpoint", TypePtr::BOTTOM);
3505 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3506
3507 // vthread epoch != current epoch
3508 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3509 record_for_igvn(epoch_compare_rgn);
3510 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3511 record_for_igvn(epoch_compare_mem);
3512 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3513 record_for_igvn(epoch_compare_io);
3514
3515 // Update control and phi nodes.
3516 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3517 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3518 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3519 epoch_compare_mem->init_req(_false_path, input_memory_state);
3520 epoch_compare_io->init_req(_true_path, i_o());
3521 epoch_compare_io->init_req(_false_path, input_io_state);
3522
3523 // excluded != true
3524 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3525 record_for_igvn(exclude_compare_rgn);
3526 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3527 record_for_igvn(exclude_compare_mem);
3528 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3529 record_for_igvn(exclude_compare_io);
3530
3531 // Update control and phi nodes.
3532 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3533 exclude_compare_rgn->init_req(_false_path, excluded);
3534 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3535 exclude_compare_mem->init_req(_false_path, input_memory_state);
3536 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3537 exclude_compare_io->init_req(_false_path, input_io_state);
3538
3539 // vthread != threadObj
3540 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3541 record_for_igvn(vthread_compare_rgn);
3542 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3543 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3544 record_for_igvn(vthread_compare_io);
3545 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3546 record_for_igvn(tid);
3547 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3548 record_for_igvn(exclusion);
3549 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3550 record_for_igvn(pinVirtualThread);
3551
3552 // Update control and phi nodes.
3553 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3554 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3555 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3556 vthread_compare_mem->init_req(_false_path, input_memory_state);
3557 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3558 vthread_compare_io->init_req(_false_path, input_io_state);
3559 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3560 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3561 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3562 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3563 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3564 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3565
3566 // Update branch state.
3567 set_control(_gvn.transform(vthread_compare_rgn));
3568 set_all_memory(_gvn.transform(vthread_compare_mem));
3569 set_i_o(_gvn.transform(vthread_compare_io));
3570
3571 // Load the event writer oop by dereferencing the jobject handle.
3572 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3573 assert(klass_EventWriter->is_loaded(), "invariant");
3574 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3575 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3576 const TypeOopPtr* const xtype = aklass->as_instance_type();
3577 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3578 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3579
3580 // Load the current thread id from the event writer object.
3581 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3582 // Get the field offset to, conditionally, store an updated tid value later.
3583 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3584 // Get the field offset to, conditionally, store an updated exclusion value later.
3585 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3586 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3587 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3588
3589 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3590 record_for_igvn(event_writer_tid_compare_rgn);
3591 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3592 record_for_igvn(event_writer_tid_compare_mem);
3593 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3594 record_for_igvn(event_writer_tid_compare_io);
3595
3596 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3597 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3598 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3599 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3600
3601 // False path, tids are the same.
3602 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3603
3604 // True path, tid is not equal, need to update the tid in the event writer.
3605 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3606 record_for_igvn(tid_is_not_equal);
3607
3608 // Store the pin state to the event writer.
3609 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3610
3611 // Store the exclusion state to the event writer.
3612 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3613 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3614
3615 // Store the tid to the event writer.
3616 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3617
3618 // Update control and phi nodes.
3619 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3620 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3621 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3622 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3623 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3624 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3625
3626 // Result of top level CFG, Memory, IO and Value.
3627 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3628 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3629 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3630 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3631
3632 // Result control.
3633 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3634 result_rgn->init_req(_false_path, jobj_is_null);
3635
3636 // Result memory.
3637 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3638 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3639
3640 // Result IO.
3641 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3642 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3643
3644 // Result value.
3645 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3646 result_value->init_req(_false_path, null()); // return null
3647
3648 // Set output state.
3649 set_control(_gvn.transform(result_rgn));
3650 set_all_memory(_gvn.transform(result_mem));
3651 set_i_o(_gvn.transform(result_io));
3652 set_result(result_rgn, result_value);
3653 return true;
3654 }
3655
3656 /*
3657 * The intrinsic is a model of this pseudo-code:
3658 *
3659 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3660 * if (carrierThread != thread) { // is virtual thread
3661 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3662 * bool excluded = vthread_epoch_raw & excluded_mask;
3663 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3664 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3665 * if (!excluded) {
3666 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3667 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3668 * }
3669 * AtomicAccess::release_store(&tl->_vthread, true);
3670 * return;
3671 * }
3672 * AtomicAccess::release_store(&tl->_vthread, false);
3673 */
3674 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3675 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3676
3677 Node* input_memory_state = reset_memory();
3678 set_all_memory(input_memory_state);
3679
3680 // The most significant bit of the u2 is used to denote thread exclusion
3681 Node* excluded_mask = _gvn.intcon(1 << 15);
3682 // The epoch generation is the range [1-32767]
3683 Node* epoch_mask = _gvn.intcon(32767);
3684
3685 Node* const carrierThread = generate_current_thread(jt);
3686 // If thread != carrierThread, this is a virtual thread.
3687 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3688 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3689 IfNode* iff_thread_not_equal_carrierThread =
3690 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3691
3692 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3693
3694 // False branch, is carrierThread.
3695 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3696 // Store release
3697 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3698
3699 set_all_memory(input_memory_state);
3700
3701 // True branch, is virtual thread.
3702 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3703 set_control(thread_not_equal_carrierThread);
3704
3705 // Load the raw epoch value from the vthread.
3706 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3707 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3708 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3709
3710 // Mask off the excluded information from the epoch.
3711 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3712
3713 // Load the tid field from the thread.
3714 Node* tid = load_field_from_object(thread, "tid", "J");
3715
3716 // Store the vthread tid to the jfr thread local.
3717 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3718 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3719
3720 // Branch is_excluded to conditionalize updating the epoch .
3721 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3722 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3723 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3724
3725 // True branch, vthread is excluded, no need to write epoch info.
3726 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3727 set_control(excluded);
3728 Node* vthread_is_excluded = _gvn.intcon(1);
3729
3730 // False branch, vthread is included, update epoch info.
3731 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3732 set_control(included);
3733 Node* vthread_is_included = _gvn.intcon(0);
3734
3735 // Get epoch value.
3736 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3737
3738 // Store the vthread epoch to the jfr thread local.
3739 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3740 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3741
3742 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3743 record_for_igvn(excluded_rgn);
3744 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3745 record_for_igvn(excluded_mem);
3746 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3747 record_for_igvn(exclusion);
3748
3749 // Merge the excluded control and memory.
3750 excluded_rgn->init_req(_true_path, excluded);
3751 excluded_rgn->init_req(_false_path, included);
3752 excluded_mem->init_req(_true_path, tid_memory);
3753 excluded_mem->init_req(_false_path, included_memory);
3754 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3755 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3756
3757 // Set intermediate state.
3758 set_control(_gvn.transform(excluded_rgn));
3759 set_all_memory(excluded_mem);
3760
3761 // Store the vthread exclusion state to the jfr thread local.
3762 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3763 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3764
3765 // Store release
3766 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3767
3768 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3769 record_for_igvn(thread_compare_rgn);
3770 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3771 record_for_igvn(thread_compare_mem);
3772 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3773 record_for_igvn(vthread);
3774
3775 // Merge the thread_compare control and memory.
3776 thread_compare_rgn->init_req(_true_path, control());
3777 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3778 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3779 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3780
3781 // Set output state.
3782 set_control(_gvn.transform(thread_compare_rgn));
3783 set_all_memory(_gvn.transform(thread_compare_mem));
3784 }
3785
3786 #endif // JFR_HAVE_INTRINSICS
3787
3788 //------------------------inline_native_currentCarrierThread------------------
3789 bool LibraryCallKit::inline_native_currentCarrierThread() {
3790 Node* junk = nullptr;
3791 set_result(generate_current_thread(junk));
3792 return true;
3793 }
3794
3795 //------------------------inline_native_currentThread------------------
3796 bool LibraryCallKit::inline_native_currentThread() {
3797 Node* junk = nullptr;
3798 set_result(generate_virtual_thread(junk));
3799 return true;
3800 }
3801
3802 //------------------------inline_native_setVthread------------------
3803 bool LibraryCallKit::inline_native_setCurrentThread() {
3804 assert(C->method()->changes_current_thread(),
3805 "method changes current Thread but is not annotated ChangesCurrentThread");
3806 Node* arr = argument(1);
3807 Node* thread = _gvn.transform(new ThreadLocalNode());
3808 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3809 Node* thread_obj_handle
3810 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3811 thread_obj_handle = _gvn.transform(thread_obj_handle);
3812 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3813 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3814
3815 // Change the _monitor_owner_id of the JavaThread
3816 Node* tid = load_field_from_object(arr, "tid", "J");
3817 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3818 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3819
3820 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3821 return true;
3822 }
3823
3824 const Type* LibraryCallKit::scopedValueCache_type() {
3825 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3826 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3827 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3828
3829 // Because we create the scopedValue cache lazily we have to make the
3830 // type of the result BotPTR.
3831 bool xk = etype->klass_is_exact();
3832 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3833 return objects_type;
3834 }
3835
3836 Node* LibraryCallKit::scopedValueCache_helper() {
3837 Node* thread = _gvn.transform(new ThreadLocalNode());
3838 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3839 // We cannot use immutable_memory() because we might flip onto a
3840 // different carrier thread, at which point we'll need to use that
3841 // carrier thread's cache.
3842 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3843 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3844 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3845 }
3846
3847 //------------------------inline_native_scopedValueCache------------------
3848 bool LibraryCallKit::inline_native_scopedValueCache() {
3849 Node* cache_obj_handle = scopedValueCache_helper();
3850 const Type* objects_type = scopedValueCache_type();
3851 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3852
3853 return true;
3854 }
3855
3856 //------------------------inline_native_setScopedValueCache------------------
3857 bool LibraryCallKit::inline_native_setScopedValueCache() {
3858 Node* arr = argument(0);
3859 Node* cache_obj_handle = scopedValueCache_helper();
3860 const Type* objects_type = scopedValueCache_type();
3861
3862 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3863 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3864
3865 return true;
3866 }
3867
3868 //------------------------inline_native_Continuation_pin and unpin-----------
3869
3870 // Shared implementation routine for both pin and unpin.
3871 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3872 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3873
3874 // Save input memory.
3875 Node* input_memory_state = reset_memory();
3876 set_all_memory(input_memory_state);
3877
3878 // TLS
3879 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3880 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3881 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3882
3883 // Null check the last continuation object.
3884 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3885 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3886 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3887
3888 // False path, last continuation is null.
3889 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3890
3891 // True path, last continuation is not null.
3892 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3893
3894 set_control(continuation_is_not_null);
3895
3896 // Load the pin count from the last continuation.
3897 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3898 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3899
3900 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3901 Node* pin_count_rhs;
3902 if (unpin) {
3903 pin_count_rhs = _gvn.intcon(0);
3904 } else {
3905 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3906 }
3907 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3908 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3909 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3910
3911 // True branch, pin count over/underflow.
3912 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3913 {
3914 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3915 // which will throw IllegalStateException for pin count over/underflow.
3916 // No memory changed so far - we can use memory create by reset_memory()
3917 // at the beginning of this intrinsic. No need to call reset_memory() again.
3918 PreserveJVMState pjvms(this);
3919 set_control(pin_count_over_underflow);
3920 uncommon_trap(Deoptimization::Reason_intrinsic,
3921 Deoptimization::Action_none);
3922 assert(stopped(), "invariant");
3923 }
3924
3925 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3926 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3927 set_control(valid_pin_count);
3928
3929 Node* next_pin_count;
3930 if (unpin) {
3931 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3932 } else {
3933 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3934 }
3935
3936 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3937
3938 // Result of top level CFG and Memory.
3939 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3940 record_for_igvn(result_rgn);
3941 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3942 record_for_igvn(result_mem);
3943
3944 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3945 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3946 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3947 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3948
3949 // Set output state.
3950 set_control(_gvn.transform(result_rgn));
3951 set_all_memory(_gvn.transform(result_mem));
3952
3953 return true;
3954 }
3955
3956 //---------------------------load_mirror_from_klass----------------------------
3957 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3958 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3959 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3960 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3961 // mirror = ((OopHandle)mirror)->resolve();
3962 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3963 }
3964
3965 //-----------------------load_klass_from_mirror_common-------------------------
3966 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3967 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3968 // and branch to the given path on the region.
3969 // If never_see_null, take an uncommon trap on null, so we can optimistically
3970 // compile for the non-null case.
3971 // If the region is null, force never_see_null = true.
3972 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3973 bool never_see_null,
3974 RegionNode* region,
3975 int null_path,
3976 int offset) {
3977 if (region == nullptr) never_see_null = true;
3978 Node* p = basic_plus_adr(mirror, offset);
3979 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3980 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3981 Node* null_ctl = top();
3982 kls = null_check_oop(kls, &null_ctl, never_see_null);
3983 if (region != nullptr) {
3984 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3985 region->init_req(null_path, null_ctl);
3986 } else {
3987 assert(null_ctl == top(), "no loose ends");
3988 }
3989 return kls;
3990 }
3991
3992 //--------------------(inline_native_Class_query helpers)---------------------
3993 // Use this for JVM_ACC_INTERFACE.
3994 // Fall through if (mods & mask) == bits, take the guard otherwise.
3995 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3996 ByteSize offset, const Type* type, BasicType bt) {
3997 // Branch around if the given klass has the given modifier bit set.
3998 // Like generate_guard, adds a new path onto the region.
3999 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4000 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4001 Node* mask = intcon(modifier_mask);
4002 Node* bits = intcon(modifier_bits);
4003 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4004 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4005 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4006 return generate_fair_guard(bol, region);
4007 }
4008 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4009 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4010 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4011 }
4012
4013 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4014 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4015 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4016 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4017 }
4018
4019 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4020 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4021 }
4022
4023 //-------------------------inline_native_Class_query-------------------
4024 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4025 const Type* return_type = TypeInt::BOOL;
4026 Node* prim_return_value = top(); // what happens if it's a primitive class?
4027 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4028 bool expect_prim = false; // most of these guys expect to work on refs
4029
4030 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4031
4032 Node* mirror = argument(0);
4033 Node* obj = top();
4034
4035 switch (id) {
4036 case vmIntrinsics::_isInstance:
4037 // nothing is an instance of a primitive type
4038 prim_return_value = intcon(0);
4039 obj = argument(1);
4040 break;
4041 case vmIntrinsics::_isHidden:
4042 prim_return_value = intcon(0);
4043 break;
4044 case vmIntrinsics::_getSuperclass:
4045 prim_return_value = null();
4046 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4047 break;
4048 default:
4049 fatal_unexpected_iid(id);
4050 break;
4051 }
4052
4053 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4054 if (mirror_con == nullptr) return false; // cannot happen?
4055
4056 #ifndef PRODUCT
4057 if (C->print_intrinsics() || C->print_inlining()) {
4058 ciType* k = mirror_con->java_mirror_type();
4059 if (k) {
4060 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4061 k->print_name();
4062 tty->cr();
4063 }
4064 }
4065 #endif
4066
4067 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4068 RegionNode* region = new RegionNode(PATH_LIMIT);
4069 record_for_igvn(region);
4070 PhiNode* phi = new PhiNode(region, return_type);
4071
4072 // The mirror will never be null of Reflection.getClassAccessFlags, however
4073 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4074 // if it is. See bug 4774291.
4075
4076 // For Reflection.getClassAccessFlags(), the null check occurs in
4077 // the wrong place; see inline_unsafe_access(), above, for a similar
4078 // situation.
4079 mirror = null_check(mirror);
4080 // If mirror or obj is dead, only null-path is taken.
4081 if (stopped()) return true;
4082
4083 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4084
4085 // Now load the mirror's klass metaobject, and null-check it.
4086 // Side-effects region with the control path if the klass is null.
4087 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4088 // If kls is null, we have a primitive mirror.
4089 phi->init_req(_prim_path, prim_return_value);
4090 if (stopped()) { set_result(region, phi); return true; }
4091 bool safe_for_replace = (region->in(_prim_path) == top());
4092
4093 Node* p; // handy temp
4094 Node* null_ctl;
4095
4096 // Now that we have the non-null klass, we can perform the real query.
4097 // For constant classes, the query will constant-fold in LoadNode::Value.
4098 Node* query_value = top();
4099 switch (id) {
4100 case vmIntrinsics::_isInstance:
4101 // nothing is an instance of a primitive type
4102 query_value = gen_instanceof(obj, kls, safe_for_replace);
4103 break;
4104
4105 case vmIntrinsics::_isHidden:
4106 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4107 if (generate_hidden_class_guard(kls, region) != nullptr)
4108 // A guard was added. If the guard is taken, it was an hidden class.
4109 phi->add_req(intcon(1));
4110 // If we fall through, it's a plain class.
4111 query_value = intcon(0);
4112 break;
4113
4114
4115 case vmIntrinsics::_getSuperclass:
4116 // The rules here are somewhat unfortunate, but we can still do better
4117 // with random logic than with a JNI call.
4118 // Interfaces store null or Object as _super, but must report null.
4119 // Arrays store an intermediate super as _super, but must report Object.
4120 // Other types can report the actual _super.
4121 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4122 if (generate_array_guard(kls, region) != nullptr) {
4123 // A guard was added. If the guard is taken, it was an array.
4124 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4125 }
4126 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4127 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4128 if (generate_interface_guard(kls, region) != nullptr) {
4129 // A guard was added. If the guard is taken, it was an interface.
4130 phi->add_req(null());
4131 }
4132 // If we fall through, it's a plain class. Get its _super.
4133 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4134 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4135 null_ctl = top();
4136 kls = null_check_oop(kls, &null_ctl);
4137 if (null_ctl != top()) {
4138 // If the guard is taken, Object.superClass is null (both klass and mirror).
4139 region->add_req(null_ctl);
4140 phi ->add_req(null());
4141 }
4142 if (!stopped()) {
4143 query_value = load_mirror_from_klass(kls);
4144 }
4145 break;
4146
4147 default:
4148 fatal_unexpected_iid(id);
4149 break;
4150 }
4151
4152 // Fall-through is the normal case of a query to a real class.
4153 phi->init_req(1, query_value);
4154 region->init_req(1, control());
4155
4156 C->set_has_split_ifs(true); // Has chance for split-if optimization
4157 set_result(region, phi);
4158 return true;
4159 }
4160
4161 //-------------------------inline_Class_cast-------------------
4162 bool LibraryCallKit::inline_Class_cast() {
4163 Node* mirror = argument(0); // Class
4164 Node* obj = argument(1);
4165 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4166 if (mirror_con == nullptr) {
4167 return false; // dead path (mirror->is_top()).
4168 }
4169 if (obj == nullptr || obj->is_top()) {
4170 return false; // dead path
4171 }
4172 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4173
4174 // First, see if Class.cast() can be folded statically.
4175 // java_mirror_type() returns non-null for compile-time Class constants.
4176 ciType* tm = mirror_con->java_mirror_type();
4177 if (tm != nullptr && tm->is_klass() &&
4178 tp != nullptr) {
4179 if (!tp->is_loaded()) {
4180 // Don't use intrinsic when class is not loaded.
4181 return false;
4182 } else {
4183 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4184 if (static_res == Compile::SSC_always_true) {
4185 // isInstance() is true - fold the code.
4186 set_result(obj);
4187 return true;
4188 } else if (static_res == Compile::SSC_always_false) {
4189 // Don't use intrinsic, have to throw ClassCastException.
4190 // If the reference is null, the non-intrinsic bytecode will
4191 // be optimized appropriately.
4192 return false;
4193 }
4194 }
4195 }
4196
4197 // Bailout intrinsic and do normal inlining if exception path is frequent.
4198 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4199 return false;
4200 }
4201
4202 // Generate dynamic checks.
4203 // Class.cast() is java implementation of _checkcast bytecode.
4204 // Do checkcast (Parse::do_checkcast()) optimizations here.
4205
4206 mirror = null_check(mirror);
4207 // If mirror is dead, only null-path is taken.
4208 if (stopped()) {
4209 return true;
4210 }
4211
4212 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4213 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4214 RegionNode* region = new RegionNode(PATH_LIMIT);
4215 record_for_igvn(region);
4216
4217 // Now load the mirror's klass metaobject, and null-check it.
4218 // If kls is null, we have a primitive mirror and
4219 // nothing is an instance of a primitive type.
4220 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4221
4222 Node* res = top();
4223 if (!stopped()) {
4224 Node* bad_type_ctrl = top();
4225 // Do checkcast optimizations.
4226 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4227 region->init_req(_bad_type_path, bad_type_ctrl);
4228 }
4229 if (region->in(_prim_path) != top() ||
4230 region->in(_bad_type_path) != top()) {
4231 // Let Interpreter throw ClassCastException.
4232 PreserveJVMState pjvms(this);
4233 set_control(_gvn.transform(region));
4234 uncommon_trap(Deoptimization::Reason_intrinsic,
4235 Deoptimization::Action_maybe_recompile);
4236 }
4237 if (!stopped()) {
4238 set_result(res);
4239 }
4240 return true;
4241 }
4242
4243
4244 //--------------------------inline_native_subtype_check------------------------
4245 // This intrinsic takes the JNI calls out of the heart of
4246 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4247 bool LibraryCallKit::inline_native_subtype_check() {
4248 // Pull both arguments off the stack.
