1 /*
2 * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/mountUnmountDisabler.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "utilities/macros.hpp"
63 #include "utilities/powerOfTwo.hpp"
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return nullptr;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
85 if (is_available && is_virtual) {
86 is_available = vmIntrinsics::does_virtual_dispatch(id);
87 }
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return nullptr;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 return kit.transfer_exceptions_into_jvms();
135 }
136
137 // The intrinsic bailed out
138 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
139 assert(jvms->map() == kit.map(), "Out of sync JVM state");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dsinh:
250 case vmIntrinsics::_dtanh:
251 case vmIntrinsics::_dcbrt:
252 case vmIntrinsics::_dabs:
253 case vmIntrinsics::_fabs:
254 case vmIntrinsics::_iabs:
255 case vmIntrinsics::_labs:
256 case vmIntrinsics::_datan2:
257 case vmIntrinsics::_dsqrt:
258 case vmIntrinsics::_dsqrt_strict:
259 case vmIntrinsics::_dexp:
260 case vmIntrinsics::_dlog:
261 case vmIntrinsics::_dlog10:
262 case vmIntrinsics::_dpow:
263 case vmIntrinsics::_dcopySign:
264 case vmIntrinsics::_fcopySign:
265 case vmIntrinsics::_dsignum:
266 case vmIntrinsics::_roundF:
267 case vmIntrinsics::_roundD:
268 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
269
270 case vmIntrinsics::_notify:
271 case vmIntrinsics::_notifyAll:
272 return inline_notify(intrinsic_id());
273
274 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
275 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
276 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
277 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
278 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
279 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
280 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
281 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
282 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
283 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
284 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
285 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
286 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
287 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
288
289 case vmIntrinsics::_arraycopy: return inline_arraycopy();
290
291 case vmIntrinsics::_arraySort: return inline_array_sort();
292 case vmIntrinsics::_arrayPartition: return inline_array_partition();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309
310 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
311
312 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
313 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
314 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
315 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
316
317 case vmIntrinsics::_compressStringC:
318 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
319 case vmIntrinsics::_inflateStringC:
320 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
321
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
331
332 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
341
342 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
343 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
344 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
345 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
346 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
347 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
348 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
349 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
350 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
351
352 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
373 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
374 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
375 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
376 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
377 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
378 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
379 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
380 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
381
382 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
383 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
384 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
385 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
386 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
387 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
388 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
389 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
390 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
391
392 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
393 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
394 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
395 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
396 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
397 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
398 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
399 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
400 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
401
402 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
417
418 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
438
439 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
454
455 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
459
460 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
465
466 case vmIntrinsics::_loadFence:
467 case vmIntrinsics::_storeFence:
468 case vmIntrinsics::_storeStoreFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
474 case vmIntrinsics::_currentThread: return inline_native_currentThread();
475 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
476
477 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
478 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
479
480 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
481 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
482
483 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
484 "endFirstTransition", true);
485 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
486 "startFinalTransition", true);
487 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
488 "startTransition", false);
489 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
490 "endTransition", false);
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
508 case vmIntrinsics::_getLength: return inline_native_getLength();
509 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