4249 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4250 args[0] = argument(0);
4251 args[1] = argument(1);
4252 Node* klasses[2]; // corresponding Klasses: superk, subk
4253 klasses[0] = klasses[1] = top();
4254
4255 enum {
4256 // A full decision tree on {superc is prim, subc is prim}:
4257 _prim_0_path = 1, // {P,N} => false
4258 // {P,P} & superc!=subc => false
4259 _prim_same_path, // {P,P} & superc==subc => true
4260 _prim_1_path, // {N,P} => false
4261 _ref_subtype_path, // {N,N} & subtype check wins => true
4262 _both_ref_path, // {N,N} & subtype check loses => false
4263 PATH_LIMIT
4264 };
4265
4266 RegionNode* region = new RegionNode(PATH_LIMIT);
4267 Node* phi = new PhiNode(region, TypeInt::BOOL);
4268 record_for_igvn(region);
4269
4270 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4271 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4272 int class_klass_offset = java_lang_Class::klass_offset();
4273
4274 // First null-check both mirrors and load each mirror's klass metaobject.
4275 int which_arg;
4276 for (which_arg = 0; which_arg <= 1; which_arg++) {
4277 Node* arg = args[which_arg];
4278 arg = null_check(arg);
4279 if (stopped()) break;
4280 args[which_arg] = arg;
4281
4282 Node* p = basic_plus_adr(arg, class_klass_offset);
4283 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4284 klasses[which_arg] = _gvn.transform(kls);
4285 }
4286
4287 // Having loaded both klasses, test each for null.
4288 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4289 for (which_arg = 0; which_arg <= 1; which_arg++) {
4290 Node* kls = klasses[which_arg];
4291 Node* null_ctl = top();
4292 kls = null_check_oop(kls, &null_ctl, never_see_null);
4293 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4294 region->init_req(prim_path, null_ctl);
4295 if (stopped()) break;
4296 klasses[which_arg] = kls;
4297 }
4298
4299 if (!stopped()) {
4300 // now we have two reference types, in klasses[0..1]
4301 Node* subk = klasses[1]; // the argument to isAssignableFrom
4302 Node* superk = klasses[0]; // the receiver
4303 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4304 // now we have a successful reference subtype check
4305 region->set_req(_ref_subtype_path, control());
4306 }
4307
4308 // If both operands are primitive (both klasses null), then
4309 // we must return true when they are identical primitives.
4310 // It is convenient to test this after the first null klass check.
4311 set_control(region->in(_prim_0_path)); // go back to first null check
4312 if (!stopped()) {
4313 // Since superc is primitive, make a guard for the superc==subc case.
4314 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4315 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4316 generate_guard(bol_eq, region, PROB_FAIR);
4317 if (region->req() == PATH_LIMIT+1) {
4318 // A guard was added. If the added guard is taken, superc==subc.
4319 region->swap_edges(PATH_LIMIT, _prim_same_path);
4320 region->del_req(PATH_LIMIT);
4321 }
4322 region->set_req(_prim_0_path, control()); // Not equal after all.
4323 }
4324
4325 // these are the only paths that produce 'true':
4326 phi->set_req(_prim_same_path, intcon(1));
4327 phi->set_req(_ref_subtype_path, intcon(1));
4328
4329 // pull together the cases:
4330 assert(region->req() == PATH_LIMIT, "sane region");
4331 for (uint i = 1; i < region->req(); i++) {
4332 Node* ctl = region->in(i);
4333 if (ctl == nullptr || ctl == top()) {
4334 region->set_req(i, top());
4335 phi ->set_req(i, top());
4336 } else if (phi->in(i) == nullptr) {
4337 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4338 }
4339 }
4340
4341 set_control(_gvn.transform(region));
4342 set_result(_gvn.transform(phi));
4343 return true;
4344 }
4345
4346 //---------------------generate_array_guard_common------------------------
4347 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4348 bool obj_array, bool not_array, Node** obj) {
4349
4350 if (stopped()) {
4351 return nullptr;
4352 }
4353
4354 // If obj_array/non_array==false/false:
4355 // Branch around if the given klass is in fact an array (either obj or prim).
4356 // If obj_array/non_array==false/true:
4357 // Branch around if the given klass is not an array klass of any kind.
4358 // If obj_array/non_array==true/true:
4359 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4360 // If obj_array/non_array==true/false:
4361 // Branch around if the kls is an oop array (Object[] or subtype)
4362 //
4363 // Like generate_guard, adds a new path onto the region.
4364 jint layout_con = 0;
4365 Node* layout_val = get_layout_helper(kls, layout_con);
4366 if (layout_val == nullptr) {
4367 bool query = (obj_array
4368 ? Klass::layout_helper_is_objArray(layout_con)
4369 : Klass::layout_helper_is_array(layout_con));
4370 if (query == not_array) {
4371 return nullptr; // never a branch
4372 } else { // always a branch
4373 Node* always_branch = control();
4374 if (region != nullptr)
4375 region->add_req(always_branch);
4376 set_control(top());
4377 return always_branch;
4378 }
4379 }
4380 // Now test the correct condition.
4381 jint nval = (obj_array
4382 ? (jint)(Klass::_lh_array_tag_type_value
4383 << Klass::_lh_array_tag_shift)
4384 : Klass::_lh_neutral_value);
4385 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4386 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4387 // invert the test if we are looking for a non-array
4388 if (not_array) btest = BoolTest(btest).negate();
4389 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4390 Node* ctrl = generate_fair_guard(bol, region);
4391 Node* is_array_ctrl = not_array ? control() : ctrl;
4392 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4393 // Keep track of the fact that 'obj' is an array to prevent
4394 // array specific accesses from floating above the guard.
4395 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4396 }
4397 return ctrl;
4398 }
4399
4400
4401 //-----------------------inline_native_newArray--------------------------
4402 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4403 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4404 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4405 Node* mirror;
4406 Node* count_val;
4407 if (uninitialized) {
4408 null_check_receiver();
4409 mirror = argument(1);
4410 count_val = argument(2);
4411 } else {
4412 mirror = argument(0);
4413 count_val = argument(1);
4414 }
4415
4416 mirror = null_check(mirror);
4417 // If mirror or obj is dead, only null-path is taken.
4418 if (stopped()) return true;
4419
4420 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4421 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4422 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4423 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4424 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4425
4426 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4427 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4428 result_reg, _slow_path);
4429 Node* normal_ctl = control();
4430 Node* no_array_ctl = result_reg->in(_slow_path);
4431
4432 // Generate code for the slow case. We make a call to newArray().
4433 set_control(no_array_ctl);
4434 if (!stopped()) {
4435 // Either the input type is void.class, or else the
4436 // array klass has not yet been cached. Either the
4437 // ensuing call will throw an exception, or else it
4438 // will cache the array klass for next time.
4439 PreserveJVMState pjvms(this);
4440 CallJavaNode* slow_call = nullptr;
4441 if (uninitialized) {
4442 // Generate optimized virtual call (holder class 'Unsafe' is final)
4443 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4444 } else {
4445 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4446 }
4447 Node* slow_result = set_results_for_java_call(slow_call);
4448 // this->control() comes from set_results_for_java_call
4449 result_reg->set_req(_slow_path, control());
4450 result_val->set_req(_slow_path, slow_result);
4451 result_io ->set_req(_slow_path, i_o());
4452 result_mem->set_req(_slow_path, reset_memory());
4453 }
4454
4455 set_control(normal_ctl);
4456 if (!stopped()) {
4457 // Normal case: The array type has been cached in the java.lang.Class.
4458 // The following call works fine even if the array type is polymorphic.
4459 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4460 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4461 result_reg->init_req(_normal_path, control());
4462 result_val->init_req(_normal_path, obj);
4463 result_io ->init_req(_normal_path, i_o());
4464 result_mem->init_req(_normal_path, reset_memory());
4465
4466 if (uninitialized) {
4467 // Mark the allocation so that zeroing is skipped
4468 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4469 alloc->maybe_set_complete(&_gvn);
4470 }
4471 }
4472
4473 // Return the combined state.
4474 set_i_o( _gvn.transform(result_io) );
4475 set_all_memory( _gvn.transform(result_mem));
4476
4477 C->set_has_split_ifs(true); // Has chance for split-if optimization
4478 set_result(result_reg, result_val);
4479 return true;
4480 }
4481
4482 //----------------------inline_native_getLength--------------------------
4483 // public static native int java.lang.reflect.Array.getLength(Object array);
4484 bool LibraryCallKit::inline_native_getLength() {
4485 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4486
4487 Node* array = null_check(argument(0));
4488 // If array is dead, only null-path is taken.
4489 if (stopped()) return true;
4490
4491 // Deoptimize if it is a non-array.
4492 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4493
4494 if (non_array != nullptr) {
4495 PreserveJVMState pjvms(this);
4496 set_control(non_array);
4497 uncommon_trap(Deoptimization::Reason_intrinsic,
4498 Deoptimization::Action_maybe_recompile);
4499 }
4500
4501 // If control is dead, only non-array-path is taken.
4502 if (stopped()) return true;
4503
4504 // The works fine even if the array type is polymorphic.
4505 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4506 Node* result = load_array_length(array);
4507
4508 C->set_has_split_ifs(true); // Has chance for split-if optimization
4509 set_result(result);
4510 return true;
4511 }
4512
4513 //------------------------inline_array_copyOf----------------------------
4514 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4515 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4516 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4517 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4518
4519 // Get the arguments.
4520 Node* original = argument(0);
4521 Node* start = is_copyOfRange? argument(1): intcon(0);
4522 Node* end = is_copyOfRange? argument(2): argument(1);
4523 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4524
4525 Node* newcopy = nullptr;
4526
4527 // Set the original stack and the reexecute bit for the interpreter to reexecute
4528 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4529 { PreserveReexecuteState preexecs(this);
4530 jvms()->set_should_reexecute(true);
4531
4532 array_type_mirror = null_check(array_type_mirror);
4533 original = null_check(original);
4534
4535 // Check if a null path was taken unconditionally.
4536 if (stopped()) return true;
4537
4538 Node* orig_length = load_array_length(original);
4539
4540 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4541 klass_node = null_check(klass_node);
4542
4543 RegionNode* bailout = new RegionNode(1);
4544 record_for_igvn(bailout);
4545
4546 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4547 // Bail out if that is so.
4548 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4549 if (not_objArray != nullptr) {
4550 // Improve the klass node's type from the new optimistic assumption:
4551 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4552 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4553 Node* cast = new CastPPNode(control(), klass_node, akls);
4554 klass_node = _gvn.transform(cast);
4555 }
4556
4557 // Bail out if either start or end is negative.
4558 generate_negative_guard(start, bailout, &start);
4559 generate_negative_guard(end, bailout, &end);
4560
4561 Node* length = end;
4562 if (_gvn.type(start) != TypeInt::ZERO) {
4563 length = _gvn.transform(new SubINode(end, start));
4564 }
4565
4566 // Bail out if length is negative (i.e., if start > end).
4567 // Without this the new_array would throw
4568 // NegativeArraySizeException but IllegalArgumentException is what
4569 // should be thrown
4570 generate_negative_guard(length, bailout, &length);
4571
4572 // Bail out if start is larger than the original length
4573 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4574 generate_negative_guard(orig_tail, bailout, &orig_tail);
4575
4576 if (bailout->req() > 1) {
4577 PreserveJVMState pjvms(this);
4578 set_control(_gvn.transform(bailout));
4579 uncommon_trap(Deoptimization::Reason_intrinsic,
4580 Deoptimization::Action_maybe_recompile);
4581 }
4582
4583 if (!stopped()) {
4584 // How many elements will we copy from the original?
4585 // The answer is MinI(orig_tail, length).
4586 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4587
4588 // Generate a direct call to the right arraycopy function(s).
4589 // We know the copy is disjoint but we might not know if the
4590 // oop stores need checking.
4591 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4592 // This will fail a store-check if x contains any non-nulls.
4593
4594 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4595 // loads/stores but it is legal only if we're sure the
4596 // Arrays.copyOf would succeed. So we need all input arguments
4597 // to the copyOf to be validated, including that the copy to the
4598 // new array won't trigger an ArrayStoreException. That subtype
4599 // check can be optimized if we know something on the type of
4600 // the input array from type speculation.
4601 if (_gvn.type(klass_node)->singleton()) {
4602 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4603 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4604
4605 int test = C->static_subtype_check(superk, subk);
4606 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4607 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4608 if (t_original->speculative_type() != nullptr) {
4609 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4610 }
4611 }
4612 }
4613
4614 bool validated = false;
4615 // Reason_class_check rather than Reason_intrinsic because we
4616 // want to intrinsify even if this traps.
4617 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4618 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4619
4620 if (not_subtype_ctrl != top()) {
4621 PreserveJVMState pjvms(this);
4622 set_control(not_subtype_ctrl);
4623 uncommon_trap(Deoptimization::Reason_class_check,
4624 Deoptimization::Action_make_not_entrant);
4625 assert(stopped(), "Should be stopped");
4626 }
4627 validated = true;
4628 }
4629
4630 if (!stopped()) {
4631 newcopy = new_array(klass_node, length, 0); // no arguments to push
4632
4633 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4634 load_object_klass(original), klass_node);
4635 if (!is_copyOfRange) {
4636 ac->set_copyof(validated);
4637 } else {
4638 ac->set_copyofrange(validated);
4639 }
4640 Node* n = _gvn.transform(ac);
4641 if (n == ac) {
4642 ac->connect_outputs(this);
4643 } else {
4644 assert(validated, "shouldn't transform if all arguments not validated");
4645 set_all_memory(n);
4646 }
4647 }
4648 }
4649 } // original reexecute is set back here
4650
4651 C->set_has_split_ifs(true); // Has chance for split-if optimization
4652 if (!stopped()) {
4653 set_result(newcopy);
4654 }
4655 return true;
4656 }
4657
4658
4659 //----------------------generate_virtual_guard---------------------------
4660 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4661 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4662 RegionNode* slow_region) {
4663 ciMethod* method = callee();
4664 int vtable_index = method->vtable_index();
4665 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4666 "bad index %d", vtable_index);
4667 // Get the Method* out of the appropriate vtable entry.
4668 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4669 vtable_index*vtableEntry::size_in_bytes() +
4670 in_bytes(vtableEntry::method_offset());
4671 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4672 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4673
4674 // Compare the target method with the expected method (e.g., Object.hashCode).
4675 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4676
4677 Node* native_call = makecon(native_call_addr);
4678 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4679 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4680
4681 return generate_slow_guard(test_native, slow_region);
4682 }
4683
4684 //-----------------------generate_method_call----------------------------
4685 // Use generate_method_call to make a slow-call to the real
4686 // method if the fast path fails. An alternative would be to
4687 // use a stub like OptoRuntime::slow_arraycopy_Java.
4688 // This only works for expanding the current library call,
4689 // not another intrinsic. (E.g., don't use this for making an
4690 // arraycopy call inside of the copyOf intrinsic.)
4691 CallJavaNode*
4692 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4693 // When compiling the intrinsic method itself, do not use this technique.
4694 guarantee(callee() != C->method(), "cannot make slow-call to self");
4695
4696 ciMethod* method = callee();
4697 // ensure the JVMS we have will be correct for this call
4698 guarantee(method_id == method->intrinsic_id(), "must match");
4699
4700 const TypeFunc* tf = TypeFunc::make(method);
4701 if (res_not_null) {
4702 assert(tf->return_type() == T_OBJECT, "");
4703 const TypeTuple* range = tf->range();
4704 const Type** fields = TypeTuple::fields(range->cnt());
4705 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4706 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4707 tf = TypeFunc::make(tf->domain(), new_range);
4708 }
4709 CallJavaNode* slow_call;
4710 if (is_static) {
4711 assert(!is_virtual, "");
4712 slow_call = new CallStaticJavaNode(C, tf,
4713 SharedRuntime::get_resolve_static_call_stub(), method);
4714 } else if (is_virtual) {
4715 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4716 int vtable_index = Method::invalid_vtable_index;
4717 if (UseInlineCaches) {
4718 // Suppress the vtable call
4719 } else {
4720 // hashCode and clone are not a miranda methods,
4721 // so the vtable index is fixed.
4722 // No need to use the linkResolver to get it.
4723 vtable_index = method->vtable_index();
4724 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4725 "bad index %d", vtable_index);
4726 }
4727 slow_call = new CallDynamicJavaNode(tf,
4728 SharedRuntime::get_resolve_virtual_call_stub(),
4729 method, vtable_index);
4730 } else { // neither virtual nor static: opt_virtual
4731 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4732 slow_call = new CallStaticJavaNode(C, tf,
4733 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4734 slow_call->set_optimized_virtual(true);
4735 }
4736 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4737 // To be able to issue a direct call (optimized virtual or virtual)
4738 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4739 // about the method being invoked should be attached to the call site to
4740 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4741 slow_call->set_override_symbolic_info(true);
4742 }
4743 set_arguments_for_java_call(slow_call);
4744 set_edges_for_java_call(slow_call);
4745 return slow_call;
4746 }
4747
4748
4749 /**
4750 * Build special case code for calls to hashCode on an object. This call may
4751 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4752 * slightly different code.
4753 */
4754 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4755 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4756 assert(!(is_virtual && is_static), "either virtual, special, or static");
4757
4758 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4759
4760 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4761 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4762 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4763 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4764 Node* obj = nullptr;
4765 if (!is_static) {
4766 // Check for hashing null object
4767 obj = null_check_receiver();
4768 if (stopped()) return true; // unconditionally null
4769 result_reg->init_req(_null_path, top());
4770 result_val->init_req(_null_path, top());
4771 } else {
4772 // Do a null check, and return zero if null.
4773 // System.identityHashCode(null) == 0
4774 obj = argument(0);
4775 Node* null_ctl = top();
4776 obj = null_check_oop(obj, &null_ctl);
4777 result_reg->init_req(_null_path, null_ctl);
4778 result_val->init_req(_null_path, _gvn.intcon(0));
4779 }
4780
4781 // Unconditionally null? Then return right away.
4782 if (stopped()) {
4783 set_control( result_reg->in(_null_path));
4784 if (!stopped())
4785 set_result(result_val->in(_null_path));
4786 return true;
4787 }
4788
4789 // We only go to the fast case code if we pass a number of guards. The
4790 // paths which do not pass are accumulated in the slow_region.
4791 RegionNode* slow_region = new RegionNode(1);
4792 record_for_igvn(slow_region);
4793
4794 // If this is a virtual call, we generate a funny guard. We pull out
4795 // the vtable entry corresponding to hashCode() from the target object.
4796 // If the target method which we are calling happens to be the native
4797 // Object hashCode() method, we pass the guard. We do not need this
4798 // guard for non-virtual calls -- the caller is known to be the native
4799 // Object hashCode().
4800 if (is_virtual) {
4801 // After null check, get the object's klass.
4802 Node* obj_klass = load_object_klass(obj);
4803 generate_virtual_guard(obj_klass, slow_region);
4804 }
4805
4806 // Get the header out of the object, use LoadMarkNode when available
4807 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4808 // The control of the load must be null. Otherwise, the load can move before
4809 // the null check after castPP removal.
4810 Node* no_ctrl = nullptr;
4811 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4812
4813 if (!UseObjectMonitorTable) {
4814 // Test the header to see if it is safe to read w.r.t. locking.
4815 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4816 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4817 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4818 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4819 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4820
4821 generate_slow_guard(test_monitor, slow_region);
4822 }
4823
4824 // Get the hash value and check to see that it has been properly assigned.
4825 // We depend on hash_mask being at most 32 bits and avoid the use of
4826 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4827 // vm: see markWord.hpp.
4828 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4829 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4830 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4831 // This hack lets the hash bits live anywhere in the mark object now, as long
4832 // as the shift drops the relevant bits into the low 32 bits. Note that
4833 // Java spec says that HashCode is an int so there's no point in capturing
4834 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4835 hshifted_header = ConvX2I(hshifted_header);
4836 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4837
4838 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4839 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4840 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4841
4842 generate_slow_guard(test_assigned, slow_region);
4843
4844 Node* init_mem = reset_memory();
4845 // fill in the rest of the null path:
4846 result_io ->init_req(_null_path, i_o());
4847 result_mem->init_req(_null_path, init_mem);
4848
4849 result_val->init_req(_fast_path, hash_val);
4850 result_reg->init_req(_fast_path, control());
4851 result_io ->init_req(_fast_path, i_o());
4852 result_mem->init_req(_fast_path, init_mem);
4853
4854 // Generate code for the slow case. We make a call to hashCode().
4855 set_control(_gvn.transform(slow_region));
4856 if (!stopped()) {
4857 // No need for PreserveJVMState, because we're using up the present state.
4858 set_all_memory(init_mem);
4859 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4860 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4861 Node* slow_result = set_results_for_java_call(slow_call);
4862 // this->control() comes from set_results_for_java_call
4863 result_reg->init_req(_slow_path, control());
4864 result_val->init_req(_slow_path, slow_result);
4865 result_io ->set_req(_slow_path, i_o());
4866 result_mem ->set_req(_slow_path, reset_memory());
4867 }
4868
4869 // Return the combined state.
4870 set_i_o( _gvn.transform(result_io) );
4871 set_all_memory( _gvn.transform(result_mem));
4872
4873 set_result(result_reg, result_val);
4874 return true;
4875 }
4876
4877 //---------------------------inline_native_getClass----------------------------
4878 // public final native Class<?> java.lang.Object.getClass();
4879 //
4880 // Build special case code for calls to getClass on an object.
4881 bool LibraryCallKit::inline_native_getClass() {
4882 Node* obj = null_check_receiver();
4883 if (stopped()) return true;
4884 set_result(load_mirror_from_klass(load_object_klass(obj)));
4885 return true;
4886 }
4887
4888 //-----------------inline_native_Reflection_getCallerClass---------------------
4889 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4890 //
4891 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4892 //
4893 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4894 // in that it must skip particular security frames and checks for
4895 // caller sensitive methods.
4896 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4897 #ifndef PRODUCT
4898 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4899 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4900 }
4901 #endif
4902
4903 if (!jvms()->has_method()) {
4904 #ifndef PRODUCT
4905 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4906 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4907 }
4908 #endif
4909 return false;
4910 }
4911
4912 // Walk back up the JVM state to find the caller at the required
4913 // depth.
4914 JVMState* caller_jvms = jvms();
4915
4916 // Cf. JVM_GetCallerClass
4917 // NOTE: Start the loop at depth 1 because the current JVM state does
4918 // not include the Reflection.getCallerClass() frame.
4919 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4920 ciMethod* m = caller_jvms->method();
4921 switch (n) {
4922 case 0:
4923 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4924 break;
4925 case 1:
4926 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4927 if (!m->caller_sensitive()) {
4928 #ifndef PRODUCT
4929 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4930 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4931 }
4932 #endif
4933 return false; // bail-out; let JVM_GetCallerClass do the work
4934 }
4935 break;
4936 default:
4937 if (!m->is_ignored_by_security_stack_walk()) {
4938 // We have reached the desired frame; return the holder class.
4939 // Acquire method holder as java.lang.Class and push as constant.
4940 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4941 ciInstance* caller_mirror = caller_klass->java_mirror();
4942 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4943
4944 #ifndef PRODUCT
4945 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4946 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());
4947 tty->print_cr(" JVM state at this point:");
4948 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4949 ciMethod* m = jvms()->of_depth(i)->method();
4950 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4951 }
4952 }
4953 #endif
4954 return true;
4955 }
4956 break;
4957 }
4958 }
4959
4960 #ifndef PRODUCT
4961 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4962 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4963 tty->print_cr(" JVM state at this point:");
4964 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4965 ciMethod* m = jvms()->of_depth(i)->method();
4966 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4967 }
4968 }
4969 #endif
4970
4971 return false; // bail-out; let JVM_GetCallerClass do the work
4972 }
4973
4974 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4975 Node* arg = argument(0);
4976 Node* result = nullptr;
4977
4978 switch (id) {
4979 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4980 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4981 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4982 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4983 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4984 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4985
4986 case vmIntrinsics::_doubleToLongBits: {
4987 // two paths (plus control) merge in a wood
4988 RegionNode *r = new RegionNode(3);
4989 Node *phi = new PhiNode(r, TypeLong::LONG);
4990
4991 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4992 // Build the boolean node
4993 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4994
4995 // Branch either way.
4996 // NaN case is less traveled, which makes all the difference.
4997 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4998 Node *opt_isnan = _gvn.transform(ifisnan);
4999 assert( opt_isnan->is_If(), "Expect an IfNode");
5000 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5001 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5002
5003 set_control(iftrue);
5004
5005 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5006 Node *slow_result = longcon(nan_bits); // return NaN
5007 phi->init_req(1, _gvn.transform( slow_result ));
5008 r->init_req(1, iftrue);
5009
5010 // Else fall through
5011 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5012 set_control(iffalse);
5013
5014 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5015 r->init_req(2, iffalse);
5016
5017 // Post merge
5018 set_control(_gvn.transform(r));
5019 record_for_igvn(r);
5020
5021 C->set_has_split_ifs(true); // Has chance for split-if optimization
5022 result = phi;
5023 assert(result->bottom_type()->isa_long(), "must be");
5024 break;
5025 }
5026
5027 case vmIntrinsics::_floatToIntBits: {
5028 // two paths (plus control) merge in a wood
5029 RegionNode *r = new RegionNode(3);
5030 Node *phi = new PhiNode(r, TypeInt::INT);
5031
5032 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5033 // Build the boolean node
5034 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5035
5036 // Branch either way.
5037 // NaN case is less traveled, which makes all the difference.
5038 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5039 Node *opt_isnan = _gvn.transform(ifisnan);
5040 assert( opt_isnan->is_If(), "Expect an IfNode");
5041 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5042 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5043
5044 set_control(iftrue);
5045
5046 static const jint nan_bits = 0x7fc00000;
5047 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5048 phi->init_req(1, _gvn.transform( slow_result ));
5049 r->init_req(1, iftrue);
5050
5051 // Else fall through
5052 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5053 set_control(iffalse);
5054
5055 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5056 r->init_req(2, iffalse);
5057
5058 // Post merge
5059 set_control(_gvn.transform(r));
5060 record_for_igvn(r);
5061
5062 C->set_has_split_ifs(true); // Has chance for split-if optimization
5063 result = phi;
5064 assert(result->bottom_type()->isa_int(), "must be");
5065 break;
5066 }
5067
5068 default:
5069 fatal_unexpected_iid(id);
5070 break;
5071 }
5072 set_result(_gvn.transform(result));
5073 return true;
5074 }
5075
5076 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5077 Node* arg = argument(0);
5078 Node* result = nullptr;
5079
5080 switch (id) {
5081 case vmIntrinsics::_floatIsInfinite:
5082 result = new IsInfiniteFNode(arg);
5083 break;
5084 case vmIntrinsics::_floatIsFinite:
5085 result = new IsFiniteFNode(arg);
5086 break;
5087 case vmIntrinsics::_doubleIsInfinite:
5088 result = new IsInfiniteDNode(arg);
5089 break;
5090 case vmIntrinsics::_doubleIsFinite:
5091 result = new IsFiniteDNode(arg);
5092 break;
5093 default:
5094 fatal_unexpected_iid(id);
5095 break;
5096 }
5097 set_result(_gvn.transform(result));
5098 return true;
5099 }
5100
5101 //----------------------inline_unsafe_copyMemory-------------------------
5102 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5103
5104 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5105 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5106 const Type* base_t = gvn.type(base);
5107
5108 bool in_native = (base_t == TypePtr::NULL_PTR);
5109 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5110 bool is_mixed = !in_heap && !in_native;
5111
5112 if (is_mixed) {
5113 return true; // mixed accesses can touch both on-heap and off-heap memory
5114 }
5115 if (in_heap) {
5116 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5117 if (!is_prim_array) {
5118 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5119 // there's not enough type information available to determine proper memory slice for it.
5120 return true;
5121 }
5122 }
5123 return false;
5124 }
5125
5126 bool LibraryCallKit::inline_unsafe_copyMemory() {
5127 if (callee()->is_static()) return false; // caller must have the capability!
5128 null_check_receiver(); // null-check receiver
5129 if (stopped()) return true;
5130
5131 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5132
5133 Node* src_base = argument(1); // type: oop
5134 Node* src_off = ConvL2X(argument(2)); // type: long
5135 Node* dst_base = argument(4); // type: oop
5136 Node* dst_off = ConvL2X(argument(5)); // type: long
5137 Node* size = ConvL2X(argument(7)); // type: long
5138
5139 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5140 "fieldOffset must be byte-scaled");
5141
5142 Node* src_addr = make_unsafe_address(src_base, src_off);
5143 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5144
5145 Node* thread = _gvn.transform(new ThreadLocalNode());
5146 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5147 BasicType doing_unsafe_access_bt = T_BYTE;
5148 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5149
5150 // update volatile field
5151 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5152
5153 int flags = RC_LEAF | RC_NO_FP;
5154
5155 const TypePtr* dst_type = TypePtr::BOTTOM;
5156
5157 // Adjust memory effects of the runtime call based on input values.
5158 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5159 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5160 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5161
5162 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5163 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5164 flags |= RC_NARROW_MEM; // narrow in memory
5165 }
5166 }
5167
5168 // Call it. Note that the length argument is not scaled.
5169 make_runtime_call(flags,
5170 OptoRuntime::fast_arraycopy_Type(),
5171 StubRoutines::unsafe_arraycopy(),
5172 "unsafe_arraycopy",
5173 dst_type,
5174 src_addr, dst_addr, size XTOP);
5175
5176 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5177
5178 return true;
5179 }
5180
5181 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5182 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5183 bool LibraryCallKit::inline_unsafe_setMemory() {
5184 if (callee()->is_static()) return false; // caller must have the capability!
5185 null_check_receiver(); // null-check receiver
5186 if (stopped()) return true;
5187
5188 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5189
5190 Node* dst_base = argument(1); // type: oop
5191 Node* dst_off = ConvL2X(argument(2)); // type: long
5192 Node* size = ConvL2X(argument(4)); // type: long
5193 Node* byte = argument(6); // type: byte
5194
5195 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5196 "fieldOffset must be byte-scaled");
5197
5198 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5199
5200 Node* thread = _gvn.transform(new ThreadLocalNode());
5201 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5202 BasicType doing_unsafe_access_bt = T_BYTE;
5203 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5204
5205 // update volatile field
5206 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5207
5208 int flags = RC_LEAF | RC_NO_FP;
5209
5210 const TypePtr* dst_type = TypePtr::BOTTOM;
5211
5212 // Adjust memory effects of the runtime call based on input values.
5213 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5214 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5215
5216 flags |= RC_NARROW_MEM; // narrow in memory
5217 }
5218
5219 // Call it. Note that the length argument is not scaled.
5220 make_runtime_call(flags,
5221 OptoRuntime::unsafe_setmemory_Type(),
5222 StubRoutines::unsafe_setmemory(),
5223 "unsafe_setmemory",
5224 dst_type,
5225 dst_addr, size XTOP, byte);
5226
5227 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5228
5229 return true;
5230 }
5231
5232 #undef XTOP
5233
5234 //------------------------clone_coping-----------------------------------
5235 // Helper function for inline_native_clone.
5236 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5237 assert(obj_size != nullptr, "");
5238 Node* raw_obj = alloc_obj->in(1);
5239 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5240
5241 AllocateNode* alloc = nullptr;
5242 if (ReduceBulkZeroing &&
5243 // If we are implementing an array clone without knowing its source type
5244 // (can happen when compiling the array-guarded branch of a reflective
5245 // Object.clone() invocation), initialize the array within the allocation.
5246 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5247 // to a runtime clone call that assumes fully initialized source arrays.
5248 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5249 // We will be completely responsible for initializing this object -
5250 // mark Initialize node as complete.
5251 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5252 // The object was just allocated - there should be no any stores!
5253 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5254 // Mark as complete_with_arraycopy so that on AllocateNode
5255 // expansion, we know this AllocateNode is initialized by an array
5256 // copy and a StoreStore barrier exists after the array copy.
5257 alloc->initialization()->set_complete_with_arraycopy();
5258 }
5259
5260 Node* size = _gvn.transform(obj_size);
5261 access_clone(obj, alloc_obj, size, is_array);
5262
5263 // Do not let reads from the cloned object float above the arraycopy.
5264 if (alloc != nullptr) {
5265 // Do not let stores that initialize this object be reordered with
5266 // a subsequent store that would make this object accessible by
5267 // other threads.
5268 // Record what AllocateNode this StoreStore protects so that
5269 // escape analysis can go from the MemBarStoreStoreNode to the
5270 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5271 // based on the escape status of the AllocateNode.
5272 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5273 } else {
5274 insert_mem_bar(Op_MemBarCPUOrder);
5275 }
5276 }
5277
5278 //------------------------inline_native_clone----------------------------
5279 // protected native Object java.lang.Object.clone();
5280 //
5281 // Here are the simple edge cases:
5282 // null receiver => normal trap
5283 // virtual and clone was overridden => slow path to out-of-line clone
5284 // not cloneable or finalizer => slow path to out-of-line Object.clone
5285 //
5286 // The general case has two steps, allocation and copying.
5287 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5288 //
5289 // Copying also has two cases, oop arrays and everything else.
5290 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5291 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5292 //
5293 // These steps fold up nicely if and when the cloned object's klass
5294 // can be sharply typed as an object array, a type array, or an instance.
5295 //
5296 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5297 PhiNode* result_val;
5298
5299 // Set the reexecute bit for the interpreter to reexecute
5300 // the bytecode that invokes Object.clone if deoptimization happens.
5301 { PreserveReexecuteState preexecs(this);
5302 jvms()->set_should_reexecute(true);
5303
5304 Node* obj = null_check_receiver();
5305 if (stopped()) return true;
5306
5307 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5308
5309 // If we are going to clone an instance, we need its exact type to
5310 // know the number and types of fields to convert the clone to
5311 // loads/stores. Maybe a speculative type can help us.
5312 if (!obj_type->klass_is_exact() &&
5313 obj_type->speculative_type() != nullptr &&
5314 obj_type->speculative_type()->is_instance_klass()) {
5315 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5316 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5317 !spec_ik->has_injected_fields()) {
5318 if (!obj_type->isa_instptr() ||
5319 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5320 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5321 }
5322 }
5323 }
5324
5325 // Conservatively insert a memory barrier on all memory slices.
5326 // Do not let writes into the original float below the clone.
5327 insert_mem_bar(Op_MemBarCPUOrder);
5328
5329 // paths into result_reg:
5330 enum {
5331 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5332 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5333 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5334 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5335 PATH_LIMIT
5336 };
5337 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5338 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5339 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5340 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5341 record_for_igvn(result_reg);
5342
5343 Node* obj_klass = load_object_klass(obj);
5344 Node* array_obj = obj;
5345 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5346 if (array_ctl != nullptr) {
5347 // It's an array.
5348 PreserveJVMState pjvms(this);
5349 set_control(array_ctl);
5350 Node* obj_length = load_array_length(array_obj);
5351 Node* array_size = nullptr; // Size of the array without object alignment padding.
5352 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5353
5354 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5355 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5356 // If it is an oop array, it requires very special treatment,
5357 // because gc barriers are required when accessing the array.
5358 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5359 if (is_obja != nullptr) {
5360 PreserveJVMState pjvms2(this);
5361 set_control(is_obja);
5362 // Generate a direct call to the right arraycopy function(s).
5363 // Clones are always tightly coupled.
5364 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5365 ac->set_clone_oop_array();
5366 Node* n = _gvn.transform(ac);
5367 assert(n == ac, "cannot disappear");
5368 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5369
5370 result_reg->init_req(_objArray_path, control());
5371 result_val->init_req(_objArray_path, alloc_obj);
5372 result_i_o ->set_req(_objArray_path, i_o());
5373 result_mem ->set_req(_objArray_path, reset_memory());
5374 }
5375 }
5376 // Otherwise, there are no barriers to worry about.
5377 // (We can dispense with card marks if we know the allocation
5378 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5379 // causes the non-eden paths to take compensating steps to
5380 // simulate a fresh allocation, so that no further
5381 // card marks are required in compiled code to initialize
5382 // the object.)
5383
5384 if (!stopped()) {
5385 copy_to_clone(array_obj, alloc_obj, array_size, true);
5386
5387 // Present the results of the copy.
5388 result_reg->init_req(_array_path, control());
5389 result_val->init_req(_array_path, alloc_obj);
5390 result_i_o ->set_req(_array_path, i_o());
5391 result_mem ->set_req(_array_path, reset_memory());
5392 }
5393 }
5394
5395 // We only go to the instance fast case code if we pass a number of guards.
5396 // The paths which do not pass are accumulated in the slow_region.
5397 RegionNode* slow_region = new RegionNode(1);
5398 record_for_igvn(slow_region);
5399 if (!stopped()) {
5400 // It's an instance (we did array above). Make the slow-path tests.
5401 // If this is a virtual call, we generate a funny guard. We grab
5402 // the vtable entry corresponding to clone() from the target object.
5403 // If the target method which we are calling happens to be the
5404 // Object clone() method, we pass the guard. We do not need this
5405 // guard for non-virtual calls; the caller is known to be the native
5406 // Object clone().
5407 if (is_virtual) {
5408 generate_virtual_guard(obj_klass, slow_region);
5409 }
5410
5411 // The object must be easily cloneable and must not have a finalizer.
5412 // Both of these conditions may be checked in a single test.
5413 // We could optimize the test further, but we don't care.
5414 generate_misc_flags_guard(obj_klass,
5415 // Test both conditions:
5416 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5417 // Must be cloneable but not finalizer:
5418 KlassFlags::_misc_is_cloneable_fast,
5419 slow_region);
5420 }
5421
5422 if (!stopped()) {
5423 // It's an instance, and it passed the slow-path tests.
5424 PreserveJVMState pjvms(this);
5425 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5426 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5427 // is reexecuted if deoptimization occurs and there could be problems when merging
5428 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5429 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5430
5431 copy_to_clone(obj, alloc_obj, obj_size, false);
5432
5433 // Present the results of the slow call.
5434 result_reg->init_req(_instance_path, control());
5435 result_val->init_req(_instance_path, alloc_obj);
5436 result_i_o ->set_req(_instance_path, i_o());
5437 result_mem ->set_req(_instance_path, reset_memory());
5438 }
5439
5440 // Generate code for the slow case. We make a call to clone().
5441 set_control(_gvn.transform(slow_region));
5442 if (!stopped()) {
5443 PreserveJVMState pjvms(this);
5444 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5445 // We need to deoptimize on exception (see comment above)
5446 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5447 // this->control() comes from set_results_for_java_call
5448 result_reg->init_req(_slow_path, control());
5449 result_val->init_req(_slow_path, slow_result);
5450 result_i_o ->set_req(_slow_path, i_o());
5451 result_mem ->set_req(_slow_path, reset_memory());
5452 }
5453
5454 // Return the combined state.
5455 set_control( _gvn.transform(result_reg));
5456 set_i_o( _gvn.transform(result_i_o));
5457 set_all_memory( _gvn.transform(result_mem));
5458 } // original reexecute is set back here
5459
5460 set_result(_gvn.transform(result_val));
5461 return true;
5462 }
5463
5464 // If we have a tightly coupled allocation, the arraycopy may take care
5465 // of the array initialization. If one of the guards we insert between
5466 // the allocation and the arraycopy causes a deoptimization, an
5467 // uninitialized array will escape the compiled method. To prevent that
5468 // we set the JVM state for uncommon traps between the allocation and
5469 // the arraycopy to the state before the allocation so, in case of
5470 // deoptimization, we'll reexecute the allocation and the
5471 // initialization.
5472 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5473 if (alloc != nullptr) {
5474 ciMethod* trap_method = alloc->jvms()->method();
5475 int trap_bci = alloc->jvms()->bci();
5476
5477 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5478 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5479 // Make sure there's no store between the allocation and the
5480 // arraycopy otherwise visible side effects could be rexecuted
5481 // in case of deoptimization and cause incorrect execution.
5482 bool no_interfering_store = true;
5483 Node* mem = alloc->in(TypeFunc::Memory);
5484 if (mem->is_MergeMem()) {
5485 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5486 Node* n = mms.memory();
5487 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5488 assert(n->is_Store(), "what else?");
5489 no_interfering_store = false;
5490 break;
5491 }
5492 }
5493 } else {
5494 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5495 Node* n = mms.memory();
5496 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5497 assert(n->is_Store(), "what else?");
5498 no_interfering_store = false;
5499 break;
5500 }
5501 }
5502 }
5503
5504 if (no_interfering_store) {
5505 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5506
5507 JVMState* saved_jvms = jvms();
5508 saved_reexecute_sp = _reexecute_sp;
5509
5510 set_jvms(sfpt->jvms());
5511 _reexecute_sp = jvms()->sp();
5512
5513 return saved_jvms;
5514 }
5515 }
5516 }
5517 return nullptr;
5518 }
5519
5520 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5521 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5522 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5523 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5524 uint size = alloc->req();
5525 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5526 old_jvms->set_map(sfpt);
5527 for (uint i = 0; i < size; i++) {
5528 sfpt->init_req(i, alloc->in(i));
5529 }
5530 // re-push array length for deoptimization
5531 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5532 old_jvms->set_sp(old_jvms->sp()+1);
5533 old_jvms->set_monoff(old_jvms->monoff()+1);
5534 old_jvms->set_scloff(old_jvms->scloff()+1);
5535 old_jvms->set_endoff(old_jvms->endoff()+1);
5536 old_jvms->set_should_reexecute(true);
5537
5538 sfpt->set_i_o(map()->i_o());
5539 sfpt->set_memory(map()->memory());
5540 sfpt->set_control(map()->control());
5541 return sfpt;
5542 }
5543
5544 // In case of a deoptimization, we restart execution at the
5545 // allocation, allocating a new array. We would leave an uninitialized
5546 // array in the heap that GCs wouldn't expect. Move the allocation
5547 // after the traps so we don't allocate the array if we
5548 // deoptimize. This is possible because tightly_coupled_allocation()
5549 // guarantees there's no observer of the allocated array at this point
5550 // and the control flow is simple enough.