510 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
511 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
512 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
513 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
514 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
515 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
516
517 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
518 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
519
520 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
521
522 case vmIntrinsics::_isInstance:
523 case vmIntrinsics::_isHidden:
524 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
535 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
536 case vmIntrinsics::_floatIsFinite:
537 case vmIntrinsics::_floatIsInfinite:
538 case vmIntrinsics::_doubleIsFinite:
539 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
540
541 case vmIntrinsics::_numberOfLeadingZeros_i:
542 case vmIntrinsics::_numberOfLeadingZeros_l:
543 case vmIntrinsics::_numberOfTrailingZeros_i:
544 case vmIntrinsics::_numberOfTrailingZeros_l:
545 case vmIntrinsics::_bitCount_i:
546 case vmIntrinsics::_bitCount_l:
547 case vmIntrinsics::_reverse_i:
548 case vmIntrinsics::_reverse_l:
549 case vmIntrinsics::_reverseBytes_i:
550 case vmIntrinsics::_reverseBytes_l:
551 case vmIntrinsics::_reverseBytes_s:
552 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
553
554 case vmIntrinsics::_compress_i:
555 case vmIntrinsics::_compress_l:
556 case vmIntrinsics::_expand_i:
557 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
558
559 case vmIntrinsics::_compareUnsigned_i:
560 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
561
562 case vmIntrinsics::_divideUnsigned_i:
563 case vmIntrinsics::_divideUnsigned_l:
564 case vmIntrinsics::_remainderUnsigned_i:
565 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
566
567 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
568
569 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
570 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
571 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
572 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
573 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
574
575 case vmIntrinsics::_Class_cast: return inline_Class_cast();
576
577 case vmIntrinsics::_aescrypt_encryptBlock:
578 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
579
580 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
581 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
582 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
583
584 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
585 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
586 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_counterMode_AESCrypt:
589 return inline_counterMode_AESCrypt(intrinsic_id());
590
591 case vmIntrinsics::_galoisCounterMode_AESCrypt:
592 return inline_galoisCounterMode_AESCrypt();
593
594 case vmIntrinsics::_md5_implCompress:
595 case vmIntrinsics::_sha_implCompress:
596 case vmIntrinsics::_sha2_implCompress:
597 case vmIntrinsics::_sha5_implCompress:
598 case vmIntrinsics::_sha3_implCompress:
599 return inline_digestBase_implCompress(intrinsic_id());
600 case vmIntrinsics::_double_keccak:
601 return inline_double_keccak();
602
603 case vmIntrinsics::_digestBase_implCompressMB:
604 return inline_digestBase_implCompressMB(predicate);
605
606 case vmIntrinsics::_multiplyToLen:
607 return inline_multiplyToLen();
608
609 case vmIntrinsics::_squareToLen:
610 return inline_squareToLen();
611
612 case vmIntrinsics::_mulAdd:
613 return inline_mulAdd();
614
615 case vmIntrinsics::_montgomeryMultiply:
616 return inline_montgomeryMultiply();
617 case vmIntrinsics::_montgomerySquare:
618 return inline_montgomerySquare();
619
620 case vmIntrinsics::_bigIntegerRightShiftWorker:
621 return inline_bigIntegerShift(true);
622 case vmIntrinsics::_bigIntegerLeftShiftWorker:
623 return inline_bigIntegerShift(false);
624
625 case vmIntrinsics::_vectorizedMismatch:
626 return inline_vectorizedMismatch();
627
628 case vmIntrinsics::_ghash_processBlocks:
629 return inline_ghash_processBlocks();
630 case vmIntrinsics::_chacha20Block:
631 return inline_chacha20Block();
632 case vmIntrinsics::_kyberNtt:
633 return inline_kyberNtt();
634 case vmIntrinsics::_kyberInverseNtt:
635 return inline_kyberInverseNtt();
636 case vmIntrinsics::_kyberNttMult:
637 return inline_kyberNttMult();
638 case vmIntrinsics::_kyberAddPoly_2:
639 return inline_kyberAddPoly_2();
640 case vmIntrinsics::_kyberAddPoly_3:
641 return inline_kyberAddPoly_3();
642 case vmIntrinsics::_kyber12To16:
643 return inline_kyber12To16();
644 case vmIntrinsics::_kyberBarrettReduce:
645 return inline_kyberBarrettReduce();
646 case vmIntrinsics::_dilithiumAlmostNtt:
647 return inline_dilithiumAlmostNtt();
648 case vmIntrinsics::_dilithiumAlmostInverseNtt:
649 return inline_dilithiumAlmostInverseNtt();
650 case vmIntrinsics::_dilithiumNttMult:
651 return inline_dilithiumNttMult();
652 case vmIntrinsics::_dilithiumMontMulByConstant:
653 return inline_dilithiumMontMulByConstant();
654 case vmIntrinsics::_dilithiumDecomposePoly:
655 return inline_dilithiumDecomposePoly();
656 case vmIntrinsics::_base64_encodeBlock:
657 return inline_base64_encodeBlock();
658 case vmIntrinsics::_base64_decodeBlock:
659 return inline_base64_decodeBlock();
660 case vmIntrinsics::_poly1305_processBlocks:
661 return inline_poly1305_processBlocks();
662 case vmIntrinsics::_intpoly_montgomeryMult_P256:
663 return inline_intpoly_montgomeryMult_P256();
664 case vmIntrinsics::_intpoly_assign:
665 return inline_intpoly_assign();
666 case vmIntrinsics::_encodeISOArray:
667 case vmIntrinsics::_encodeByteISOArray:
668 return inline_encodeISOArray(false);
669 case vmIntrinsics::_encodeAsciiArray:
670 return inline_encodeISOArray(true);
671
672 case vmIntrinsics::_updateCRC32:
673 return inline_updateCRC32();
674 case vmIntrinsics::_updateBytesCRC32:
675 return inline_updateBytesCRC32();
676 case vmIntrinsics::_updateByteBufferCRC32:
677 return inline_updateByteBufferCRC32();
678
679 case vmIntrinsics::_updateBytesCRC32C:
680 return inline_updateBytesCRC32C();
681 case vmIntrinsics::_updateDirectByteBufferCRC32C:
682 return inline_updateDirectByteBufferCRC32C();
683
684 case vmIntrinsics::_updateBytesAdler32:
685 return inline_updateBytesAdler32();
686 case vmIntrinsics::_updateByteBufferAdler32:
687 return inline_updateByteBufferAdler32();
688
689 case vmIntrinsics::_profileBoolean:
690 return inline_profileBoolean();
691 case vmIntrinsics::_isCompileConstant:
692 return inline_isCompileConstant();
693
694 case vmIntrinsics::_countPositives:
695 return inline_countPositives();
696
697 case vmIntrinsics::_fmaD:
698 case vmIntrinsics::_fmaF:
699 return inline_fma(intrinsic_id());
700
701 case vmIntrinsics::_isDigit:
702 case vmIntrinsics::_isLowerCase:
703 case vmIntrinsics::_isUpperCase:
704 case vmIntrinsics::_isWhitespace:
705 return inline_character_compare(intrinsic_id());
706
707 case vmIntrinsics::_min:
708 case vmIntrinsics::_max:
709 case vmIntrinsics::_min_strict:
710 case vmIntrinsics::_max_strict:
711 case vmIntrinsics::_minL:
712 case vmIntrinsics::_maxL:
713 case vmIntrinsics::_minF:
714 case vmIntrinsics::_maxF:
715 case vmIntrinsics::_minD:
716 case vmIntrinsics::_maxD:
717 case vmIntrinsics::_minF_strict:
718 case vmIntrinsics::_maxF_strict:
719 case vmIntrinsics::_minD_strict:
720 case vmIntrinsics::_maxD_strict:
721 return inline_min_max(intrinsic_id());
722
723 case vmIntrinsics::_VectorUnaryOp:
724 return inline_vector_nary_operation(1);
725 case vmIntrinsics::_VectorBinaryOp:
726 return inline_vector_nary_operation(2);
727 case vmIntrinsics::_VectorUnaryLibOp:
728 return inline_vector_call(1);
729 case vmIntrinsics::_VectorBinaryLibOp:
730 return inline_vector_call(2);
731 case vmIntrinsics::_VectorTernaryOp:
732 return inline_vector_nary_operation(3);
733 case vmIntrinsics::_VectorFromBitsCoerced:
734 return inline_vector_frombits_coerced();
735 case vmIntrinsics::_VectorMaskOp:
736 return inline_vector_mask_operation();
737 case vmIntrinsics::_VectorLoadOp:
738 return inline_vector_mem_operation(/*is_store=*/false);
739 case vmIntrinsics::_VectorLoadMaskedOp:
740 return inline_vector_mem_masked_operation(/*is_store*/false);
741 case vmIntrinsics::_VectorStoreOp:
742 return inline_vector_mem_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorStoreMaskedOp:
744 return inline_vector_mem_masked_operation(/*is_store=*/true);
745 case vmIntrinsics::_VectorGatherOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ false);
747 case vmIntrinsics::_VectorScatterOp:
748 return inline_vector_gather_scatter(/*is_scatter*/ true);
749 case vmIntrinsics::_VectorReductionCoerced:
750 return inline_vector_reduction();
751 case vmIntrinsics::_VectorTest:
752 return inline_vector_test();
753 case vmIntrinsics::_VectorBlend:
754 return inline_vector_blend();
755 case vmIntrinsics::_VectorRearrange:
756 return inline_vector_rearrange();
757 case vmIntrinsics::_VectorSelectFrom:
758 return inline_vector_select_from();
759 case vmIntrinsics::_VectorCompare:
760 return inline_vector_compare();
761 case vmIntrinsics::_VectorBroadcastInt:
762 return inline_vector_broadcast_int();
763 case vmIntrinsics::_VectorConvert:
764 return inline_vector_convert();
765 case vmIntrinsics::_VectorInsert:
766 return inline_vector_insert();
767 case vmIntrinsics::_VectorExtract:
768 return inline_vector_extract();
769 case vmIntrinsics::_VectorCompressExpand:
770 return inline_vector_compress_expand();
771 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
772 return inline_vector_select_from_two_vectors();
773 case vmIntrinsics::_IndexVector:
774 return inline_index_vector();
775 case vmIntrinsics::_IndexPartiallyInUpperRange:
776 return inline_index_partially_in_upper_range();
777
778 case vmIntrinsics::_getObjectSize:
779 return inline_getObjectSize();
780
781 case vmIntrinsics::_blackhole:
782 return inline_blackhole();
783
784 default:
785 // If you get here, it may be that someone has added a new intrinsic
786 // to the list in vmIntrinsics.hpp without implementing it here.
787 #ifndef PRODUCT
788 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
789 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
790 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
791 }
792 #endif
793 return false;
794 }
795 }
796
797 Node* LibraryCallKit::try_to_predicate(int predicate) {
798 if (!jvms()->has_method()) {
799 // Root JVMState has a null method.
800 assert(map()->memory()->Opcode() == Op_Parm, "");
801 // Insert the memory aliasing node
802 set_all_memory(reset_memory());
803 }
804 assert(merged_memory(), "");
805
806 switch (intrinsic_id()) {
807 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(false);
809 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt_predicate(true);
811 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(false);
813 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
814 return inline_electronicCodeBook_AESCrypt_predicate(true);
815 case vmIntrinsics::_counterMode_AESCrypt:
816 return inline_counterMode_AESCrypt_predicate();
817 case vmIntrinsics::_digestBase_implCompressMB:
818 return inline_digestBase_implCompressMB_predicate(predicate);
819 case vmIntrinsics::_galoisCounterMode_AESCrypt:
820 return inline_galoisCounterMode_AESCrypt_predicate();
821
822 default:
823 // If you get here, it may be that someone has added a new intrinsic
824 // to the list in vmIntrinsics.hpp without implementing it here.
825 #ifndef PRODUCT
826 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
827 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
828 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
829 }
830 #endif
831 Node* slow_ctl = control();
832 set_control(top()); // No fast path intrinsic
833 return slow_ctl;
834 }
835 }
836
837 //------------------------------set_result-------------------------------
838 // Helper function for finishing intrinsics.
839 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
840 record_for_igvn(region);
841 set_control(_gvn.transform(region));
842 set_result( _gvn.transform(value));
843 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
844 }
845
846 //------------------------------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, bool with_opaque) {
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 if (with_opaque) {
902 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
903 }
904 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
905 if (is_neg != nullptr && pos_index != nullptr) {
906 // Emulate effect of Parse::adjust_map_after_if.
907 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
908 (*pos_index) = _gvn.transform(ccast);
909 }
910 return is_neg;
911 }
912
913 // Make sure that 'position' is a valid limit index, in [0..length].
914 // There are two equivalent plans for checking this:
915 // A. (offset + copyLength) unsigned<= arrayLength
916 // B. offset <= (arrayLength - copyLength)
917 // We require that all of the values above, except for the sum and
918 // difference, are already known to be non-negative.
919 // Plan A is robust in the face of overflow, if offset and copyLength
920 // are both hugely positive.
921 //
922 // Plan B is less direct and intuitive, but it does not overflow at
923 // all, since the difference of two non-negatives is always
924 // representable. Whenever Java methods must perform the equivalent
925 // check they generally use Plan B instead of Plan A.