5551 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5552 int saved_reexecute_sp, uint new_idx) {
5553 if (saved_jvms_before_guards != nullptr && !stopped()) {
5554 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5555
5556 assert(alloc != nullptr, "only with a tightly coupled allocation");
5557 // restore JVM state to the state at the arraycopy
5558 saved_jvms_before_guards->map()->set_control(map()->control());
5559 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5560 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5561 // If we've improved the types of some nodes (null check) while
5562 // emitting the guards, propagate them to the current state
5563 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5564 set_jvms(saved_jvms_before_guards);
5565 _reexecute_sp = saved_reexecute_sp;
5566
5567 // Remove the allocation from above the guards
5568 CallProjections callprojs;
5569 alloc->extract_projections(&callprojs, true);
5570 InitializeNode* init = alloc->initialization();
5571 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5572 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5573 init->replace_mem_projs_by(alloc_mem, C);
5574
5575 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5576 // the allocation (i.e. is only valid if the allocation succeeds):
5577 // 1) replace CastIINode with AllocateArrayNode's length here
5578 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5579 //
5580 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5581 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5582 Node* init_control = init->proj_out(TypeFunc::Control);
5583 Node* alloc_length = alloc->Ideal_length();
5584 #ifdef ASSERT
5585 Node* prev_cast = nullptr;
5586 #endif
5587 for (uint i = 0; i < init_control->outcnt(); i++) {
5588 Node* init_out = init_control->raw_out(i);
5589 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5590 #ifdef ASSERT
5591 if (prev_cast == nullptr) {
5592 prev_cast = init_out;
5593 } else {
5594 if (prev_cast->cmp(*init_out) == false) {
5595 prev_cast->dump();
5596 init_out->dump();
5597 assert(false, "not equal CastIINode");
5598 }
5599 }
5600 #endif
5601 C->gvn_replace_by(init_out, alloc_length);
5602 }
5603 }
5604 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5605
5606 // move the allocation here (after the guards)
5607 _gvn.hash_delete(alloc);
5608 alloc->set_req(TypeFunc::Control, control());
5609 alloc->set_req(TypeFunc::I_O, i_o());
5610 Node *mem = reset_memory();
5611 set_all_memory(mem);
5612 alloc->set_req(TypeFunc::Memory, mem);
5613 set_control(init->proj_out_or_null(TypeFunc::Control));
5614 set_i_o(callprojs.fallthrough_ioproj);
5615
5616 // Update memory as done in GraphKit::set_output_for_allocation()
5617 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5618 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5619 if (ary_type->isa_aryptr() && length_type != nullptr) {
5620 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5621 }
5622 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5623 int elemidx = C->get_alias_index(telemref);
5624 // Need to properly move every memory projection for the Initialize
5625 #ifdef ASSERT
5626 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5627 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
5628 #endif
5629 auto move_proj = [&](ProjNode* proj) {
5630 int alias_idx = C->get_alias_index(proj->adr_type());
5631 assert(alias_idx == Compile::AliasIdxRaw ||
5632 alias_idx == elemidx ||
5633 alias_idx == mark_idx ||
5634 alias_idx == klass_idx, "should be raw memory or array element type");
5635 set_memory(proj, alias_idx);
5636 };
5637 init->for_each_proj(move_proj, TypeFunc::Memory);
5638
5639 Node* allocx = _gvn.transform(alloc);
5640 assert(allocx == alloc, "where has the allocation gone?");
5641 assert(dest->is_CheckCastPP(), "not an allocation result?");
5642
5643 _gvn.hash_delete(dest);
5644 dest->set_req(0, control());
5645 Node* destx = _gvn.transform(dest);
5646 assert(destx == dest, "where has the allocation result gone?");
5647
5648 array_ideal_length(alloc, ary_type, true);
5649 }
5650 }
5651
5652 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5653 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5654 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5655 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5656 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5657 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5658 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5659 JVMState* saved_jvms_before_guards) {
5660 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5661 // There is at least one unrelated uncommon trap which needs to be replaced.
5662 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5663
5664 JVMState* saved_jvms = jvms();
5665 const int saved_reexecute_sp = _reexecute_sp;
5666 set_jvms(sfpt->jvms());
5667 _reexecute_sp = jvms()->sp();
5668
5669 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5670
5671 // Restore state
5672 set_jvms(saved_jvms);
5673 _reexecute_sp = saved_reexecute_sp;
5674 }
5675 }
5676
5677 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5678 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5679 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5680 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5681 while (if_proj->is_IfProj()) {
5682 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5683 if (uncommon_trap != nullptr) {
5684 create_new_uncommon_trap(uncommon_trap);
5685 }
5686 assert(if_proj->in(0)->is_If(), "must be If");
5687 if_proj = if_proj->in(0)->in(0);
5688 }
5689 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5690 "must have reached control projection of init node");
5691 }
5692
5693 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5694 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5695 assert(trap_request != 0, "no valid UCT trap request");
5696 PreserveJVMState pjvms(this);
5697 set_control(uncommon_trap_call->in(0));
5698 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5699 Deoptimization::trap_request_action(trap_request));
5700 assert(stopped(), "Should be stopped");
5701 _gvn.hash_delete(uncommon_trap_call);
5702 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5703 }
5704
5705 // Common checks for array sorting intrinsics arguments.
5706 // Returns `true` if checks passed.
5707 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5708 // check address of the class
5709 if (elementType == nullptr || elementType->is_top()) {
5710 return false; // dead path
5711 }
5712 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5713 if (elem_klass == nullptr) {
5714 return false; // dead path
5715 }
5716 // java_mirror_type() returns non-null for compile-time Class constants only
5717 ciType* elem_type = elem_klass->java_mirror_type();
5718 if (elem_type == nullptr) {
5719 return false;
5720 }
5721 bt = elem_type->basic_type();
5722 // Disable the intrinsic if the CPU does not support SIMD sort
5723 if (!Matcher::supports_simd_sort(bt)) {
5724 return false;
5725 }
5726 // check address of the array
5727 if (obj == nullptr || obj->is_top()) {
5728 return false; // dead path
5729 }
5730 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5731 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5732 return false; // failed input validation
5733 }
5734 return true;
5735 }
5736
5737 //------------------------------inline_array_partition-----------------------
5738 bool LibraryCallKit::inline_array_partition() {
5739 address stubAddr = StubRoutines::select_array_partition_function();
5740 if (stubAddr == nullptr) {
5741 return false; // Intrinsic's stub is not implemented on this platform
5742 }
5743 assert(callee()->signature()->size() == 9, "arrayPartition has 8 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 Node* indexPivot1 = argument(6);
5752 Node* indexPivot2 = argument(7);
5753 // PartitionOperation: argument(8) is ignored
5754
5755 Node* pivotIndices = nullptr;
5756 BasicType bt = T_ILLEGAL;
5757
5758 if (!check_array_sort_arguments(elementType, obj, bt)) {
5759 return false;
5760 }
5761 null_check(obj);
5762 // If obj is dead, only null-path is taken.
5763 if (stopped()) {
5764 return true;
5765 }
5766 // Set the original stack and the reexecute bit for the interpreter to reexecute
5767 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5768 { PreserveReexecuteState preexecs(this);
5769 jvms()->set_should_reexecute(true);
5770
5771 Node* obj_adr = make_unsafe_address(obj, offset);
5772
5773 // create the pivotIndices array of type int and size = 2
5774 Node* size = intcon(2);
5775 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5776 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5777 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5778 guarantee(alloc != nullptr, "created above");
5779 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5780
5781 // pass the basic type enum to the stub
5782 Node* elemType = intcon(bt);
5783
5784 // Call the stub
5785 const char *stubName = "array_partition_stub";
5786 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5787 stubAddr, stubName, TypePtr::BOTTOM,
5788 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5789 indexPivot1, indexPivot2);
5790
5791 } // original reexecute is set back here
5792
5793 if (!stopped()) {
5794 set_result(pivotIndices);
5795 }
5796
5797 return true;
5798 }
5799
5800
5801 //------------------------------inline_array_sort-----------------------
5802 bool LibraryCallKit::inline_array_sort() {
5803 address stubAddr = StubRoutines::select_arraysort_function();
5804 if (stubAddr == nullptr) {
5805 return false; // Intrinsic's stub is not implemented on this platform
5806 }
5807 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5808
5809 // no receiver because it is a static method
5810 Node* elementType = argument(0);
5811 Node* obj = argument(1);
5812 Node* offset = argument(2); // long
5813 Node* fromIndex = argument(4);
5814 Node* toIndex = argument(5);
5815 // SortOperation: argument(6) is ignored
5816
5817 BasicType bt = T_ILLEGAL;
5818
5819 if (!check_array_sort_arguments(elementType, obj, bt)) {
5820 return false;
5821 }
5822 null_check(obj);
5823 // If obj is dead, only null-path is taken.
5824 if (stopped()) {
5825 return true;
5826 }
5827 Node* obj_adr = make_unsafe_address(obj, offset);
5828
5829 // pass the basic type enum to the stub
5830 Node* elemType = intcon(bt);
5831
5832 // Call the stub.
5833 const char *stubName = "arraysort_stub";
5834 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5835 stubAddr, stubName, TypePtr::BOTTOM,
5836 obj_adr, elemType, fromIndex, toIndex);
5837
5838 return true;
5839 }
5840
5841
5842 //------------------------------inline_arraycopy-----------------------
5843 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5844 // Object dest, int destPos,
5845 // int length);
5846 bool LibraryCallKit::inline_arraycopy() {
5847 // Get the arguments.
5848 Node* src = argument(0); // type: oop
5849 Node* src_offset = argument(1); // type: int
5850 Node* dest = argument(2); // type: oop
5851 Node* dest_offset = argument(3); // type: int
5852 Node* length = argument(4); // type: int
5853
5854 uint new_idx = C->unique();
5855
5856 // Check for allocation before we add nodes that would confuse
5857 // tightly_coupled_allocation()
5858 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5859
5860 int saved_reexecute_sp = -1;
5861 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5862 // See arraycopy_restore_alloc_state() comment
5863 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5864 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5865 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5866 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5867
5868 // The following tests must be performed
5869 // (1) src and dest are arrays.
5870 // (2) src and dest arrays must have elements of the same BasicType
5871 // (3) src and dest must not be null.
5872 // (4) src_offset must not be negative.
5873 // (5) dest_offset must not be negative.
5874 // (6) length must not be negative.
5875 // (7) src_offset + length must not exceed length of src.
5876 // (8) dest_offset + length must not exceed length of dest.
5877 // (9) each element of an oop array must be assignable
5878
5879 // (3) src and dest must not be null.
5880 // always do this here because we need the JVM state for uncommon traps
5881 Node* null_ctl = top();
5882 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5883 assert(null_ctl->is_top(), "no null control here");
5884 dest = null_check(dest, T_ARRAY);
5885
5886 if (!can_emit_guards) {
5887 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5888 // guards but the arraycopy node could still take advantage of a
5889 // tightly allocated allocation. tightly_coupled_allocation() is
5890 // called again to make sure it takes the null check above into
5891 // account: the null check is mandatory and if it caused an
5892 // uncommon trap to be emitted then the allocation can't be
5893 // considered tightly coupled in this context.
5894 alloc = tightly_coupled_allocation(dest);
5895 }
5896
5897 bool validated = false;
5898
5899 const Type* src_type = _gvn.type(src);
5900 const Type* dest_type = _gvn.type(dest);
5901 const TypeAryPtr* top_src = src_type->isa_aryptr();
5902 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5903
5904 // Do we have the type of src?
5905 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5906 // Do we have the type of dest?
5907 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5908 // Is the type for src from speculation?
5909 bool src_spec = false;
5910 // Is the type for dest from speculation?
5911 bool dest_spec = false;
5912
5913 if ((!has_src || !has_dest) && can_emit_guards) {
5914 // We don't have sufficient type information, let's see if
5915 // speculative types can help. We need to have types for both src
5916 // and dest so that it pays off.
5917
5918 // Do we already have or could we have type information for src
5919 bool could_have_src = has_src;
5920 // Do we already have or could we have type information for dest
5921 bool could_have_dest = has_dest;
5922
5923 ciKlass* src_k = nullptr;
5924 if (!has_src) {
5925 src_k = src_type->speculative_type_not_null();
5926 if (src_k != nullptr && src_k->is_array_klass()) {
5927 could_have_src = true;
5928 }
5929 }
5930
5931 ciKlass* dest_k = nullptr;
5932 if (!has_dest) {
5933 dest_k = dest_type->speculative_type_not_null();
5934 if (dest_k != nullptr && dest_k->is_array_klass()) {
5935 could_have_dest = true;
5936 }
5937 }
5938
5939 if (could_have_src && could_have_dest) {
5940 // This is going to pay off so emit the required guards
5941 if (!has_src) {
5942 src = maybe_cast_profiled_obj(src, src_k, true);
5943 src_type = _gvn.type(src);
5944 top_src = src_type->isa_aryptr();
5945 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5946 src_spec = true;
5947 }
5948 if (!has_dest) {
5949 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5950 dest_type = _gvn.type(dest);
5951 top_dest = dest_type->isa_aryptr();
5952 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5953 dest_spec = true;
5954 }
5955 }
5956 }
5957
5958 if (has_src && has_dest && can_emit_guards) {
5959 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5960 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5961 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5962 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5963
5964 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5965 // If both arrays are object arrays then having the exact types
5966 // for both will remove the need for a subtype check at runtime
5967 // before the call and may make it possible to pick a faster copy
5968 // routine (without a subtype check on every element)
5969 // Do we have the exact type of src?
5970 bool could_have_src = src_spec;
5971 // Do we have the exact type of dest?
5972 bool could_have_dest = dest_spec;
5973 ciKlass* src_k = nullptr;
5974 ciKlass* dest_k = nullptr;
5975 if (!src_spec) {
5976 src_k = src_type->speculative_type_not_null();
5977 if (src_k != nullptr && src_k->is_array_klass()) {
5978 could_have_src = true;
5979 }
5980 }
5981 if (!dest_spec) {
5982 dest_k = dest_type->speculative_type_not_null();
5983 if (dest_k != nullptr && dest_k->is_array_klass()) {
5984 could_have_dest = true;
5985 }
5986 }
5987 if (could_have_src && could_have_dest) {
5988 // If we can have both exact types, emit the missing guards
5989 if (could_have_src && !src_spec) {
5990 src = maybe_cast_profiled_obj(src, src_k, true);
5991 }
5992 if (could_have_dest && !dest_spec) {
5993 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5994 }
5995 }
5996 }
5997 }
5998
5999 ciMethod* trap_method = method();
6000 int trap_bci = bci();
6001 if (saved_jvms_before_guards != nullptr) {
6002 trap_method = alloc->jvms()->method();
6003 trap_bci = alloc->jvms()->bci();
6004 }
6005
6006 bool negative_length_guard_generated = false;
6007
6008 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6009 can_emit_guards &&
6010 !src->is_top() && !dest->is_top()) {
6011 // validate arguments: enables transformation the ArrayCopyNode
6012 validated = true;
6013
6014 RegionNode* slow_region = new RegionNode(1);
6015 record_for_igvn(slow_region);
6016
6017 // (1) src and dest are arrays.
6018 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6019 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6020
6021 // (2) src and dest arrays must have elements of the same BasicType
6022 // done at macro expansion or at Ideal transformation time
6023
6024 // (4) src_offset must not be negative.
6025 generate_negative_guard(src_offset, slow_region);
6026
6027 // (5) dest_offset must not be negative.
6028 generate_negative_guard(dest_offset, slow_region);
6029
6030 // (7) src_offset + length must not exceed length of src.
6031 generate_limit_guard(src_offset, length,
6032 load_array_length(src),
6033 slow_region);
6034
6035 // (8) dest_offset + length must not exceed length of dest.
6036 generate_limit_guard(dest_offset, length,
6037 load_array_length(dest),
6038 slow_region);
6039
6040 // (6) length must not be negative.
6041 // This is also checked in generate_arraycopy() during macro expansion, but
6042 // we also have to check it here for the case where the ArrayCopyNode will
6043 // be eliminated by Escape Analysis.
6044 if (EliminateAllocations) {
6045 generate_negative_guard(length, slow_region);
6046 negative_length_guard_generated = true;
6047 }
6048
6049 // (9) each element of an oop array must be assignable
6050 Node* dest_klass = load_object_klass(dest);
6051 if (src != dest) {
6052 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6053
6054 if (not_subtype_ctrl != top()) {
6055 PreserveJVMState pjvms(this);
6056 set_control(not_subtype_ctrl);
6057 uncommon_trap(Deoptimization::Reason_intrinsic,
6058 Deoptimization::Action_make_not_entrant);
6059 assert(stopped(), "Should be stopped");
6060 }
6061 }
6062 {
6063 PreserveJVMState pjvms(this);
6064 set_control(_gvn.transform(slow_region));
6065 uncommon_trap(Deoptimization::Reason_intrinsic,
6066 Deoptimization::Action_make_not_entrant);
6067 assert(stopped(), "Should be stopped");
6068 }
6069
6070 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6071 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6072 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6073 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6074 }
6075
6076 if (stopped()) {
6077 return true;
6078 }
6079
6080 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6081 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6082 // so the compiler has a chance to eliminate them: during macro expansion,
6083 // we have to set their control (CastPP nodes are eliminated).
6084 load_object_klass(src), load_object_klass(dest),
6085 load_array_length(src), load_array_length(dest));
6086
6087 ac->set_arraycopy(validated);
6088
6089 Node* n = _gvn.transform(ac);
6090 if (n == ac) {
6091 ac->connect_outputs(this);
6092 } else {
6093 assert(validated, "shouldn't transform if all arguments not validated");
6094 set_all_memory(n);
6095 }
6096 clear_upper_avx();
6097
6098
6099 return true;
6100 }
6101
6102
6103 // Helper function which determines if an arraycopy immediately follows
6104 // an allocation, with no intervening tests or other escapes for the object.
6105 AllocateArrayNode*
6106 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6107 if (stopped()) return nullptr; // no fast path
6108 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6109
6110 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6111 if (alloc == nullptr) return nullptr;
6112
6113 Node* rawmem = memory(Compile::AliasIdxRaw);
6114 // Is the allocation's memory state untouched?
6115 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6116 // Bail out if there have been raw-memory effects since the allocation.
6117 // (Example: There might have been a call or safepoint.)
6118 return nullptr;
6119 }
6120 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6121 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6122 return nullptr;
6123 }
6124
6125 // There must be no unexpected observers of this allocation.
6126 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6127 Node* obs = ptr->fast_out(i);
6128 if (obs != this->map()) {
6129 return nullptr;
6130 }
6131 }
6132
6133 // This arraycopy must unconditionally follow the allocation of the ptr.
6134 Node* alloc_ctl = ptr->in(0);
6135 Node* ctl = control();
6136 while (ctl != alloc_ctl) {
6137 // There may be guards which feed into the slow_region.
6138 // Any other control flow means that we might not get a chance
6139 // to finish initializing the allocated object.
6140 // Various low-level checks bottom out in uncommon traps. These
6141 // are considered safe since we've already checked above that
6142 // there is no unexpected observer of this allocation.
6143 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6144 assert(ctl->in(0)->is_If(), "must be If");
6145 ctl = ctl->in(0)->in(0);
6146 } else {
6147 return nullptr;
6148 }
6149 }
6150
6151 // If we get this far, we have an allocation which immediately
6152 // precedes the arraycopy, and we can take over zeroing the new object.
6153 // The arraycopy will finish the initialization, and provide
6154 // a new control state to which we will anchor the destination pointer.
6155
6156 return alloc;
6157 }
6158
6159 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6160 if (node->is_IfProj()) {
6161 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6162 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6163 Node* obs = other_proj->fast_out(j);
6164 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6165 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6166 return obs->as_CallStaticJava();
6167 }
6168 }
6169 }
6170 return nullptr;
6171 }
6172
6173 //-------------inline_encodeISOArray-----------------------------------
6174 // encode char[] to byte[] in ISO_8859_1 or ASCII
6175 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6176 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6177 // no receiver since it is static method
6178 Node *src = argument(0);
6179 Node *src_offset = argument(1);
6180 Node *dst = argument(2);
6181 Node *dst_offset = argument(3);
6182 Node *length = argument(4);
6183
6184 src = must_be_not_null(src, true);
6185 dst = must_be_not_null(dst, true);
6186
6187 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6188 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6189 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6190 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6191 // failed array check
6192 return false;
6193 }
6194
6195 // Figure out the size and type of the elements we will be copying.
6196 BasicType src_elem = src_type->elem()->array_element_basic_type();
6197 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6198 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6199 return false;
6200 }
6201
6202 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6203 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6204 // 'src_start' points to src array + scaled offset
6205 // 'dst_start' points to dst array + scaled offset
6206
6207 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6208 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6209 enc = _gvn.transform(enc);
6210 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6211 set_memory(res_mem, mtype);
6212 set_result(enc);
6213 clear_upper_avx();
6214
6215 return true;
6216 }
6217
6218 //-------------inline_multiplyToLen-----------------------------------
6219 bool LibraryCallKit::inline_multiplyToLen() {
6220 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6221
6222 address stubAddr = StubRoutines::multiplyToLen();
6223 if (stubAddr == nullptr) {
6224 return false; // Intrinsic's stub is not implemented on this platform
6225 }
6226 const char* stubName = "multiplyToLen";
6227
6228 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6229
6230 // no receiver because it is a static method
6231 Node* x = argument(0);
6232 Node* xlen = argument(1);
6233 Node* y = argument(2);
6234 Node* ylen = argument(3);
6235 Node* z = argument(4);
6236
6237 x = must_be_not_null(x, true);
6238 y = must_be_not_null(y, true);
6239
6240 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6241 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6242 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6243 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6244 // failed array check
6245 return false;
6246 }
6247
6248 BasicType x_elem = x_type->elem()->array_element_basic_type();
6249 BasicType y_elem = y_type->elem()->array_element_basic_type();
6250 if (x_elem != T_INT || y_elem != T_INT) {
6251 return false;
6252 }
6253
6254 Node* x_start = array_element_address(x, intcon(0), x_elem);
6255 Node* y_start = array_element_address(y, intcon(0), y_elem);
6256 // 'x_start' points to x array + scaled xlen
6257 // 'y_start' points to y array + scaled ylen
6258
6259 Node* z_start = array_element_address(z, intcon(0), T_INT);
6260
6261 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6262 OptoRuntime::multiplyToLen_Type(),
6263 stubAddr, stubName, TypePtr::BOTTOM,
6264 x_start, xlen, y_start, ylen, z_start);
6265
6266 C->set_has_split_ifs(true); // Has chance for split-if optimization
6267 set_result(z);
6268 return true;
6269 }
6270
6271 //-------------inline_squareToLen------------------------------------
6272 bool LibraryCallKit::inline_squareToLen() {
6273 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6274
6275 address stubAddr = StubRoutines::squareToLen();
6276 if (stubAddr == nullptr) {
6277 return false; // Intrinsic's stub is not implemented on this platform
6278 }
6279 const char* stubName = "squareToLen";
6280
6281 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6282
6283 Node* x = argument(0);
6284 Node* len = argument(1);
6285 Node* z = argument(2);
6286 Node* zlen = argument(3);
6287
6288 x = must_be_not_null(x, true);
6289 z = must_be_not_null(z, true);
6290
6291 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6292 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6293 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6294 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6295 // failed array check
6296 return false;
6297 }
6298
6299 BasicType x_elem = x_type->elem()->array_element_basic_type();
6300 BasicType z_elem = z_type->elem()->array_element_basic_type();
6301 if (x_elem != T_INT || z_elem != T_INT) {
6302 return false;
6303 }
6304
6305
6306 Node* x_start = array_element_address(x, intcon(0), x_elem);
6307 Node* z_start = array_element_address(z, intcon(0), z_elem);
6308
6309 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6310 OptoRuntime::squareToLen_Type(),
6311 stubAddr, stubName, TypePtr::BOTTOM,
6312 x_start, len, z_start, zlen);
6313
6314 set_result(z);
6315 return true;
6316 }
6317
6318 //-------------inline_mulAdd------------------------------------------
6319 bool LibraryCallKit::inline_mulAdd() {
6320 assert(UseMulAddIntrinsic, "not implemented on this platform");
6321
6322 address stubAddr = StubRoutines::mulAdd();
6323 if (stubAddr == nullptr) {
6324 return false; // Intrinsic's stub is not implemented on this platform
6325 }
6326 const char* stubName = "mulAdd";
6327
6328 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6329
6330 Node* out = argument(0);
6331 Node* in = argument(1);
6332 Node* offset = argument(2);
6333 Node* len = argument(3);
6334 Node* k = argument(4);
6335
6336 in = must_be_not_null(in, true);
6337 out = must_be_not_null(out, true);
6338
6339 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6340 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6341 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6342 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6343 // failed array check
6344 return false;
6345 }
6346
6347 BasicType out_elem = out_type->elem()->array_element_basic_type();
6348 BasicType in_elem = in_type->elem()->array_element_basic_type();
6349 if (out_elem != T_INT || in_elem != T_INT) {
6350 return false;
6351 }
6352
6353 Node* outlen = load_array_length(out);
6354 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6355 Node* out_start = array_element_address(out, intcon(0), out_elem);
6356 Node* in_start = array_element_address(in, intcon(0), in_elem);
6357
6358 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6359 OptoRuntime::mulAdd_Type(),
6360 stubAddr, stubName, TypePtr::BOTTOM,
6361 out_start,in_start, new_offset, len, k);
6362 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6363 set_result(result);
6364 return true;
6365 }
6366
6367 //-------------inline_montgomeryMultiply-----------------------------------
6368 bool LibraryCallKit::inline_montgomeryMultiply() {
6369 address stubAddr = StubRoutines::montgomeryMultiply();
6370 if (stubAddr == nullptr) {
6371 return false; // Intrinsic's stub is not implemented on this platform
6372 }
6373
6374 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6375 const char* stubName = "montgomery_multiply";
6376
6377 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6378
6379 Node* a = argument(0);
6380 Node* b = argument(1);
6381 Node* n = argument(2);
6382 Node* len = argument(3);
6383 Node* inv = argument(4);
6384 Node* m = argument(6);
6385
6386 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6387 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6388 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6389 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6390 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6391 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6392 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6393 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6394 // failed array check
6395 return false;
6396 }
6397
6398 BasicType a_elem = a_type->elem()->array_element_basic_type();
6399 BasicType b_elem = b_type->elem()->array_element_basic_type();
6400 BasicType n_elem = n_type->elem()->array_element_basic_type();
6401 BasicType m_elem = m_type->elem()->array_element_basic_type();
6402 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6403 return false;
6404 }
6405
6406 // Make the call
6407 {
6408 Node* a_start = array_element_address(a, intcon(0), a_elem);
6409 Node* b_start = array_element_address(b, intcon(0), b_elem);
6410 Node* n_start = array_element_address(n, intcon(0), n_elem);
6411 Node* m_start = array_element_address(m, intcon(0), m_elem);
6412
6413 Node* call = make_runtime_call(RC_LEAF,
6414 OptoRuntime::montgomeryMultiply_Type(),
6415 stubAddr, stubName, TypePtr::BOTTOM,
6416 a_start, b_start, n_start, len, inv, top(),
6417 m_start);
6418 set_result(m);
6419 }
6420
6421 return true;
6422 }
6423
6424 bool LibraryCallKit::inline_montgomerySquare() {
6425 address stubAddr = StubRoutines::montgomerySquare();
6426 if (stubAddr == nullptr) {
6427 return false; // Intrinsic's stub is not implemented on this platform
6428 }
6429
6430 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6431 const char* stubName = "montgomery_square";
6432
6433 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6434
6435 Node* a = argument(0);
6436 Node* n = argument(1);
6437 Node* len = argument(2);
6438 Node* inv = argument(3);
6439 Node* m = argument(5);
6440
6441 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6442 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6443 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6444 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6445 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6446 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6447 // failed array check
6448 return false;
6449 }
6450
6451 BasicType a_elem = a_type->elem()->array_element_basic_type();
6452 BasicType n_elem = n_type->elem()->array_element_basic_type();
6453 BasicType m_elem = m_type->elem()->array_element_basic_type();
6454 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6455 return false;
6456 }
6457
6458 // Make the call
6459 {
6460 Node* a_start = array_element_address(a, intcon(0), a_elem);
6461 Node* n_start = array_element_address(n, intcon(0), n_elem);
6462 Node* m_start = array_element_address(m, intcon(0), m_elem);
6463
6464 Node* call = make_runtime_call(RC_LEAF,
6465 OptoRuntime::montgomerySquare_Type(),
6466 stubAddr, stubName, TypePtr::BOTTOM,
6467 a_start, n_start, len, inv, top(),
6468 m_start);
6469 set_result(m);
6470 }
6471
6472 return true;
6473 }
6474
6475 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6476 address stubAddr = nullptr;
6477 const char* stubName = nullptr;
6478
6479 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6480 if (stubAddr == nullptr) {
6481 return false; // Intrinsic's stub is not implemented on this platform
6482 }
6483
6484 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6485
6486 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6487
6488 Node* newArr = argument(0);
6489 Node* oldArr = argument(1);
6490 Node* newIdx = argument(2);
6491 Node* shiftCount = argument(3);
6492 Node* numIter = argument(4);
6493
6494 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6495 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6496 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6497 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6498 return false;
6499 }
6500
6501 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6502 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6503 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6504 return false;
6505 }
6506
6507 // Make the call
6508 {
6509 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6510 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6511
6512 Node* call = make_runtime_call(RC_LEAF,
6513 OptoRuntime::bigIntegerShift_Type(),
6514 stubAddr,
6515 stubName,
6516 TypePtr::BOTTOM,
6517 newArr_start,
6518 oldArr_start,
6519 newIdx,
6520 shiftCount,
6521 numIter);
6522 }
6523
6524 return true;
6525 }
6526
6527 //-------------inline_vectorizedMismatch------------------------------
6528 bool LibraryCallKit::inline_vectorizedMismatch() {
6529 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6530
6531 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6532 Node* obja = argument(0); // Object
6533 Node* aoffset = argument(1); // long
6534 Node* objb = argument(3); // Object
6535 Node* boffset = argument(4); // long
6536 Node* length = argument(6); // int
6537 Node* scale = argument(7); // int
6538
6539 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6540 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6541 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6542 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6543 scale == top()) {
6544 return false; // failed input validation
6545 }
6546
6547 Node* obja_adr = make_unsafe_address(obja, aoffset);
6548 Node* objb_adr = make_unsafe_address(objb, boffset);
6549
6550 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6551 //
6552 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6553 // if (length <= inline_limit) {
6554 // inline_path:
6555 // vmask = VectorMaskGen length
6556 // vload1 = LoadVectorMasked obja, vmask
6557 // vload2 = LoadVectorMasked objb, vmask
6558 // result1 = VectorCmpMasked vload1, vload2, vmask
6559 // } else {
6560 // call_stub_path:
6561 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6562 // }
6563 // exit_block:
6564 // return Phi(result1, result2);
6565 //
6566 enum { inline_path = 1, // input is small enough to process it all at once
6567 stub_path = 2, // input is too large; call into the VM
6568 PATH_LIMIT = 3
6569 };
6570
6571 Node* exit_block = new RegionNode(PATH_LIMIT);
6572 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6573 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6574
6575 Node* call_stub_path = control();
6576
6577 BasicType elem_bt = T_ILLEGAL;
6578
6579 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6580 if (scale_t->is_con()) {
6581 switch (scale_t->get_con()) {
6582 case 0: elem_bt = T_BYTE; break;
6583 case 1: elem_bt = T_SHORT; break;
6584 case 2: elem_bt = T_INT; break;
6585 case 3: elem_bt = T_LONG; break;
6586
6587 default: elem_bt = T_ILLEGAL; break; // not supported
6588 }
6589 }
6590
6591 int inline_limit = 0;
6592 bool do_partial_inline = false;
6593
6594 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6595 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6596 do_partial_inline = inline_limit >= 16;
6597 }
6598
6599 if (do_partial_inline) {
6600 assert(elem_bt != T_ILLEGAL, "sanity");
6601
6602 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6603 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6604 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6605
6606 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6607 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6608 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6609
6610 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6611
6612 if (!stopped()) {
6613 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6614
6615 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6616 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6617 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6618 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6619
6620 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6621 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6622 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6623 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6624
6625 exit_block->init_req(inline_path, control());
6626 memory_phi->init_req(inline_path, map()->memory());
6627 result_phi->init_req(inline_path, result);
6628
6629 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6630 clear_upper_avx();
6631 }
6632 }
6633 }
6634
6635 if (call_stub_path != nullptr) {
6636 set_control(call_stub_path);
6637
6638 Node* call = make_runtime_call(RC_LEAF,
6639 OptoRuntime::vectorizedMismatch_Type(),
6640 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6641 obja_adr, objb_adr, length, scale);
6642
6643 exit_block->init_req(stub_path, control());
6644 memory_phi->init_req(stub_path, map()->memory());
6645 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6646 }
6647
6648 exit_block = _gvn.transform(exit_block);
6649 memory_phi = _gvn.transform(memory_phi);
6650 result_phi = _gvn.transform(result_phi);
6651
6652 record_for_igvn(exit_block);
6653 record_for_igvn(memory_phi);
6654 record_for_igvn(result_phi);
6655
6656 set_control(exit_block);
6657 set_all_memory(memory_phi);
6658 set_result(result_phi);
6659
6660 return true;
6661 }
6662
6663 //------------------------------inline_vectorizedHashcode----------------------------
6664 bool LibraryCallKit::inline_vectorizedHashCode() {
6665 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6666
6667 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6668 Node* array = argument(0);
6669 Node* offset = argument(1);
6670 Node* length = argument(2);
6671 Node* initialValue = argument(3);
6672 Node* basic_type = argument(4);
6673
6674 if (basic_type == top()) {
6675 return false; // failed input validation
6676 }
6677
6678 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6679 if (!basic_type_t->is_con()) {
6680 return false; // Only intrinsify if mode argument is constant
6681 }
6682
6683 array = must_be_not_null(array, true);
6684
6685 BasicType bt = (BasicType)basic_type_t->get_con();
6686
6687 // Resolve address of first element
6688 Node* array_start = array_element_address(array, offset, bt);
6689
6690 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6691 array_start, length, initialValue, basic_type)));
6692 clear_upper_avx();
6693
6694 return true;
6695 }
6696
6697 /**
6698 * Calculate CRC32 for byte.
6699 * int java.util.zip.CRC32.update(int crc, int b)
6700 */
6701 bool LibraryCallKit::inline_updateCRC32() {
6702 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6703 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6704 // no receiver since it is static method
6705 Node* crc = argument(0); // type: int
6706 Node* b = argument(1); // type: int
6707
6708 /*
6709 * int c = ~ crc;
6710 * b = timesXtoThe32[(b ^ c) & 0xFF];
6711 * b = b ^ (c >>> 8);
6712 * crc = ~b;
6713 */
6714
6715 Node* M1 = intcon(-1);
6716 crc = _gvn.transform(new XorINode(crc, M1));
6717 Node* result = _gvn.transform(new XorINode(crc, b));
6718 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6719
6720 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6721 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6722 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6723 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6724
6725 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6726 result = _gvn.transform(new XorINode(crc, result));
6727 result = _gvn.transform(new XorINode(result, M1));
6728 set_result(result);
6729 return true;
6730 }
6731
6732 /**
6733 * Calculate CRC32 for byte[] array.
6734 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6735 */
6736 bool LibraryCallKit::inline_updateBytesCRC32() {
6737 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6738 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6739 // no receiver since it is static method
6740 Node* crc = argument(0); // type: int
6741 Node* src = argument(1); // type: oop
6742 Node* offset = argument(2); // type: int
6743 Node* length = argument(3); // type: int
6744
6745 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6746 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6747 // failed array check
6748 return false;
6749 }
6750
6751 // Figure out the size and type of the elements we will be copying.
6752 BasicType src_elem = src_type->elem()->array_element_basic_type();
6753 if (src_elem != T_BYTE) {
6754 return false;
6755 }
6756
6757 // 'src_start' points to src array + scaled offset
6758 src = must_be_not_null(src, true);
6759 Node* src_start = array_element_address(src, offset, src_elem);
6760
6761 // We assume that range check is done by caller.
6762 // TODO: generate range check (offset+length < src.length) in debug VM.
6763
6764 // Call the stub.
6765 address stubAddr = StubRoutines::updateBytesCRC32();
6766 const char *stubName = "updateBytesCRC32";
6767
6768 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6769 stubAddr, stubName, TypePtr::BOTTOM,
6770 crc, src_start, length);
6771 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6772 set_result(result);
6773 return true;
6774 }
6775
6776 /**
6777 * Calculate CRC32 for ByteBuffer.
6778 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6779 */
6780 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6781 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6782 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6783 // no receiver since it is static method
6784 Node* crc = argument(0); // type: int
6785 Node* src = argument(1); // type: long
6786 Node* offset = argument(3); // type: int
6787 Node* length = argument(4); // type: int
6788
6789 src = ConvL2X(src); // adjust Java long to machine word
6790 Node* base = _gvn.transform(new CastX2PNode(src));
6791 offset = ConvI2X(offset);
6792
6793 // 'src_start' points to src array + scaled offset
6794 Node* src_start = basic_plus_adr(top(), base, offset);
6795
6796 // Call the stub.
6797 address stubAddr = StubRoutines::updateBytesCRC32();
6798 const char *stubName = "updateBytesCRC32";
6799
6800 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6801 stubAddr, stubName, TypePtr::BOTTOM,
6802 crc, src_start, length);
6803 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6804 set_result(result);
6805 return true;
6806 }
6807
6808 //------------------------------get_table_from_crc32c_class-----------------------
6809 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6810 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6811 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6812
6813 return table;
6814 }
6815
6816 //------------------------------inline_updateBytesCRC32C-----------------------
6817 //
6818 // Calculate CRC32C for byte[] array.
6819 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6820 //
6821 bool LibraryCallKit::inline_updateBytesCRC32C() {
6822 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6823 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6824 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6825 // no receiver since it is a static method
6826 Node* crc = argument(0); // type: int
6827 Node* src = argument(1); // type: oop
6828 Node* offset = argument(2); // type: int
6829 Node* end = argument(3); // type: int
6830
6831 Node* length = _gvn.transform(new SubINode(end, offset));
6832
6833 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6834 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6835 // failed array check
6836 return false;
6837 }
6838
6839 // Figure out the size and type of the elements we will be copying.
6840 BasicType src_elem = src_type->elem()->array_element_basic_type();
6841 if (src_elem != T_BYTE) {
6842 return false;
6843 }
6844
6845 // 'src_start' points to src array + scaled offset
6846 src = must_be_not_null(src, true);
6847 Node* src_start = array_element_address(src, offset, src_elem);
6848
6849 // static final int[] byteTable in class CRC32C
6850 Node* table = get_table_from_crc32c_class(callee()->holder());
6851 table = must_be_not_null(table, true);
6852 Node* table_start = array_element_address(table, intcon(0), T_INT);
6853
6854 // We assume that range check is done by caller.
6855 // TODO: generate range check (offset+length < src.length) in debug VM.
6856
6857 // Call the stub.
6858 address stubAddr = StubRoutines::updateBytesCRC32C();
6859 const char *stubName = "updateBytesCRC32C";
6860
6861 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6862 stubAddr, stubName, TypePtr::BOTTOM,
6863 crc, src_start, length, table_start);
6864 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6865 set_result(result);
6866 return true;
6867 }
6868
6869 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6870 //
6871 // Calculate CRC32C for DirectByteBuffer.
6872 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6873 //
6874 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6875 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6876 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6877 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6878 // no receiver since it is a static method
6879 Node* crc = argument(0); // type: int
6880 Node* src = argument(1); // type: long
6881 Node* offset = argument(3); // type: int
6882 Node* end = argument(4); // type: int
6883
6884 Node* length = _gvn.transform(new SubINode(end, offset));
6885
6886 src = ConvL2X(src); // adjust Java long to machine word
6887 Node* base = _gvn.transform(new CastX2PNode(src));
6888 offset = ConvI2X(offset);
6889
6890 // 'src_start' points to src array + scaled offset
6891 Node* src_start = basic_plus_adr(top(), base, offset);
6892
6893 // static final int[] byteTable in class CRC32C
6894 Node* table = get_table_from_crc32c_class(callee()->holder());
6895 table = must_be_not_null(table, true);
6896 Node* table_start = array_element_address(table, intcon(0), T_INT);
6897
6898 // Call the stub.
6899 address stubAddr = StubRoutines::updateBytesCRC32C();
6900 const char *stubName = "updateBytesCRC32C";
6901
6902 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6903 stubAddr, stubName, TypePtr::BOTTOM,
6904 crc, src_start, length, table_start);
6905 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6906 set_result(result);
6907 return true;
6908 }
6909
6910 //------------------------------inline_updateBytesAdler32----------------------
6911 //
6912 // Calculate Adler32 checksum for byte[] array.
6913 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6914 //
6915 bool LibraryCallKit::inline_updateBytesAdler32() {
6916 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6917 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6918 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6919 // no receiver since it is static method
6920 Node* crc = argument(0); // type: int
6921 Node* src = argument(1); // type: oop
6922 Node* offset = argument(2); // type: int
6923 Node* length = argument(3); // type: int
6924
6925 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6926 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6927 // failed array check
6928 return false;
6929 }
6930
6931 // Figure out the size and type of the elements we will be copying.
6932 BasicType src_elem = src_type->elem()->array_element_basic_type();
6933 if (src_elem != T_BYTE) {
6934 return false;
6935 }
6936
6937 // 'src_start' points to src array + scaled offset
6938 Node* src_start = array_element_address(src, offset, src_elem);
6939
6940 // We assume that range check is done by caller.
6941 // TODO: generate range check (offset+length < src.length) in debug VM.
6942
6943 // Call the stub.
6944 address stubAddr = StubRoutines::updateBytesAdler32();
6945 const char *stubName = "updateBytesAdler32";
6946
6947 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6948 stubAddr, stubName, TypePtr::BOTTOM,
6949 crc, src_start, length);
6950 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6951 set_result(result);
6952 return true;
6953 }
6954
6955 //------------------------------inline_updateByteBufferAdler32---------------
6956 //
6957 // Calculate Adler32 checksum for DirectByteBuffer.
6958 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6959 //
6960 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6961 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6962 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6963 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6964 // no receiver since it is static method
6965 Node* crc = argument(0); // type: int
6966 Node* src = argument(1); // type: long
6967 Node* offset = argument(3); // type: int
6968 Node* length = argument(4); // type: int
6969
6970 src = ConvL2X(src); // adjust Java long to machine word
6971 Node* base = _gvn.transform(new CastX2PNode(src));
6972 offset = ConvI2X(offset);
6973
6974 // 'src_start' points to src array + scaled offset
6975 Node* src_start = basic_plus_adr(top(), base, offset);
6976
6977 // Call the stub.