926 // For the moment we use Plan A.
927 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
928 Node* subseq_length,
929 Node* array_length,
930 RegionNode* region,
931 bool with_opaque) {
932 if (stopped())
933 return nullptr; // already stopped
934 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
935 if (zero_offset && subseq_length->eqv_uncast(array_length))
936 return nullptr; // common case of whole-array copy
937 Node* last = subseq_length;
938 if (!zero_offset) // last += offset
939 last = _gvn.transform(new AddINode(last, offset));
940 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
941 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
942 if (with_opaque) {
943 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
944 }
945 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
946 return is_over;
947 }
948
949 // Emit range checks for the given String.value byte array
950 void LibraryCallKit::generate_string_range_check(Node* array,
951 Node* offset,
952 Node* count,
953 bool char_count,
954 bool halt_on_oob) {
955 if (stopped()) {
956 return; // already stopped
957 }
958 RegionNode* bailout = new RegionNode(1);
959 record_for_igvn(bailout);
960 if (char_count) {
961 // Convert char count to byte count
962 count = _gvn.transform(new LShiftINode(count, intcon(1)));
963 }
964
965 // Offset and count must not be negative
966 generate_negative_guard(offset, bailout, nullptr, halt_on_oob);
967 generate_negative_guard(count, bailout, nullptr, halt_on_oob);
968 // Offset + count must not exceed length of array
969 generate_limit_guard(offset, count, load_array_length(array), bailout, halt_on_oob);
970
971 if (bailout->req() > 1) {
972 if (halt_on_oob) {
973 bailout = _gvn.transform(bailout)->as_Region();
974 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
975 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
976 C->root()->add_req(halt);
977 } else {
978 PreserveJVMState pjvms(this);
979 set_control(_gvn.transform(bailout));
980 uncommon_trap(Deoptimization::Reason_intrinsic,
981 Deoptimization::Action_maybe_recompile);
982 }
983 }
984 }
985
986 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
987 bool is_immutable) {
988 ciKlass* thread_klass = env()->Thread_klass();
989 const Type* thread_type
990 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
991
992 Node* thread = _gvn.transform(new ThreadLocalNode());
993 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
994 tls_output = thread;
995
996 Node* thread_obj_handle
997 = (is_immutable
998 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
999 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1000 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1001 thread_obj_handle = _gvn.transform(thread_obj_handle);
1002
1003 DecoratorSet decorators = IN_NATIVE;
1004 if (is_immutable) {
1005 decorators |= C2_IMMUTABLE_MEMORY;
1006 }
1007 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1008 }
1009
1010 //--------------------------generate_current_thread--------------------
1011 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1012 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1013 /*is_immutable*/false);
1014 }
1015
1016 //--------------------------generate_virtual_thread--------------------
1017 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1018 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1019 !C->method()->changes_current_thread());
1020 }
1021
1022 //------------------------------make_string_method_node------------------------
1023 // Helper method for String intrinsic functions. This version is called with
1024 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1025 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1026 // containing the lengths of str1 and str2.
1027 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1028 Node* result = nullptr;
1029 switch (opcode) {
1030 case Op_StrIndexOf:
1031 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1032 str1_start, cnt1, str2_start, cnt2, ae);
1033 break;
1034 case Op_StrComp:
1035 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1036 str1_start, cnt1, str2_start, cnt2, ae);
1037 break;
1038 case Op_StrEquals:
1039 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1040 // Use the constant length if there is one because optimized match rule may exist.
1041 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1042 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1043 break;
1044 default:
1045 ShouldNotReachHere();
1046 return nullptr;
1047 }
1048
1049 // All these intrinsics have checks.
1050 C->set_has_split_ifs(true); // Has chance for split-if optimization
1051 clear_upper_avx();
1052
1053 return _gvn.transform(result);
1054 }
1055
1056 //------------------------------inline_string_compareTo------------------------
1057 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1058 Node* arg1 = argument(0);
1059 Node* arg2 = argument(1);
1060
1061 arg1 = must_be_not_null(arg1, true);
1062 arg2 = must_be_not_null(arg2, true);
1063
1064 // Get start addr and length of first argument
1065 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1066 Node* arg1_cnt = load_array_length(arg1);
1067
1068 // Get start addr and length of second argument
1069 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1070 Node* arg2_cnt = load_array_length(arg2);
1071
1072 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1073 set_result(result);
1074 return true;
1075 }
1076
1077 //------------------------------inline_string_equals------------------------
1078 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1079 Node* arg1 = argument(0);
1080 Node* arg2 = argument(1);
1081
1082 // paths (plus control) merge
1083 RegionNode* region = new RegionNode(3);
1084 Node* phi = new PhiNode(region, TypeInt::BOOL);
1085
1086 if (!stopped()) {
1087
1088 arg1 = must_be_not_null(arg1, true);
1089 arg2 = must_be_not_null(arg2, true);
1090
1091 // Get start addr and length of first argument
1092 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1093 Node* arg1_cnt = load_array_length(arg1);
1094
1095 // Get start addr and length of second argument
1096 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1097 Node* arg2_cnt = load_array_length(arg2);
1098
1099 // Check for arg1_cnt != arg2_cnt
1100 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1101 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1102 Node* if_ne = generate_slow_guard(bol, nullptr);
1103 if (if_ne != nullptr) {
1104 phi->init_req(2, intcon(0));
1105 region->init_req(2, if_ne);
1106 }
1107
1108 // Check for count == 0 is done by assembler code for StrEquals.