6978 address stubAddr = StubRoutines::updateBytesAdler32();
6979 const char *stubName = "updateBytesAdler32";
6980
6981 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6982 stubAddr, stubName, TypePtr::BOTTOM,
6983 crc, src_start, length);
6984
6985 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6986 set_result(result);
6987 return true;
6988 }
6989
6990 //----------------------------inline_reference_get0----------------------------
6991 // public T java.lang.ref.Reference.get();
6992 bool LibraryCallKit::inline_reference_get0() {
6993 const int referent_offset = java_lang_ref_Reference::referent_offset();
6994
6995 // Get the argument:
6996 Node* reference_obj = null_check_receiver();
6997 if (stopped()) return true;
6998
6999 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7000 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7001 decorators, /*is_static*/ false, nullptr);
7002 if (result == nullptr) return false;
7003
7004 // Add memory barrier to prevent commoning reads from this field
7005 // across safepoint since GC can change its value.
7006 insert_mem_bar(Op_MemBarCPUOrder);
7007
7008 set_result(result);
7009 return true;
7010 }
7011
7012 //----------------------------inline_reference_refersTo0----------------------------
7013 // bool java.lang.ref.Reference.refersTo0();
7014 // bool java.lang.ref.PhantomReference.refersTo0();
7015 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7016 // Get arguments:
7017 Node* reference_obj = null_check_receiver();
7018 Node* other_obj = argument(1);
7019 if (stopped()) return true;
7020
7021 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7022 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7023 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7024 decorators, /*is_static*/ false, nullptr);
7025 if (referent == nullptr) return false;
7026
7027 // Add memory barrier to prevent commoning reads from this field
7028 // across safepoint since GC can change its value.
7029 insert_mem_bar(Op_MemBarCPUOrder);
7030
7031 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7032 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7033 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7034
7035 RegionNode* region = new RegionNode(3);
7036 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7037
7038 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7039 region->init_req(1, if_true);
7040 phi->init_req(1, intcon(1));
7041
7042 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7043 region->init_req(2, if_false);
7044 phi->init_req(2, intcon(0));
7045
7046 set_control(_gvn.transform(region));
7047 record_for_igvn(region);
7048 set_result(_gvn.transform(phi));
7049 return true;
7050 }
7051
7052 //----------------------------inline_reference_clear0----------------------------
7053 // void java.lang.ref.Reference.clear0();
7054 // void java.lang.ref.PhantomReference.clear0();
7055 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7056 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7057
7058 // Get arguments
7059 Node* reference_obj = null_check_receiver();
7060 if (stopped()) return true;
7061
7062 // Common access parameters
7063 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7064 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7065 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7066 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7067 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7068
7069 Node* referent = access_load_at(reference_obj,
7070 referent_field_addr,
7071 referent_field_addr_type,
7072 val_type,
7073 T_OBJECT,
7074 decorators);
7075
7076 IdealKit ideal(this);
7077 #define __ ideal.
7078 __ if_then(referent, BoolTest::ne, null());
7079 sync_kit(ideal);
7080 access_store_at(reference_obj,
7081 referent_field_addr,
7082 referent_field_addr_type,
7083 null(),
7084 val_type,
7085 T_OBJECT,
7086 decorators);
7087 __ sync_kit(this);
7088 __ end_if();
7089 final_sync(ideal);
7090 #undef __
7091
7092 return true;
7093 }
7094
7095 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7096 DecoratorSet decorators, bool is_static,
7097 ciInstanceKlass* fromKls) {
7098 if (fromKls == nullptr) {
7099 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7100 assert(tinst != nullptr, "obj is null");
7101 assert(tinst->is_loaded(), "obj is not loaded");
7102 fromKls = tinst->instance_klass();
7103 } else {
7104 assert(is_static, "only for static field access");
7105 }
7106 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7107 ciSymbol::make(fieldTypeString),
7108 is_static);
7109
7110 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7111 if (field == nullptr) return (Node *) nullptr;
7112
7113 if (is_static) {
7114 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7115 fromObj = makecon(tip);
7116 }
7117
7118 // Next code copied from Parse::do_get_xxx():
7119
7120 // Compute address and memory type.
7121 int offset = field->offset_in_bytes();
7122 bool is_vol = field->is_volatile();
7123 ciType* field_klass = field->type();
7124 assert(field_klass->is_loaded(), "should be loaded");
7125 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7126 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7127 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7128 "slice of address and input slice don't match");
7129 BasicType bt = field->layout_type();
7130
7131 // Build the resultant type of the load
7132 const Type *type;
7133 if (bt == T_OBJECT) {
7134 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7135 } else {
7136 type = Type::get_const_basic_type(bt);
7137 }
7138
7139 if (is_vol) {
7140 decorators |= MO_SEQ_CST;
7141 }
7142
7143 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7144 }
7145
7146 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7147 bool is_exact /* true */, bool is_static /* false */,
7148 ciInstanceKlass * fromKls /* nullptr */) {
7149 if (fromKls == nullptr) {
7150 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7151 assert(tinst != nullptr, "obj is null");
7152 assert(tinst->is_loaded(), "obj is not loaded");
7153 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7154 fromKls = tinst->instance_klass();
7155 }
7156 else {
7157 assert(is_static, "only for static field access");
7158 }
7159 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7160 ciSymbol::make(fieldTypeString),
7161 is_static);
7162
7163 assert(field != nullptr, "undefined field");
7164 assert(!field->is_volatile(), "not defined for volatile fields");
7165
7166 if (is_static) {
7167 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7168 fromObj = makecon(tip);
7169 }
7170
7171 // Next code copied from Parse::do_get_xxx():
7172
7173 // Compute address and memory type.
7174 int offset = field->offset_in_bytes();
7175 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7176
7177 return adr;
7178 }
7179
7180 //------------------------------inline_aescrypt_Block-----------------------
7181 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7182 address stubAddr = nullptr;
7183 const char *stubName;
7184 bool is_decrypt = false;
7185 assert(UseAES, "need AES instruction support");
7186
7187 switch(id) {
7188 case vmIntrinsics::_aescrypt_encryptBlock:
7189 stubAddr = StubRoutines::aescrypt_encryptBlock();
7190 stubName = "aescrypt_encryptBlock";
7191 break;
7192 case vmIntrinsics::_aescrypt_decryptBlock:
7193 stubAddr = StubRoutines::aescrypt_decryptBlock();
7194 stubName = "aescrypt_decryptBlock";
7195 is_decrypt = true;
7196 break;
7197 default:
7198 break;
7199 }
7200 if (stubAddr == nullptr) return false;
7201
7202 Node* aescrypt_object = argument(0);
7203 Node* src = argument(1);
7204 Node* src_offset = argument(2);
7205 Node* dest = argument(3);
7206 Node* dest_offset = argument(4);
7207
7208 src = must_be_not_null(src, true);
7209 dest = must_be_not_null(dest, true);
7210
7211 // (1) src and dest are arrays.
7212 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7213 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7214 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7215 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7216
7217 // for the quick and dirty code we will skip all the checks.
7218 // we are just trying to get the call to be generated.
7219 Node* src_start = src;
7220 Node* dest_start = dest;
7221 if (src_offset != nullptr || dest_offset != nullptr) {
7222 assert(src_offset != nullptr && dest_offset != nullptr, "");
7223 src_start = array_element_address(src, src_offset, T_BYTE);
7224 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7225 }
7226
7227 // now need to get the start of its expanded key array
7228 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7229 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7230 if (k_start == nullptr) return false;
7231
7232 // Call the stub.
7233 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7234 stubAddr, stubName, TypePtr::BOTTOM,
7235 src_start, dest_start, k_start);
7236
7237 return true;
7238 }
7239
7240 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7241 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7242 address stubAddr = nullptr;
7243 const char *stubName = nullptr;
7244 bool is_decrypt = false;
7245 assert(UseAES, "need AES instruction support");
7246
7247 switch(id) {
7248 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7249 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7250 stubName = "cipherBlockChaining_encryptAESCrypt";
7251 break;
7252 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7253 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7254 stubName = "cipherBlockChaining_decryptAESCrypt";
7255 is_decrypt = true;
7256 break;
7257 default:
7258 break;
7259 }
7260 if (stubAddr == nullptr) return false;
7261
7262 Node* cipherBlockChaining_object = argument(0);
7263 Node* src = argument(1);
7264 Node* src_offset = argument(2);
7265 Node* len = argument(3);
7266 Node* dest = argument(4);
7267 Node* dest_offset = argument(5);
7268
7269 src = must_be_not_null(src, false);
7270 dest = must_be_not_null(dest, false);
7271
7272 // (1) src and dest are arrays.
7273 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7274 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7275 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7276 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7277
7278 // checks are the responsibility of the caller
7279 Node* src_start = src;
7280 Node* dest_start = dest;
7281 if (src_offset != nullptr || dest_offset != nullptr) {
7282 assert(src_offset != nullptr && dest_offset != nullptr, "");
7283 src_start = array_element_address(src, src_offset, T_BYTE);
7284 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7285 }
7286
7287 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7288 // (because of the predicated logic executed earlier).
7289 // so we cast it here safely.
7290 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7291
7292 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7293 if (embeddedCipherObj == nullptr) return false;
7294
7295 // cast it to what we know it will be at runtime
7296 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7297 assert(tinst != nullptr, "CBC obj is null");
7298 assert(tinst->is_loaded(), "CBC obj is not loaded");
7299 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7300 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7301
7302 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7303 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7304 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7305 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7306 aescrypt_object = _gvn.transform(aescrypt_object);
7307
7308 // we need to get the start of the aescrypt_object's expanded key array
7309 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7310 if (k_start == nullptr) return false;
7311
7312 // similarly, get the start address of the r vector
7313 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7314 if (objRvec == nullptr) return false;
7315 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7316
7317 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7318 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7319 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7320 stubAddr, stubName, TypePtr::BOTTOM,
7321 src_start, dest_start, k_start, r_start, len);
7322
7323 // return cipher length (int)
7324 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7325 set_result(retvalue);
7326 return true;
7327 }
7328
7329 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7330 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7331 address stubAddr = nullptr;
7332 const char *stubName = nullptr;
7333 bool is_decrypt = false;
7334 assert(UseAES, "need AES instruction support");
7335
7336 switch (id) {
7337 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7338 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7339 stubName = "electronicCodeBook_encryptAESCrypt";
7340 break;
7341 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7342 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7343 stubName = "electronicCodeBook_decryptAESCrypt";
7344 is_decrypt = true;
7345 break;
7346 default:
7347 break;
7348 }
7349
7350 if (stubAddr == nullptr) return false;
7351
7352 Node* electronicCodeBook_object = argument(0);
7353 Node* src = argument(1);
7354 Node* src_offset = argument(2);
7355 Node* len = argument(3);
7356 Node* dest = argument(4);
7357 Node* dest_offset = argument(5);
7358
7359 // (1) src and dest are arrays.
7360 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7361 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7362 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7363 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7364
7365 // checks are the responsibility of the caller
7366 Node* src_start = src;
7367 Node* dest_start = dest;
7368 if (src_offset != nullptr || dest_offset != nullptr) {
7369 assert(src_offset != nullptr && dest_offset != nullptr, "");
7370 src_start = array_element_address(src, src_offset, T_BYTE);
7371 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7372 }
7373
7374 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7375 // (because of the predicated logic executed earlier).
7376 // so we cast it here safely.
7377 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7378
7379 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7380 if (embeddedCipherObj == nullptr) return false;
7381
7382 // cast it to what we know it will be at runtime
7383 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7384 assert(tinst != nullptr, "ECB obj is null");
7385 assert(tinst->is_loaded(), "ECB obj is not loaded");
7386 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7387 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7388
7389 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7390 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7391 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7392 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7393 aescrypt_object = _gvn.transform(aescrypt_object);
7394
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, is_decrypt);
7397 if (k_start == nullptr) return false;
7398
7399 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7400 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7401 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7402 stubAddr, stubName, TypePtr::BOTTOM,
7403 src_start, dest_start, k_start, len);
7404
7405 // return cipher length (int)
7406 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7407 set_result(retvalue);
7408 return true;
7409 }
7410
7411 //------------------------------inline_counterMode_AESCrypt-----------------------
7412 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7413 assert(UseAES, "need AES instruction support");
7414 if (!UseAESCTRIntrinsics) return false;
7415
7416 address stubAddr = nullptr;
7417 const char *stubName = nullptr;
7418 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7419 stubAddr = StubRoutines::counterMode_AESCrypt();
7420 stubName = "counterMode_AESCrypt";
7421 }
7422 if (stubAddr == nullptr) return false;
7423
7424 Node* counterMode_object = argument(0);
7425 Node* src = argument(1);
7426 Node* src_offset = argument(2);
7427 Node* len = argument(3);
7428 Node* dest = argument(4);
7429 Node* dest_offset = argument(5);
7430
7431 // (1) src and dest are arrays.
7432 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7433 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7434 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7435 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7436
7437 // checks are the responsibility of the caller
7438 Node* src_start = src;
7439 Node* dest_start = dest;
7440 if (src_offset != nullptr || dest_offset != nullptr) {
7441 assert(src_offset != nullptr && dest_offset != nullptr, "");
7442 src_start = array_element_address(src, src_offset, T_BYTE);
7443 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7444 }
7445
7446 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7447 // (because of the predicated logic executed earlier).
7448 // so we cast it here safely.
7449 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7450 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7451 if (embeddedCipherObj == nullptr) return false;
7452 // cast it to what we know it will be at runtime
7453 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7454 assert(tinst != nullptr, "CTR obj is null");
7455 assert(tinst->is_loaded(), "CTR obj is not loaded");
7456 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7457 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7458 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7459 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7460 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7461 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7462 aescrypt_object = _gvn.transform(aescrypt_object);
7463 // we need to get the start of the aescrypt_object's expanded key array
7464 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7465 if (k_start == nullptr) return false;
7466 // similarly, get the start address of the r vector
7467 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7468 if (obj_counter == nullptr) return false;
7469 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7470
7471 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7472 if (saved_encCounter == nullptr) return false;
7473 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7474 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7475
7476 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7477 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7478 OptoRuntime::counterMode_aescrypt_Type(),
7479 stubAddr, stubName, TypePtr::BOTTOM,
7480 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7481
7482 // return cipher length (int)
7483 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7484 set_result(retvalue);
7485 return true;
7486 }
7487
7488 //------------------------------get_key_start_from_aescrypt_object-----------------------
7489 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7490 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7491 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7492 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7493 // The following platform specific stubs of encryption and decryption use the same round keys.
7494 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7495 bool use_decryption_key = false;
7496 #else
7497 bool use_decryption_key = is_decrypt;
7498 #endif
7499 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7500 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7501 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7502
7503 // now have the array, need to get the start address of the selected key array
7504 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7505 return k_start;
7506 }
7507
7508 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7509 // Return node representing slow path of predicate check.
7510 // the pseudo code we want to emulate with this predicate is:
7511 // for encryption:
7512 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7513 // for decryption:
7514 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7515 // note cipher==plain is more conservative than the original java code but that's OK
7516 //
7517 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7518 // The receiver was checked for null already.
7519 Node* objCBC = argument(0);
7520
7521 Node* src = argument(1);
7522 Node* dest = argument(4);
7523
7524 // Load embeddedCipher field of CipherBlockChaining object.
7525 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7526
7527 // get AESCrypt klass for instanceOf check
7528 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7529 // will have same classloader as CipherBlockChaining object
7530 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7531 assert(tinst != nullptr, "CBCobj is null");
7532 assert(tinst->is_loaded(), "CBCobj is not loaded");
7533
7534 // we want to do an instanceof comparison against the AESCrypt class
7535 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7536 if (!klass_AESCrypt->is_loaded()) {
7537 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7538 Node* ctrl = control();
7539 set_control(top()); // no regular fast path
7540 return ctrl;
7541 }
7542
7543 src = must_be_not_null(src, true);
7544 dest = must_be_not_null(dest, true);
7545
7546 // Resolve oops to stable for CmpP below.
7547 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7548
7549 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7550 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7551 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7552
7553 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7554
7555 // for encryption, we are done
7556 if (!decrypting)
7557 return instof_false; // even if it is null
7558
7559 // for decryption, we need to add a further check to avoid
7560 // taking the intrinsic path when cipher and plain are the same
7561 // see the original java code for why.
7562 RegionNode* region = new RegionNode(3);
7563 region->init_req(1, instof_false);
7564
7565 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7566 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7567 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7568 region->init_req(2, src_dest_conjoint);
7569
7570 record_for_igvn(region);
7571 return _gvn.transform(region);
7572 }
7573
7574 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7575 // Return node representing slow path of predicate check.
7576 // the pseudo code we want to emulate with this predicate is:
7577 // for encryption:
7578 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7579 // for decryption:
7580 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7581 // note cipher==plain is more conservative than the original java code but that's OK
7582 //
7583 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7584 // The receiver was checked for null already.
7585 Node* objECB = argument(0);
7586
7587 // Load embeddedCipher field of ElectronicCodeBook object.
7588 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7589
7590 // get AESCrypt klass for instanceOf check
7591 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7592 // will have same classloader as ElectronicCodeBook object
7593 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7594 assert(tinst != nullptr, "ECBobj is null");
7595 assert(tinst->is_loaded(), "ECBobj is not loaded");
7596
7597 // we want to do an instanceof comparison against the AESCrypt class
7598 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7599 if (!klass_AESCrypt->is_loaded()) {
7600 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7601 Node* ctrl = control();
7602 set_control(top()); // no regular fast path
7603 return ctrl;
7604 }
7605 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7606
7607 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7608 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7609 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7610
7611 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7612
7613 // for encryption, we are done
7614 if (!decrypting)
7615 return instof_false; // even if it is null
7616
7617 // for decryption, we need to add a further check to avoid
7618 // taking the intrinsic path when cipher and plain are the same
7619 // see the original java code for why.
7620 RegionNode* region = new RegionNode(3);
7621 region->init_req(1, instof_false);
7622 Node* src = argument(1);
7623 Node* dest = argument(4);
7624 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7625 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7626 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7627 region->init_req(2, src_dest_conjoint);
7628
7629 record_for_igvn(region);
7630 return _gvn.transform(region);
7631 }
7632
7633 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7634 // Return node representing slow path of predicate check.
7635 // the pseudo code we want to emulate with this predicate is:
7636 // for encryption:
7637 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7638 // for decryption:
7639 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7640 // note cipher==plain is more conservative than the original java code but that's OK
7641 //
7642
7643 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7644 // The receiver was checked for null already.
7645 Node* objCTR = argument(0);
7646
7647 // Load embeddedCipher field of CipherBlockChaining object.
7648 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7649
7650 // get AESCrypt klass for instanceOf check
7651 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7652 // will have same classloader as CipherBlockChaining object
7653 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7654 assert(tinst != nullptr, "CTRobj is null");
7655 assert(tinst->is_loaded(), "CTRobj is not loaded");
7656
7657 // we want to do an instanceof comparison against the AESCrypt class
7658 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7659 if (!klass_AESCrypt->is_loaded()) {
7660 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7661 Node* ctrl = control();
7662 set_control(top()); // no regular fast path
7663 return ctrl;
7664 }
7665
7666 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7667 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7668 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7669 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7670 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7671
7672 return instof_false; // even if it is null
7673 }
7674
7675 //------------------------------inline_ghash_processBlocks
7676 bool LibraryCallKit::inline_ghash_processBlocks() {
7677 address stubAddr;
7678 const char *stubName;
7679 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7680
7681 stubAddr = StubRoutines::ghash_processBlocks();
7682 stubName = "ghash_processBlocks";
7683
7684 Node* data = argument(0);
7685 Node* offset = argument(1);
7686 Node* len = argument(2);
7687 Node* state = argument(3);
7688 Node* subkeyH = argument(4);
7689
7690 state = must_be_not_null(state, true);
7691 subkeyH = must_be_not_null(subkeyH, true);
7692 data = must_be_not_null(data, true);
7693
7694 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7695 assert(state_start, "state is null");
7696 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7697 assert(subkeyH_start, "subkeyH is null");
7698 Node* data_start = array_element_address(data, offset, T_BYTE);
7699 assert(data_start, "data is null");
7700
7701 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7702 OptoRuntime::ghash_processBlocks_Type(),
7703 stubAddr, stubName, TypePtr::BOTTOM,
7704 state_start, subkeyH_start, data_start, len);
7705 return true;
7706 }
7707
7708 //------------------------------inline_chacha20Block
7709 bool LibraryCallKit::inline_chacha20Block() {
7710 address stubAddr;
7711 const char *stubName;
7712 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7713
7714 stubAddr = StubRoutines::chacha20Block();
7715 stubName = "chacha20Block";
7716
7717 Node* state = argument(0);
7718 Node* result = argument(1);
7719
7720 state = must_be_not_null(state, true);
7721 result = must_be_not_null(result, true);
7722
7723 Node* state_start = array_element_address(state, intcon(0), T_INT);
7724 assert(state_start, "state is null");
7725 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7726 assert(result_start, "result is null");
7727
7728 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7729 OptoRuntime::chacha20Block_Type(),
7730 stubAddr, stubName, TypePtr::BOTTOM,
7731 state_start, result_start);
7732 // return key stream length (int)
7733 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7734 set_result(retvalue);
7735 return true;
7736 }
7737
7738 //------------------------------inline_kyberNtt
7739 bool LibraryCallKit::inline_kyberNtt() {
7740 address stubAddr;
7741 const char *stubName;
7742 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7743 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7744
7745 stubAddr = StubRoutines::kyberNtt();
7746 stubName = "kyberNtt";
7747 if (!stubAddr) return false;
7748
7749 Node* coeffs = argument(0);
7750 Node* ntt_zetas = argument(1);
7751
7752 coeffs = must_be_not_null(coeffs, true);
7753 ntt_zetas = must_be_not_null(ntt_zetas, true);
7754
7755 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7756 assert(coeffs_start, "coeffs is null");
7757 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7758 assert(ntt_zetas_start, "ntt_zetas is null");
7759 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7760 OptoRuntime::kyberNtt_Type(),
7761 stubAddr, stubName, TypePtr::BOTTOM,
7762 coeffs_start, ntt_zetas_start);
7763 // return an int
7764 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7765 set_result(retvalue);
7766 return true;
7767 }
7768
7769 //------------------------------inline_kyberInverseNtt
7770 bool LibraryCallKit::inline_kyberInverseNtt() {
7771 address stubAddr;
7772 const char *stubName;
7773 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7774 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7775
7776 stubAddr = StubRoutines::kyberInverseNtt();
7777 stubName = "kyberInverseNtt";
7778 if (!stubAddr) return false;
7779
7780 Node* coeffs = argument(0);
7781 Node* zetas = argument(1);
7782
7783 coeffs = must_be_not_null(coeffs, true);
7784 zetas = must_be_not_null(zetas, true);
7785
7786 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7787 assert(coeffs_start, "coeffs is null");
7788 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7789 assert(zetas_start, "inverseNtt_zetas is null");
7790 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7791 OptoRuntime::kyberInverseNtt_Type(),
7792 stubAddr, stubName, TypePtr::BOTTOM,
7793 coeffs_start, zetas_start);
7794
7795 // return an int
7796 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7797 set_result(retvalue);
7798 return true;
7799 }
7800
7801 //------------------------------inline_kyberNttMult
7802 bool LibraryCallKit::inline_kyberNttMult() {
7803 address stubAddr;
7804 const char *stubName;
7805 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7806 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7807
7808 stubAddr = StubRoutines::kyberNttMult();
7809 stubName = "kyberNttMult";
7810 if (!stubAddr) return false;
7811
7812 Node* result = argument(0);
7813 Node* ntta = argument(1);
7814 Node* nttb = argument(2);
7815 Node* zetas = argument(3);
7816
7817 result = must_be_not_null(result, true);
7818 ntta = must_be_not_null(ntta, true);
7819 nttb = must_be_not_null(nttb, true);
7820 zetas = must_be_not_null(zetas, true);
7821
7822 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7823 assert(result_start, "result is null");
7824 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7825 assert(ntta_start, "ntta is null");
7826 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7827 assert(nttb_start, "nttb is null");
7828 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7829 assert(zetas_start, "nttMult_zetas is null");
7830 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7831 OptoRuntime::kyberNttMult_Type(),
7832 stubAddr, stubName, TypePtr::BOTTOM,
7833 result_start, ntta_start, nttb_start,
7834 zetas_start);
7835
7836 // return an int
7837 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7838 set_result(retvalue);
7839
7840 return true;
7841 }
7842
7843 //------------------------------inline_kyberAddPoly_2
7844 bool LibraryCallKit::inline_kyberAddPoly_2() {
7845 address stubAddr;
7846 const char *stubName;
7847 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7848 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7849
7850 stubAddr = StubRoutines::kyberAddPoly_2();
7851 stubName = "kyberAddPoly_2";
7852 if (!stubAddr) return false;
7853
7854 Node* result = argument(0);
7855 Node* a = argument(1);
7856 Node* b = argument(2);
7857
7858 result = must_be_not_null(result, true);
7859 a = must_be_not_null(a, true);
7860 b = must_be_not_null(b, true);
7861
7862 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7863 assert(result_start, "result is null");
7864 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7865 assert(a_start, "a is null");
7866 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7867 assert(b_start, "b is null");
7868 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7869 OptoRuntime::kyberAddPoly_2_Type(),
7870 stubAddr, stubName, TypePtr::BOTTOM,
7871 result_start, a_start, b_start);
7872 // return an int
7873 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7874 set_result(retvalue);
7875 return true;
7876 }
7877
7878 //------------------------------inline_kyberAddPoly_3
7879 bool LibraryCallKit::inline_kyberAddPoly_3() {
7880 address stubAddr;
7881 const char *stubName;
7882 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7883 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7884
7885 stubAddr = StubRoutines::kyberAddPoly_3();
7886 stubName = "kyberAddPoly_3";
7887 if (!stubAddr) return false;
7888
7889 Node* result = argument(0);
7890 Node* a = argument(1);
7891 Node* b = argument(2);
7892 Node* c = argument(3);
7893
7894 result = must_be_not_null(result, true);
7895 a = must_be_not_null(a, true);
7896 b = must_be_not_null(b, true);
7897 c = must_be_not_null(c, true);
7898
7899 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7900 assert(result_start, "result is null");
7901 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7902 assert(a_start, "a is null");
7903 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7904 assert(b_start, "b is null");
7905 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7906 assert(c_start, "c is null");
7907 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7908 OptoRuntime::kyberAddPoly_3_Type(),
7909 stubAddr, stubName, TypePtr::BOTTOM,
7910 result_start, a_start, b_start, c_start);
7911 // return an int
7912 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7913 set_result(retvalue);
7914 return true;
7915 }
7916
7917 //------------------------------inline_kyber12To16
7918 bool LibraryCallKit::inline_kyber12To16() {
7919 address stubAddr;
7920 const char *stubName;
7921 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7922 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7923
7924 stubAddr = StubRoutines::kyber12To16();
7925 stubName = "kyber12To16";
7926 if (!stubAddr) return false;
7927
7928 Node* condensed = argument(0);
7929 Node* condensedOffs = argument(1);
7930 Node* parsed = argument(2);
7931 Node* parsedLength = argument(3);
7932
7933 condensed = must_be_not_null(condensed, true);
7934 parsed = must_be_not_null(parsed, true);
7935
7936 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7937 assert(condensed_start, "condensed is null");
7938 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7939 assert(parsed_start, "parsed is null");
7940 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7941 OptoRuntime::kyber12To16_Type(),
7942 stubAddr, stubName, TypePtr::BOTTOM,
7943 condensed_start, condensedOffs, parsed_start, parsedLength);
7944 // return an int
7945 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7946 set_result(retvalue);
7947 return true;
7948
7949 }
7950
7951 //------------------------------inline_kyberBarrettReduce
7952 bool LibraryCallKit::inline_kyberBarrettReduce() {
7953 address stubAddr;
7954 const char *stubName;
7955 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7956 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7957
7958 stubAddr = StubRoutines::kyberBarrettReduce();
7959 stubName = "kyberBarrettReduce";
7960 if (!stubAddr) return false;
7961
7962 Node* coeffs = argument(0);
7963
7964 coeffs = must_be_not_null(coeffs, true);
7965
7966 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7967 assert(coeffs_start, "coeffs is null");
7968 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7969 OptoRuntime::kyberBarrettReduce_Type(),
7970 stubAddr, stubName, TypePtr::BOTTOM,
7971 coeffs_start);
7972 // return an int
7973 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7974 set_result(retvalue);
7975 return true;
7976 }
7977
7978 //------------------------------inline_dilithiumAlmostNtt
7979 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7980 address stubAddr;
7981 const char *stubName;
7982 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7983 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7984
7985 stubAddr = StubRoutines::dilithiumAlmostNtt();
7986 stubName = "dilithiumAlmostNtt";
7987 if (!stubAddr) return false;
7988
7989 Node* coeffs = argument(0);
7990 Node* ntt_zetas = argument(1);
7991
7992 coeffs = must_be_not_null(coeffs, true);
7993 ntt_zetas = must_be_not_null(ntt_zetas, true);
7994
7995 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7996 assert(coeffs_start, "coeffs is null");
7997 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7998 assert(ntt_zetas_start, "ntt_zetas is null");
7999 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8000 OptoRuntime::dilithiumAlmostNtt_Type(),
8001 stubAddr, stubName, TypePtr::BOTTOM,
8002 coeffs_start, ntt_zetas_start);
8003 // return an int
8004 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8005 set_result(retvalue);
8006 return true;
8007 }
8008
8009 //------------------------------inline_dilithiumAlmostInverseNtt
8010 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8011 address stubAddr;
8012 const char *stubName;
8013 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8014 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8015
8016 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8017 stubName = "dilithiumAlmostInverseNtt";
8018 if (!stubAddr) return false;
8019
8020 Node* coeffs = argument(0);
8021 Node* zetas = argument(1);
8022
8023 coeffs = must_be_not_null(coeffs, true);
8024 zetas = must_be_not_null(zetas, true);
8025
8026 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8027 assert(coeffs_start, "coeffs is null");
8028 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8029 assert(zetas_start, "inverseNtt_zetas is null");
8030 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8031 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8032 stubAddr, stubName, TypePtr::BOTTOM,
8033 coeffs_start, zetas_start);
8034 // return an int
8035 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8036 set_result(retvalue);
8037 return true;
8038 }
8039
8040 //------------------------------inline_dilithiumNttMult
8041 bool LibraryCallKit::inline_dilithiumNttMult() {
8042 address stubAddr;
8043 const char *stubName;
8044 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8045 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8046
8047 stubAddr = StubRoutines::dilithiumNttMult();
8048 stubName = "dilithiumNttMult";
8049 if (!stubAddr) return false;
8050
8051 Node* result = argument(0);
8052 Node* ntta = argument(1);
8053 Node* nttb = argument(2);
8054 Node* zetas = argument(3);
8055
8056 result = must_be_not_null(result, true);
8057 ntta = must_be_not_null(ntta, true);
8058 nttb = must_be_not_null(nttb, true);
8059 zetas = must_be_not_null(zetas, true);
8060
8061 Node* result_start = array_element_address(result, intcon(0), T_INT);
8062 assert(result_start, "result is null");
8063 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8064 assert(ntta_start, "ntta is null");
8065 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8066 assert(nttb_start, "nttb is null");
8067 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8068 OptoRuntime::dilithiumNttMult_Type(),
8069 stubAddr, stubName, TypePtr::BOTTOM,
8070 result_start, ntta_start, nttb_start);
8071
8072 // return an int
8073 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8074 set_result(retvalue);
8075
8076 return true;
8077 }
8078
8079 //------------------------------inline_dilithiumMontMulByConstant
8080 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8081 address stubAddr;
8082 const char *stubName;
8083 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8084 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8085
8086 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8087 stubName = "dilithiumMontMulByConstant";
8088 if (!stubAddr) return false;
8089
8090 Node* coeffs = argument(0);
8091 Node* constant = argument(1);
8092
8093 coeffs = must_be_not_null(coeffs, true);
8094
8095 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8096 assert(coeffs_start, "coeffs is null");
8097 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8098 OptoRuntime::dilithiumMontMulByConstant_Type(),
8099 stubAddr, stubName, TypePtr::BOTTOM,
8100 coeffs_start, constant);
8101
8102 // return an int
8103 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8104 set_result(retvalue);
8105 return true;
8106 }
8107
8108
8109 //------------------------------inline_dilithiumDecomposePoly
8110 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8111 address stubAddr;
8112 const char *stubName;
8113 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8114 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8115
8116 stubAddr = StubRoutines::dilithiumDecomposePoly();
8117 stubName = "dilithiumDecomposePoly";
8118 if (!stubAddr) return false;
8119
8120 Node* input = argument(0);
8121 Node* lowPart = argument(1);
8122 Node* highPart = argument(2);
8123 Node* twoGamma2 = argument(3);
8124 Node* multiplier = argument(4);
8125
8126 input = must_be_not_null(input, true);
8127 lowPart = must_be_not_null(lowPart, true);
8128 highPart = must_be_not_null(highPart, true);
8129
8130 Node* input_start = array_element_address(input, intcon(0), T_INT);
8131 assert(input_start, "input is null");
8132 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8133 assert(lowPart_start, "lowPart is null");
8134 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8135 assert(highPart_start, "highPart is null");
8136
8137 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8138 OptoRuntime::dilithiumDecomposePoly_Type(),
8139 stubAddr, stubName, TypePtr::BOTTOM,
8140 input_start, lowPart_start, highPart_start,
8141 twoGamma2, multiplier);
8142
8143 // return an int
8144 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8145 set_result(retvalue);
8146 return true;
8147 }
8148
8149 bool LibraryCallKit::inline_base64_encodeBlock() {
8150 address stubAddr;
8151 const char *stubName;
8152 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8153 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8154 stubAddr = StubRoutines::base64_encodeBlock();
8155 stubName = "encodeBlock";
8156
8157 if (!stubAddr) return false;
8158 Node* base64obj = argument(0);
8159 Node* src = argument(1);
8160 Node* offset = argument(2);
8161 Node* len = argument(3);
8162 Node* dest = argument(4);
8163 Node* dp = argument(5);
8164 Node* isURL = argument(6);
8165
8166 src = must_be_not_null(src, true);
8167 dest = must_be_not_null(dest, true);
8168
8169 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8170 assert(src_start, "source array is null");
8171 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8172 assert(dest_start, "destination array is null");
8173
8174 Node* base64 = make_runtime_call(RC_LEAF,
8175 OptoRuntime::base64_encodeBlock_Type(),
8176 stubAddr, stubName, TypePtr::BOTTOM,
8177 src_start, offset, len, dest_start, dp, isURL);
8178 return true;
8179 }
8180
8181 bool LibraryCallKit::inline_base64_decodeBlock() {
8182 address stubAddr;
8183 const char *stubName;
8184 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8185 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8186 stubAddr = StubRoutines::base64_decodeBlock();
8187 stubName = "decodeBlock";
8188
8189 if (!stubAddr) return false;
8190 Node* base64obj = argument(0);
8191 Node* src = argument(1);
8192 Node* src_offset = argument(2);
8193 Node* len = argument(3);
8194 Node* dest = argument(4);
8195 Node* dest_offset = argument(5);
8196 Node* isURL = argument(6);
8197 Node* isMIME = argument(7);
8198
8199 src = must_be_not_null(src, true);
8200 dest = must_be_not_null(dest, true);
8201
8202 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8203 assert(src_start, "source array is null");
8204 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8205 assert(dest_start, "destination array is null");
8206
8207 Node* call = make_runtime_call(RC_LEAF,
8208 OptoRuntime::base64_decodeBlock_Type(),
8209 stubAddr, stubName, TypePtr::BOTTOM,
8210 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8211 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8212 set_result(result);
8213 return true;
8214 }
8215
8216 bool LibraryCallKit::inline_poly1305_processBlocks() {
8217 address stubAddr;
8218 const char *stubName;
8219 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8220 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8221 stubAddr = StubRoutines::poly1305_processBlocks();
8222 stubName = "poly1305_processBlocks";
8223
8224 if (!stubAddr) return false;
8225 null_check_receiver(); // null-check receiver
8226 if (stopped()) return true;
8227
8228 Node* input = argument(1);
8229 Node* input_offset = argument(2);
8230 Node* len = argument(3);
8231 Node* alimbs = argument(4);
8232 Node* rlimbs = argument(5);
8233
8234 input = must_be_not_null(input, true);
8235 alimbs = must_be_not_null(alimbs, true);
8236 rlimbs = must_be_not_null(rlimbs, true);
8237
8238 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8239 assert(input_start, "input array is null");
8240 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8241 assert(acc_start, "acc array is null");
8242 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8243 assert(r_start, "r array is null");
8244
8245 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8246 OptoRuntime::poly1305_processBlocks_Type(),
8247 stubAddr, stubName, TypePtr::BOTTOM,
8248 input_start, len, acc_start, r_start);
8249 return true;
8250 }
8251
8252 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8253 address stubAddr;
8254 const char *stubName;
8255 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8256 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8257 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8258 stubName = "intpoly_montgomeryMult_P256";
8259
8260 if (!stubAddr) return false;
8261 null_check_receiver(); // null-check receiver
8262 if (stopped()) return true;
8263
8264 Node* a = argument(1);
8265 Node* b = argument(2);
8266 Node* r = argument(3);
8267
8268 a = must_be_not_null(a, true);
8269 b = must_be_not_null(b, true);
8270 r = must_be_not_null(r, true);
8271
8272 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8273 assert(a_start, "a array is null");
8274 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8275 assert(b_start, "b array is null");
8276 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8277 assert(r_start, "r array is null");
8278
8279 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8280 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8281 stubAddr, stubName, TypePtr::BOTTOM,
8282 a_start, b_start, r_start);
8283 return true;
8284 }
8285
8286 bool LibraryCallKit::inline_intpoly_assign() {
8287 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8288 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8289 const char *stubName = "intpoly_assign";
8290 address stubAddr = StubRoutines::intpoly_assign();
8291 if (!stubAddr) return false;
8292
8293 Node* set = argument(0);
8294 Node* a = argument(1);
8295 Node* b = argument(2);
8296 Node* arr_length = load_array_length(a);
8297
8298 a = must_be_not_null(a, true);
8299 b = must_be_not_null(b, true);
8300
8301 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8302 assert(a_start, "a array is null");
8303 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8304 assert(b_start, "b array is null");
8305
8306 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8307 OptoRuntime::intpoly_assign_Type(),
8308 stubAddr, stubName, TypePtr::BOTTOM,
8309 set, a_start, b_start, arr_length);
8310 return true;
8311 }
8312
8313 //------------------------------inline_digestBase_implCompress-----------------------
8314 //
8315 // Calculate MD5 for single-block byte[] array.
8316 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8317 //
8318 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8319 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8320 //
8321 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8322 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8323 //
8324 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8325 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8326 //
8327 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8328 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8329 //
8330 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8331 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8332
8333 Node* digestBase_obj = argument(0);
8334 Node* src = argument(1); // type oop
8335 Node* ofs = argument(2); // type int
8336
8337 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8338 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8339 // failed array check
8340 return false;
8341 }
8342 // Figure out the size and type of the elements we will be copying.
8343 BasicType src_elem = src_type->elem()->array_element_basic_type();
8344 if (src_elem != T_BYTE) {
8345 return false;
8346 }
8347 // 'src_start' points to src array + offset
8348 src = must_be_not_null(src, true);
8349 Node* src_start = array_element_address(src, ofs, src_elem);
8350 Node* state = nullptr;
8351 Node* block_size = nullptr;
8352 address stubAddr;
8353 const char *stubName;
8354
8355 switch(id) {
8356 case vmIntrinsics::_md5_implCompress:
8357 assert(UseMD5Intrinsics, "need MD5 instruction support");
8358 state = get_state_from_digest_object(digestBase_obj, T_INT);
8359 stubAddr = StubRoutines::md5_implCompress();
8360 stubName = "md5_implCompress";
8361 break;
8362 case vmIntrinsics::_sha_implCompress:
8363 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8364 state = get_state_from_digest_object(digestBase_obj, T_INT);
8365 stubAddr = StubRoutines::sha1_implCompress();
8366 stubName = "sha1_implCompress";
8367 break;
8368 case vmIntrinsics::_sha2_implCompress:
8369 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8370 state = get_state_from_digest_object(digestBase_obj, T_INT);
8371 stubAddr = StubRoutines::sha256_implCompress();
8372 stubName = "sha256_implCompress";
8373 break;
8374 case vmIntrinsics::_sha5_implCompress:
8375 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8376 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8377 stubAddr = StubRoutines::sha512_implCompress();
8378 stubName = "sha512_implCompress";
8379 break;
8380 case vmIntrinsics::_sha3_implCompress:
8381 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8382 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8383 stubAddr = StubRoutines::sha3_implCompress();
8384 stubName = "sha3_implCompress";
8385 block_size = get_block_size_from_digest_object(digestBase_obj);
8386 if (block_size == nullptr) return false;
8387 break;
8388 default:
8389 fatal_unexpected_iid(id);
8390 return false;
8391 }
8392 if (state == nullptr) return false;
8393
8394 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8395 if (stubAddr == nullptr) return false;
8396
8397 // Call the stub.
8398 Node* call;
8399 if (block_size == nullptr) {
8400 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8401 stubAddr, stubName, TypePtr::BOTTOM,
8402 src_start, state);
8403 } else {
8404 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8405 stubAddr, stubName, TypePtr::BOTTOM,
8406 src_start, state, block_size);
8407 }
8408
8409 return true;
8410 }
8411
8412 //------------------------------inline_double_keccak
8413 bool LibraryCallKit::inline_double_keccak() {
8414 address stubAddr;
8415 const char *stubName;
8416 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8417 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8418
8419 stubAddr = StubRoutines::double_keccak();
8420 stubName = "double_keccak";
8421 if (!stubAddr) return false;
8422
8423 Node* status0 = argument(0);
8424 Node* status1 = argument(1);
8425
8426 status0 = must_be_not_null(status0, true);
8427 status1 = must_be_not_null(status1, true);
8428
8429 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8430 assert(status0_start, "status0 is null");
8431 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8432 assert(status1_start, "status1 is null");
8433 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8434 OptoRuntime::double_keccak_Type(),
8435 stubAddr, stubName, TypePtr::BOTTOM,
8436 status0_start, status1_start);
8437 // return an int
8438 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8439 set_result(retvalue);
8440 return true;
8441 }
8442
8443
8444 //------------------------------inline_digestBase_implCompressMB-----------------------
8445 //
8446 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8447 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8448 //
8449 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8450 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8451 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8452 assert((uint)predicate < 5, "sanity");
8453 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8454
8455 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8456 Node* src = argument(1); // byte[] array
8457 Node* ofs = argument(2); // type int
8458 Node* limit = argument(3); // type int
8459
8460 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8461 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8462 // failed array check
8463 return false;
8464 }
8465 // Figure out the size and type of the elements we will be copying.