1109
1110 if (!stopped()) {
1111 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1112 phi->init_req(1, equals);
1113 region->init_req(1, control());
1114 }
1115 }
1116
1117 // post merge
1118 set_control(_gvn.transform(region));
1119 record_for_igvn(region);
1120
1121 set_result(_gvn.transform(phi));
1122 return true;
1123 }
1124
1125 //------------------------------inline_array_equals----------------------------
1126 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1127 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1128 Node* arg1 = argument(0);
1129 Node* arg2 = argument(1);
1130
1131 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1132 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1133 clear_upper_avx();
1134
1135 return true;
1136 }
1137
1138
1139 //------------------------------inline_countPositives------------------------------
1140 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1141 bool LibraryCallKit::inline_countPositives() {
1142 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1143 return false;
1144 }
1145
1146 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1147 // no receiver since it is static method
1148 Node* ba = argument(0);
1149 Node* offset = argument(1);
1150 Node* len = argument(2);
1151
1152 ba = must_be_not_null(ba, true);
1153 generate_string_range_check(ba, offset, len, false, true);
1154 if (stopped()) {
1155 return true;
1156 }
1157
1158 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1159 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1160 set_result(_gvn.transform(result));
1161 clear_upper_avx();
1162 return true;
1163 }
1164
1165 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1166 Node* index = argument(0);
1167 Node* length = bt == T_INT ? argument(1) : argument(2);
1168 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1169 return false;
1170 }
1171
1172 // check that length is positive
1173 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1174 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1175
1176 {
1177 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1178 uncommon_trap(Deoptimization::Reason_intrinsic,
1179 Deoptimization::Action_make_not_entrant);
1180 }
1181
1182 if (stopped()) {
1183 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1184 return true;
1185 }
1186
1187 // length is now known positive, add a cast node to make this explicit
1188 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1189 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1190 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1191 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1192 casted_length = _gvn.transform(casted_length);
1193 replace_in_map(length, casted_length);
1194 length = casted_length;
1195
1196 // Use an unsigned comparison for the range check itself
1197 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1198 BoolTest::mask btest = BoolTest::lt;
1199 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1200 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1201 _gvn.set_type(rc, rc->Value(&_gvn));
1202 if (!rc_bool->is_Con()) {
1203 record_for_igvn(rc);
1204 }
1205 set_control(_gvn.transform(new IfTrueNode(rc)));
1206 {
1207 PreserveJVMState pjvms(this);
1208 set_control(_gvn.transform(new IfFalseNode(rc)));
1209 uncommon_trap(Deoptimization::Reason_range_check,
1210 Deoptimization::Action_make_not_entrant);
1211 }
1212
1213 if (stopped()) {
1214 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1215 return true;
1216 }
1217
1218 // index is now known to be >= 0 and < length, cast it
1219 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1220 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1221 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1222 result = _gvn.transform(result);
1223 set_result(result);
1224 replace_in_map(index, result);
1225 return true;
1226 }
1227
1228 //------------------------------inline_string_indexOf------------------------
1229 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1230 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1231 return false;
1232 }
1233 Node* src = argument(0);
1234 Node* tgt = argument(1);
1235
1236 // Make the merge point
1237 RegionNode* result_rgn = new RegionNode(4);
1238 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1239
1240 src = must_be_not_null(src, true);
1241 tgt = must_be_not_null(tgt, true);
1242
1243 // Get start addr and length of source string
1244 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1245 Node* src_count = load_array_length(src);
1246
1247 // Get start addr and length of substring
1248 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1249 Node* tgt_count = load_array_length(tgt);
1250
1251 Node* result = nullptr;
1252 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1253
1254 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1255 // Divide src size by 2 if String is UTF16 encoded
1256 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1257 }
1258 if (ae == StrIntrinsicNode::UU) {
1259 // Divide substring size by 2 if String is UTF16 encoded
1260 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1261 }
1262
1263 if (call_opt_stub) {
1264 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1265 StubRoutines::_string_indexof_array[ae],
1266 "stringIndexOf", TypePtr::BOTTOM, src_start,
1267 src_count, tgt_start, tgt_count);
1268 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1269 } else {
1270 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1271 result_rgn, result_phi, ae);
1272 }
1273 if (result != nullptr) {
1274 result_phi->init_req(3, result);
1275 result_rgn->init_req(3, control());
1276 }
1277 set_control(_gvn.transform(result_rgn));
1278 record_for_igvn(result_rgn);
1279 set_result(_gvn.transform(result_phi));
1280
1281 return true;
1282 }
1283
1284 //-----------------------------inline_string_indexOfI-----------------------
1285 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1286 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1287 return false;
1288 }
1289 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1290 return false;
1291 }
1292
1293 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1294 Node* src = argument(0); // byte[]
1295 Node* src_count = argument(1); // char count
1296 Node* tgt = argument(2); // byte[]
1297 Node* tgt_count = argument(3); // char count
1298 Node* from_index = argument(4); // char index
1299
1300 src = must_be_not_null(src, true);
1301 tgt = must_be_not_null(tgt, true);
1302
1303 // Multiply byte array index by 2 if String is UTF16 encoded
1304 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1305 src_count = _gvn.transform(new SubINode(src_count, from_index));
1306 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1307 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1308
1309 // Range checks
1310 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1311 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1312 if (stopped()) {
1313 return true;
1314 }
1315
1316 RegionNode* region = new RegionNode(5);
1317 Node* phi = new PhiNode(region, TypeInt::INT);
1318 Node* result = nullptr;
1319
1320 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1321
1322 if (call_opt_stub) {
1323 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1324 StubRoutines::_string_indexof_array[ae],
1325 "stringIndexOf", TypePtr::BOTTOM, src_start,
1326 src_count, tgt_start, tgt_count);
1327 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1328 } else {
1329 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1330 region, phi, ae);
1331 }
1332 if (result != nullptr) {
1333 // The result is index relative to from_index if substring was found, -1 otherwise.
1334 // Generate code which will fold into cmove.
1335 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1336 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1337
1338 Node* if_lt = generate_slow_guard(bol, nullptr);
1339 if (if_lt != nullptr) {
1340 // result == -1
1341 phi->init_req(3, result);
1342 region->init_req(3, if_lt);
1343 }
1344 if (!stopped()) {
1345 result = _gvn.transform(new AddINode(result, from_index));
1346 phi->init_req(4, result);
1347 region->init_req(4, control());
1348 }
1349 }
1350
1351 set_control(_gvn.transform(region));
1352 record_for_igvn(region);
1353 set_result(_gvn.transform(phi));
1354 clear_upper_avx();
1355
1356 return true;
1357 }
1358
1359 // Create StrIndexOfNode with fast path checks
1360 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1361 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1362 // Check for substr count > string count
1363 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1364 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1365 Node* if_gt = generate_slow_guard(bol, nullptr);
1366 if (if_gt != nullptr) {
1367 phi->init_req(1, intcon(-1));
1368 region->init_req(1, if_gt);
1369 }
1370 if (!stopped()) {
1371 // Check for substr count == 0
1372 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1373 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1374 Node* if_zero = generate_slow_guard(bol, nullptr);
1375 if (if_zero != nullptr) {
1376 phi->init_req(2, intcon(0));
1377 region->init_req(2, if_zero);
1378 }
1379 }
1380 if (!stopped()) {
1381 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1382 }
1383 return nullptr;
1384 }
1385
1386 //-----------------------------inline_string_indexOfChar-----------------------
1387 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1388 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1389 return false;
1390 }
1391 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1392 return false;
1393 }
1394 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1395 Node* src = argument(0); // byte[]
1396 Node* int_ch = argument(1);
1397 Node* from_index = argument(2);
1398 Node* max = argument(3);
1399
1400 src = must_be_not_null(src, true);
1401
1402 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1403 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1404 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1405
1406 // Range checks
1407 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1408
1409 // Check for int_ch >= 0
1410 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1411 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1412 {
1413 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1414 uncommon_trap(Deoptimization::Reason_intrinsic,
1415 Deoptimization::Action_maybe_recompile);
1416 }
1417 if (stopped()) {
1418 return true;
1419 }
1420
1421 RegionNode* region = new RegionNode(3);
1422 Node* phi = new PhiNode(region, TypeInt::INT);
1423
1424 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1425 C->set_has_split_ifs(true); // Has chance for split-if optimization
1426 _gvn.transform(result);
1427
1428 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1429 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1430
1431 Node* if_lt = generate_slow_guard(bol, nullptr);
1432 if (if_lt != nullptr) {
1433 // result == -1
1434 phi->init_req(2, result);
1435 region->init_req(2, if_lt);
1436 }
1437 if (!stopped()) {
1438 result = _gvn.transform(new AddINode(result, from_index));
1439 phi->init_req(1, result);
1440 region->init_req(1, control());
1441 }
1442 set_control(_gvn.transform(region));
1443 record_for_igvn(region);
1444 set_result(_gvn.transform(phi));
1445 clear_upper_avx();
1446
1447 return true;
1448 }
1449 //---------------------------inline_string_copy---------------------
1450 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1451 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1452 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1453 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1454 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1455 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1456 bool LibraryCallKit::inline_string_copy(bool compress) {
1457 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1458 return false;
1459 }
1460 int nargs = 5; // 2 oops, 3 ints
1461 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1462
1463 Node* src = argument(0);
1464 Node* src_offset = argument(1);
1465 Node* dst = argument(2);
1466 Node* dst_offset = argument(3);
1467 Node* length = argument(4);
1468
1469 // Check for allocation before we add nodes that would confuse
1470 // tightly_coupled_allocation()
1471 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1472
1473 // Figure out the size and type of the elements we will be copying.
1474 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1475 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1476 if (src_type == nullptr || dst_type == nullptr) {
1477 return false;
1478 }
1479 BasicType src_elem = src_type->elem()->array_element_basic_type();
1480 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1481 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1482 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1483 "Unsupported array types for inline_string_copy");
1484
1485 src = must_be_not_null(src, true);
1486 dst = must_be_not_null(dst, true);
1487
1488 // Convert char[] offsets to byte[] offsets
1489 bool convert_src = (compress && src_elem == T_BYTE);
1490 bool convert_dst = (!compress && dst_elem == T_BYTE);
1491 if (convert_src) {
1492 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1493 } else if (convert_dst) {
1494 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1495 }
1496
1497 // Range checks
1498 generate_string_range_check(src, src_offset, length, convert_src);
1499 generate_string_range_check(dst, dst_offset, length, convert_dst);
1500 if (stopped()) {
1501 return true;
1502 }
1503
1504 Node* src_start = array_element_address(src, src_offset, src_elem);
1505 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1506 // 'src_start' points to src array + scaled offset
1507 // 'dst_start' points to dst array + scaled offset
1508 Node* count = nullptr;
1509 if (compress) {
1510 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1511 } else {
1512 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1513 }
1514
1515 if (alloc != nullptr) {
1516 if (alloc->maybe_set_complete(&_gvn)) {
1517 // "You break it, you buy it."