8466 BasicType src_elem = src_type->elem()->array_element_basic_type();
8467 if (src_elem != T_BYTE) {
8468 return false;
8469 }
8470 // 'src_start' points to src array + offset
8471 src = must_be_not_null(src, false);
8472 Node* src_start = array_element_address(src, ofs, src_elem);
8473
8474 const char* klass_digestBase_name = nullptr;
8475 const char* stub_name = nullptr;
8476 address stub_addr = nullptr;
8477 BasicType elem_type = T_INT;
8478
8479 switch (predicate) {
8480 case 0:
8481 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8482 klass_digestBase_name = "sun/security/provider/MD5";
8483 stub_name = "md5_implCompressMB";
8484 stub_addr = StubRoutines::md5_implCompressMB();
8485 }
8486 break;
8487 case 1:
8488 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8489 klass_digestBase_name = "sun/security/provider/SHA";
8490 stub_name = "sha1_implCompressMB";
8491 stub_addr = StubRoutines::sha1_implCompressMB();
8492 }
8493 break;
8494 case 2:
8495 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8496 klass_digestBase_name = "sun/security/provider/SHA2";
8497 stub_name = "sha256_implCompressMB";
8498 stub_addr = StubRoutines::sha256_implCompressMB();
8499 }
8500 break;
8501 case 3:
8502 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8503 klass_digestBase_name = "sun/security/provider/SHA5";
8504 stub_name = "sha512_implCompressMB";
8505 stub_addr = StubRoutines::sha512_implCompressMB();
8506 elem_type = T_LONG;
8507 }
8508 break;
8509 case 4:
8510 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8511 klass_digestBase_name = "sun/security/provider/SHA3";
8512 stub_name = "sha3_implCompressMB";
8513 stub_addr = StubRoutines::sha3_implCompressMB();
8514 elem_type = T_LONG;
8515 }
8516 break;
8517 default:
8518 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8519 }
8520 if (klass_digestBase_name != nullptr) {
8521 assert(stub_addr != nullptr, "Stub is generated");
8522 if (stub_addr == nullptr) return false;
8523
8524 // get DigestBase klass to lookup for SHA klass
8525 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8526 assert(tinst != nullptr, "digestBase_obj is not instance???");
8527 assert(tinst->is_loaded(), "DigestBase is not loaded");
8528
8529 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8530 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8531 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8532 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8533 }
8534 return false;
8535 }
8536
8537 //------------------------------inline_digestBase_implCompressMB-----------------------
8538 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8539 BasicType elem_type, address stubAddr, const char *stubName,
8540 Node* src_start, Node* ofs, Node* limit) {
8541 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8542 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8543 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8544 digest_obj = _gvn.transform(digest_obj);
8545
8546 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8547 if (state == nullptr) return false;
8548
8549 Node* block_size = nullptr;
8550 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8551 block_size = get_block_size_from_digest_object(digest_obj);
8552 if (block_size == nullptr) return false;
8553 }
8554
8555 // Call the stub.
8556 Node* call;
8557 if (block_size == nullptr) {
8558 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8559 OptoRuntime::digestBase_implCompressMB_Type(false),
8560 stubAddr, stubName, TypePtr::BOTTOM,
8561 src_start, state, ofs, limit);
8562 } else {
8563 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8564 OptoRuntime::digestBase_implCompressMB_Type(true),
8565 stubAddr, stubName, TypePtr::BOTTOM,
8566 src_start, state, block_size, ofs, limit);
8567 }
8568
8569 // return ofs (int)
8570 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8571 set_result(result);
8572
8573 return true;
8574 }
8575
8576 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8577 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8578 assert(UseAES, "need AES instruction support");
8579 address stubAddr = nullptr;
8580 const char *stubName = nullptr;
8581 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8582 stubName = "galoisCounterMode_AESCrypt";
8583
8584 if (stubAddr == nullptr) return false;
8585
8586 Node* in = argument(0);
8587 Node* inOfs = argument(1);
8588 Node* len = argument(2);
8589 Node* ct = argument(3);
8590 Node* ctOfs = argument(4);
8591 Node* out = argument(5);
8592 Node* outOfs = argument(6);
8593 Node* gctr_object = argument(7);
8594 Node* ghash_object = argument(8);
8595
8596 // (1) in, ct and out are arrays.
8597 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8598 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8599 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8600 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8601 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8602 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8603
8604 // checks are the responsibility of the caller
8605 Node* in_start = in;
8606 Node* ct_start = ct;
8607 Node* out_start = out;
8608 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8609 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8610 in_start = array_element_address(in, inOfs, T_BYTE);
8611 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8612 out_start = array_element_address(out, outOfs, T_BYTE);
8613 }
8614
8615 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8616 // (because of the predicated logic executed earlier).
8617 // so we cast it here safely.
8618 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8619 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8620 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8621 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8622 Node* state = load_field_from_object(ghash_object, "state", "[J");
8623
8624 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8625 return false;
8626 }
8627 // cast it to what we know it will be at runtime
8628 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8629 assert(tinst != nullptr, "GCTR obj is null");
8630 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8631 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8632 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8633 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8634 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8635 const TypeOopPtr* xtype = aklass->as_instance_type();
8636 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8637 aescrypt_object = _gvn.transform(aescrypt_object);
8638 // we need to get the start of the aescrypt_object's expanded key array
8639 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8640 if (k_start == nullptr) return false;
8641 // similarly, get the start address of the r vector
8642 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8643 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8644 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8645
8646
8647 // Call the stub, passing params
8648 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8649 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8650 stubAddr, stubName, TypePtr::BOTTOM,
8651 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8652
8653 // return cipher length (int)
8654 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8655 set_result(retvalue);
8656
8657 return true;
8658 }
8659
8660 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8661 // Return node representing slow path of predicate check.
8662 // the pseudo code we want to emulate with this predicate is:
8663 // for encryption:
8664 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8665 // for decryption:
8666 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8667 // note cipher==plain is more conservative than the original java code but that's OK
8668 //
8669
8670 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8671 // The receiver was checked for null already.
8672 Node* objGCTR = argument(7);
8673 // Load embeddedCipher field of GCTR object.
8674 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8675 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8676
8677 // get AESCrypt klass for instanceOf check
8678 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8679 // will have same classloader as CipherBlockChaining object
8680 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8681 assert(tinst != nullptr, "GCTR obj is null");
8682 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8683
8684 // we want to do an instanceof comparison against the AESCrypt class
8685 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8686 if (!klass_AESCrypt->is_loaded()) {
8687 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8688 Node* ctrl = control();
8689 set_control(top()); // no regular fast path
8690 return ctrl;
8691 }
8692
8693 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8694 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8695 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8696 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8697 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8698
8699 return instof_false; // even if it is null
8700 }
8701
8702 //------------------------------get_state_from_digest_object-----------------------
8703 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8704 const char* state_type;
8705 switch (elem_type) {
8706 case T_BYTE: state_type = "[B"; break;
8707 case T_INT: state_type = "[I"; break;
8708 case T_LONG: state_type = "[J"; break;
8709 default: ShouldNotReachHere();
8710 }
8711 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8712 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8713 if (digest_state == nullptr) return (Node *) nullptr;
8714
8715 // now have the array, need to get the start address of the state array
8716 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8717 return state;
8718 }
8719
8720 //------------------------------get_block_size_from_sha3_object----------------------------------
8721 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8722 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8723 assert (block_size != nullptr, "sanity");
8724 return block_size;
8725 }
8726
8727 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8728 // Return node representing slow path of predicate check.
8729 // the pseudo code we want to emulate with this predicate is:
8730 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8731 //
8732 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8733 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8734 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8735 assert((uint)predicate < 5, "sanity");
8736
8737 // The receiver was checked for null already.
8738 Node* digestBaseObj = argument(0);
8739
8740 // get DigestBase klass for instanceOf check
8741 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8742 assert(tinst != nullptr, "digestBaseObj is null");
8743 assert(tinst->is_loaded(), "DigestBase is not loaded");
8744
8745 const char* klass_name = nullptr;
8746 switch (predicate) {
8747 case 0:
8748 if (UseMD5Intrinsics) {
8749 // we want to do an instanceof comparison against the MD5 class
8750 klass_name = "sun/security/provider/MD5";
8751 }
8752 break;
8753 case 1:
8754 if (UseSHA1Intrinsics) {
8755 // we want to do an instanceof comparison against the SHA class
8756 klass_name = "sun/security/provider/SHA";
8757 }
8758 break;
8759 case 2:
8760 if (UseSHA256Intrinsics) {
8761 // we want to do an instanceof comparison against the SHA2 class
8762 klass_name = "sun/security/provider/SHA2";
8763 }
8764 break;
8765 case 3:
8766 if (UseSHA512Intrinsics) {
8767 // we want to do an instanceof comparison against the SHA5 class
8768 klass_name = "sun/security/provider/SHA5";
8769 }
8770 break;
8771 case 4:
8772 if (UseSHA3Intrinsics) {
8773 // we want to do an instanceof comparison against the SHA3 class
8774 klass_name = "sun/security/provider/SHA3";
8775 }
8776 break;
8777 default:
8778 fatal("unknown SHA intrinsic predicate: %d", predicate);
8779 }
8780
8781 ciKlass* klass = nullptr;
8782 if (klass_name != nullptr) {
8783 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8784 }
8785 if ((klass == nullptr) || !klass->is_loaded()) {
8786 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8787 Node* ctrl = control();
8788 set_control(top()); // no intrinsic path
8789 return ctrl;
8790 }
8791 ciInstanceKlass* instklass = klass->as_instance_klass();
8792
8793 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8794 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8795 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8796 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8797
8798 return instof_false; // even if it is null
8799 }
8800
8801 //-------------inline_fma-----------------------------------
8802 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8803 Node *a = nullptr;
8804 Node *b = nullptr;
8805 Node *c = nullptr;
8806 Node* result = nullptr;
8807 switch (id) {
8808 case vmIntrinsics::_fmaD:
8809 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8810 // no receiver since it is static method
8811 a = argument(0);
8812 b = argument(2);
8813 c = argument(4);
8814 result = _gvn.transform(new FmaDNode(a, b, c));
8815 break;
8816 case vmIntrinsics::_fmaF:
8817 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8818 a = argument(0);
8819 b = argument(1);
8820 c = argument(2);
8821 result = _gvn.transform(new FmaFNode(a, b, c));
8822 break;
8823 default:
8824 fatal_unexpected_iid(id); break;
8825 }
8826 set_result(result);
8827 return true;
8828 }
8829
8830 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8831 // argument(0) is receiver
8832 Node* codePoint = argument(1);
8833 Node* n = nullptr;
8834
8835 switch (id) {
8836 case vmIntrinsics::_isDigit :
8837 n = new DigitNode(control(), codePoint);
8838 break;
8839 case vmIntrinsics::_isLowerCase :
8840 n = new LowerCaseNode(control(), codePoint);
8841 break;
8842 case vmIntrinsics::_isUpperCase :
8843 n = new UpperCaseNode(control(), codePoint);
8844 break;
8845 case vmIntrinsics::_isWhitespace :
8846 n = new WhitespaceNode(control(), codePoint);
8847 break;
8848 default:
8849 fatal_unexpected_iid(id);
8850 }
8851
8852 set_result(_gvn.transform(n));
8853 return true;
8854 }
8855
8856 bool LibraryCallKit::inline_profileBoolean() {
8857 Node* counts = argument(1);
8858 const TypeAryPtr* ary = nullptr;
8859 ciArray* aobj = nullptr;
8860 if (counts->is_Con()
8861 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8862 && (aobj = ary->const_oop()->as_array()) != nullptr
8863 && (aobj->length() == 2)) {
8864 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8865 jint false_cnt = aobj->element_value(0).as_int();
8866 jint true_cnt = aobj->element_value(1).as_int();
8867
8868 if (C->log() != nullptr) {
8869 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8870 false_cnt, true_cnt);
8871 }
8872
8873 if (false_cnt + true_cnt == 0) {
8874 // According to profile, never executed.
8875 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8876 Deoptimization::Action_reinterpret);
8877 return true;
8878 }
8879
8880 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8881 // is a number of each value occurrences.
8882 Node* result = argument(0);
8883 if (false_cnt == 0 || true_cnt == 0) {
8884 // According to profile, one value has been never seen.
8885 int expected_val = (false_cnt == 0) ? 1 : 0;
8886
8887 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8888 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8889
8890 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8891 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8892 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8893
8894 { // Slow path: uncommon trap for never seen value and then reexecute
8895 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8896 // the value has been seen at least once.
8897 PreserveJVMState pjvms(this);
8898 PreserveReexecuteState preexecs(this);
8899 jvms()->set_should_reexecute(true);
8900
8901 set_control(slow_path);
8902 set_i_o(i_o());
8903
8904 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8905 Deoptimization::Action_reinterpret);
8906 }
8907 // The guard for never seen value enables sharpening of the result and
8908 // returning a constant. It allows to eliminate branches on the same value
8909 // later on.
8910 set_control(fast_path);
8911 result = intcon(expected_val);
8912 }
8913 // Stop profiling.
8914 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8915 // By replacing method body with profile data (represented as ProfileBooleanNode
8916 // on IR level) we effectively disable profiling.
8917 // It enables full speed execution once optimized code is generated.
8918 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8919 C->record_for_igvn(profile);
8920 set_result(profile);
8921 return true;
8922 } else {
8923 // Continue profiling.
8924 // Profile data isn't available at the moment. So, execute method's bytecode version.
8925 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8926 // is compiled and counters aren't available since corresponding MethodHandle
8927 // isn't a compile-time constant.
8928 return false;
8929 }
8930 }
8931
8932 bool LibraryCallKit::inline_isCompileConstant() {
8933 Node* n = argument(0);
8934 set_result(n->is_Con() ? intcon(1) : intcon(0));
8935 return true;
8936 }
8937
8938 //------------------------------- inline_getObjectSize --------------------------------------
8939 //
8940 // Calculate the runtime size of the object/array.
8941 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8942 //
8943 bool LibraryCallKit::inline_getObjectSize() {
8944 Node* obj = argument(3);
8945 Node* klass_node = load_object_klass(obj);
8946
8947 jint layout_con = Klass::_lh_neutral_value;
8948 Node* layout_val = get_layout_helper(klass_node, layout_con);
8949 int layout_is_con = (layout_val == nullptr);
8950
8951 if (layout_is_con) {
8952 // Layout helper is constant, can figure out things at compile time.
8953
8954 if (Klass::layout_helper_is_instance(layout_con)) {
8955 // Instance case: layout_con contains the size itself.
8956 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8957 set_result(size);
8958 } else {
8959 // Array case: size is round(header + element_size*arraylength).
8960 // Since arraylength is different for every array instance, we have to
8961 // compute the whole thing at runtime.
8962
8963 Node* arr_length = load_array_length(obj);
8964
8965 int round_mask = MinObjAlignmentInBytes - 1;
8966 int hsize = Klass::layout_helper_header_size(layout_con);
8967 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8968
8969 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8970 round_mask = 0; // strength-reduce it if it goes away completely
8971 }
8972 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8973 Node* header_size = intcon(hsize + round_mask);
8974
8975 Node* lengthx = ConvI2X(arr_length);
8976 Node* headerx = ConvI2X(header_size);
8977
8978 Node* abody = lengthx;
8979 if (eshift != 0) {
8980 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8981 }
8982 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8983 if (round_mask != 0) {
8984 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8985 }
8986 size = ConvX2L(size);
8987 set_result(size);
8988 }
8989 } else {
8990 // Layout helper is not constant, need to test for array-ness at runtime.
8991
8992 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8993 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8994 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8995 record_for_igvn(result_reg);
8996
8997 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8998 if (array_ctl != nullptr) {
8999 // Array case: size is round(header + element_size*arraylength).
9000 // Since arraylength is different for every array instance, we have to
9001 // compute the whole thing at runtime.
9002
9003 PreserveJVMState pjvms(this);
9004 set_control(array_ctl);
9005 Node* arr_length = load_array_length(obj);
9006
9007 int round_mask = MinObjAlignmentInBytes - 1;
9008 Node* mask = intcon(round_mask);
9009
9010 Node* hss = intcon(Klass::_lh_header_size_shift);
9011 Node* hsm = intcon(Klass::_lh_header_size_mask);
9012 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9013 header_size = _gvn.transform(new AndINode(header_size, hsm));
9014 header_size = _gvn.transform(new AddINode(header_size, mask));
9015
9016 // There is no need to mask or shift this value.
9017 // The semantics of LShiftINode include an implicit mask to 0x1F.
9018 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9019 Node* elem_shift = layout_val;
9020
9021 Node* lengthx = ConvI2X(arr_length);
9022 Node* headerx = ConvI2X(header_size);
9023
9024 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9025 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9026 if (round_mask != 0) {
9027 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9028 }
9029 size = ConvX2L(size);
9030
9031 result_reg->init_req(_array_path, control());
9032 result_val->init_req(_array_path, size);
9033 }
9034
9035 if (!stopped()) {
9036 // Instance case: the layout helper gives us instance size almost directly,
9037 // but we need to mask out the _lh_instance_slow_path_bit.
9038 Node* size = ConvI2X(layout_val);
9039 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9040 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9041 size = _gvn.transform(new AndXNode(size, mask));
9042 size = ConvX2L(size);
9043
9044 result_reg->init_req(_instance_path, control());
9045 result_val->init_req(_instance_path, size);
9046 }
9047
9048 set_result(result_reg, result_val);
9049 }
9050
9051 return true;
9052 }
9053
9054 //------------------------------- inline_blackhole --------------------------------------
9055 //
9056 // Make sure all arguments to this node are alive.
9057 // This matches methods that were requested to be blackholed through compile commands.
9058 //
9059 bool LibraryCallKit::inline_blackhole() {
9060 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9061 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9062 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9063
9064 // Blackhole node pinches only the control, not memory. This allows
9065 // the blackhole to be pinned in the loop that computes blackholed
9066 // values, but have no other side effects, like breaking the optimizations
9067 // across the blackhole.
9068
9069 Node* bh = _gvn.transform(new BlackholeNode(control()));
9070 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9071
9072 // Bind call arguments as blackhole arguments to keep them alive
9073 uint nargs = callee()->arg_size();
9074 for (uint i = 0; i < nargs; i++) {
9075 bh->add_req(argument(i));
9076 }
9077
9078 return true;
9079 }
9080
9081 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9082 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9083 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9084 return nullptr; // box klass is not Float16
9085 }
9086
9087 // Null check; get notnull casted pointer
9088 Node* null_ctl = top();
9089 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9090 // If not_null_box is dead, only null-path is taken
9091 if (stopped()) {
9092 set_control(null_ctl);
9093 return nullptr;
9094 }
9095 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9096 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9097 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9098 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9099 }
9100
9101 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9102 PreserveReexecuteState preexecs(this);
9103 jvms()->set_should_reexecute(true);
9104
9105 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9106 Node* klass_node = makecon(klass_type);
9107 Node* box = new_instance(klass_node);
9108
9109 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9110 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9111
9112 Node* field_store = _gvn.transform(access_store_at(box,
9113 value_field,
9114 value_adr_type,
9115 value,
9116 TypeInt::SHORT,
9117 T_SHORT,
9118 IN_HEAP));
9119 set_memory(field_store, value_adr_type);
9120 return box;
9121 }
9122
9123 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9124 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9125 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9126 return false;
9127 }
9128
9129 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9130 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9131 return false;
9132 }
9133
9134 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9135 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9136 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9137 ciSymbols::short_signature(),
9138 false);
9139 assert(field != nullptr, "");
9140
9141 // Transformed nodes
9142 Node* fld1 = nullptr;
9143 Node* fld2 = nullptr;
9144 Node* fld3 = nullptr;
9145 switch(num_args) {
9146 case 3:
9147 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9148 if (fld3 == nullptr) {
9149 return false;
9150 }
9151 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9152 // fall-through
9153 case 2:
9154 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9155 if (fld2 == nullptr) {
9156 return false;
9157 }
9158 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9159 // fall-through
9160 case 1:
9161 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9162 if (fld1 == nullptr) {
9163 return false;
9164 }
9165 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9166 break;
9167 default: fatal("Unsupported number of arguments %d", num_args);
9168 }
9169
9170 Node* result = nullptr;
9171 switch (id) {
9172 // Unary operations
9173 case vmIntrinsics::_sqrt_float16:
9174 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9175 break;
9176 // Ternary operations
9177 case vmIntrinsics::_fma_float16:
9178 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9179 break;
9180 default:
9181 fatal_unexpected_iid(id);
9182 break;
9183 }
9184 result = _gvn.transform(new ReinterpretHF2SNode(result));
9185 set_result(box_fp16_value(float16_box_type, field, result));
9186 return true;
9187 }
9188