1518 InitializeNode* init = alloc->initialization();
1519 assert(init->is_complete(), "we just did this");
1520 init->set_complete_with_arraycopy();
1521 assert(dst->is_CheckCastPP(), "sanity");
1522 assert(dst->in(0)->in(0) == init, "dest pinned");
1523 }
1524 // Do not let stores that initialize this object be reordered with
1525 // a subsequent store that would make this object accessible by
1526 // other threads.
1527 // Record what AllocateNode this StoreStore protects so that
1528 // escape analysis can go from the MemBarStoreStoreNode to the
1529 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1530 // based on the escape status of the AllocateNode.
1531 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1532 }
1533 if (compress) {
1534 set_result(_gvn.transform(count));
1535 }
1536 clear_upper_avx();
1537
1538 return true;
1539 }
1540
1541 #ifdef _LP64
1542 #define XTOP ,top() /*additional argument*/
1543 #else //_LP64
1544 #define XTOP /*no additional argument*/
1545 #endif //_LP64
1546
1547 //------------------------inline_string_toBytesU--------------------------
1548 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1549 bool LibraryCallKit::inline_string_toBytesU() {
1550 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1551 return false;
1552 }
1553 // Get the arguments.
1554 Node* value = argument(0);
1555 Node* offset = argument(1);
1556 Node* length = argument(2);
1557
1558 Node* newcopy = nullptr;
1559
1560 // Set the original stack and the reexecute bit for the interpreter to reexecute
1561 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1562 { PreserveReexecuteState preexecs(this);
1563 jvms()->set_should_reexecute(true);
1564
1565 // Check if a null path was taken unconditionally.
1566 value = null_check(value);
1567
1568 RegionNode* bailout = new RegionNode(1);
1569 record_for_igvn(bailout);
1570
1571 // Range checks
1572 generate_negative_guard(offset, bailout);
1573 generate_negative_guard(length, bailout);
1574 generate_limit_guard(offset, length, load_array_length(value), bailout);
1575 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1576 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1577
1578 if (bailout->req() > 1) {
1579 PreserveJVMState pjvms(this);
1580 set_control(_gvn.transform(bailout));
1581 uncommon_trap(Deoptimization::Reason_intrinsic,
1582 Deoptimization::Action_maybe_recompile);
1583 }
1584 if (stopped()) {
1585 return true;
1586 }
1587
1588 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1589 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1590 newcopy = new_array(klass_node, size, 0); // no arguments to push
1591 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1592 guarantee(alloc != nullptr, "created above");
1593
1594 // Calculate starting addresses.
1595 Node* src_start = array_element_address(value, offset, T_CHAR);
1596 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1597
1598 // Check if dst array address is aligned to HeapWordSize
1599 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1600 // If true, then check if src array address is aligned to HeapWordSize
1601 if (aligned) {
1602 const TypeInt* toffset = gvn().type(offset)->is_int();
1603 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1604 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1605 }
1606
1607 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1608 const char* copyfunc_name = "arraycopy";
1609 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1610 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1611 OptoRuntime::fast_arraycopy_Type(),
1612 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1613 src_start, dst_start, ConvI2X(length) XTOP);
1614 // Do not let reads from the cloned object float above the arraycopy.
1615 if (alloc->maybe_set_complete(&_gvn)) {
1616 // "You break it, you buy it."
1617 InitializeNode* init = alloc->initialization();
1618 assert(init->is_complete(), "we just did this");
1619 init->set_complete_with_arraycopy();
1620 assert(newcopy->is_CheckCastPP(), "sanity");
1621 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1622 }
1623 // Do not let stores that initialize this object be reordered with
1624 // a subsequent store that would make this object accessible by
1625 // other threads.
1626 // Record what AllocateNode this StoreStore protects so that
1627 // escape analysis can go from the MemBarStoreStoreNode to the
1628 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1629 // based on the escape status of the AllocateNode.
1630 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1631 } // original reexecute is set back here
1632
1633 C->set_has_split_ifs(true); // Has chance for split-if optimization
1634 if (!stopped()) {
1635 set_result(newcopy);
1636 }
1637 clear_upper_avx();
1638
1639 return true;
1640 }
1641
1642 //------------------------inline_string_getCharsU--------------------------
1643 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1644 bool LibraryCallKit::inline_string_getCharsU() {
1645 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1646 return false;
1647 }
1648
1649 // Get the arguments.
1650 Node* src = argument(0);
1651 Node* src_begin = argument(1);
1652 Node* src_end = argument(2); // exclusive offset (i < src_end)
1653 Node* dst = argument(3);
1654 Node* dst_begin = argument(4);
1655
1656 // Check for allocation before we add nodes that would confuse
1657 // tightly_coupled_allocation()
1658 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1659
1660 // Check if a null path was taken unconditionally.
1661 src = null_check(src);
1662 dst = null_check(dst);
1663 if (stopped()) {
1664 return true;
1665 }
1666
1667 // Get length and convert char[] offset to byte[] offset
1668 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1669 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1670
1671 // Range checks
1672 generate_string_range_check(src, src_begin, length, true);
1673 generate_string_range_check(dst, dst_begin, length, false);
1674 if (stopped()) {
1675 return true;
1676 }
1677
1678 if (!stopped()) {
1679 // Calculate starting addresses.
1680 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1681 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1682
1683 // Check if array addresses are aligned to HeapWordSize
1684 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1685 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1686 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1687 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1688
1689 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1690 const char* copyfunc_name = "arraycopy";
1691 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1692 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1693 OptoRuntime::fast_arraycopy_Type(),
1694 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1695 src_start, dst_start, ConvI2X(length) XTOP);
1696 // Do not let reads from the cloned object float above the arraycopy.
1697 if (alloc != nullptr) {
1698 if (alloc->maybe_set_complete(&_gvn)) {
1699 // "You break it, you buy it."
1700 InitializeNode* init = alloc->initialization();
1701 assert(init->is_complete(), "we just did this");
1702 init->set_complete_with_arraycopy();
1703 assert(dst->is_CheckCastPP(), "sanity");
1704 assert(dst->in(0)->in(0) == init, "dest pinned");
1705 }
1706 // Do not let stores that initialize this object be reordered with
1707 // a subsequent store that would make this object accessible by
1708 // other threads.
1709 // Record what AllocateNode this StoreStore protects so that
1710 // escape analysis can go from the MemBarStoreStoreNode to the
1711 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1712 // based on the escape status of the AllocateNode.
1713 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1714 } else {
1715 insert_mem_bar(Op_MemBarCPUOrder);
1716 }
1717 }
1718
1719 C->set_has_split_ifs(true); // Has chance for split-if optimization
1720 return true;
1721 }
1722
1723 //----------------------inline_string_char_access----------------------------
1724 // Store/Load char to/from byte[] array.
1725 // static void StringUTF16.putChar(byte[] val, int index, int c)
1726 // static char StringUTF16.getChar(byte[] val, int index)
1727 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1728 Node* value = argument(0);
1729 Node* index = argument(1);
1730 Node* ch = is_store ? argument(2) : nullptr;
1731
1732 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1733 // correctly requires matched array shapes.
1734 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1735 "sanity: byte[] and char[] bases agree");
1736 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1737 "sanity: byte[] and char[] scales agree");
1738
1739 // Bail when getChar over constants is requested: constant folding would
1740 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1741 // Java method would constant fold nicely instead.
1742 if (!is_store && value->is_Con() && index->is_Con()) {
1743 return false;
1744 }
1745
1746 // Save state and restore on bailout
1747 SavedState old_state(this);
1748
1749 value = must_be_not_null(value, true);
1750
1751 Node* adr = array_element_address(value, index, T_CHAR);
1752 if (adr->is_top()) {
1753 return false;
1754 }
1755 old_state.discard();
1756 if (is_store) {
1757 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1758 } else {
1759 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);
1760 set_result(ch);
1761 }
1762 return true;
1763 }
1764
1765
1766 //------------------------------inline_math-----------------------------------
1767 // public static double Math.abs(double)
1768 // public static double Math.sqrt(double)
1769 // public static double Math.log(double)
1770 // public static double Math.log10(double)
1771 // public static double Math.round(double)
1772 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1773 Node* arg = argument(0);
1774 Node* n = nullptr;
1775 switch (id) {
1776 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1777 case vmIntrinsics::_dsqrt:
1778 case vmIntrinsics::_dsqrt_strict:
1779 n = new SqrtDNode(C, control(), arg); break;
1780 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1781 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1782 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1783 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1784 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1785 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1786 default: fatal_unexpected_iid(id); break;
1787 }
1788 set_result(_gvn.transform(n));
1789 return true;
1790 }
1791
1792 //------------------------------inline_math-----------------------------------
1793 // public static float Math.abs(float)
1794 // public static int Math.abs(int)
1795 // public static long Math.abs(long)
1796 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1797 Node* arg = argument(0);
1798 Node* n = nullptr;
1799 switch (id) {
1800 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1801 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1802 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1803 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1804 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1805 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1806 default: fatal_unexpected_iid(id); break;
1807 }
1808 set_result(_gvn.transform(n));
1809 return true;
1810 }
1811
1812 //------------------------------runtime_math-----------------------------
1813 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1814 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1815 "must be (DD)D or (D)D type");
1816
1817 // Inputs
1818 Node* a = argument(0);
1819 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1820
1821 const TypePtr* no_memory_effects = nullptr;
1822 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1823 no_memory_effects,
1824 a, top(), b, b ? top() : nullptr);
1825 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1826 #ifdef ASSERT
1827 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1828 assert(value_top == top(), "second value must be top");
1829 #endif
1830
1831 set_result(value);
1832 return true;
1833 }
1834
1835 //------------------------------inline_math_pow-----------------------------
1836 bool LibraryCallKit::inline_math_pow() {
1837 Node* exp = argument(2);
1838 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1839 if (d != nullptr) {
1840 if (d->getd() == 2.0) {
1841 // Special case: pow(x, 2.0) => x * x
1842 Node* base = argument(0);
1843 set_result(_gvn.transform(new MulDNode(base, base)));
1844 return true;
1845 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1846 // Special case: pow(x, 0.5) => sqrt(x)
1847 Node* base = argument(0);
1848 Node* zero = _gvn.zerocon(T_DOUBLE);
1849
1850 RegionNode* region = new RegionNode(3);
1851 Node* phi = new PhiNode(region, Type::DOUBLE);
1852
1853 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1854 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1855 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1856 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1857 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1858
1859 Node* if_pow = generate_slow_guard(test, nullptr);
1860 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1861 phi->init_req(1, value_sqrt);
1862 region->init_req(1, control());
1863
1864 if (if_pow != nullptr) {
1865 set_control(if_pow);
1866 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1867 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1868 const TypePtr* no_memory_effects = nullptr;
1869 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1870 no_memory_effects, base, top(), exp, top());
1871 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1872 #ifdef ASSERT
1873 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1874 assert(value_top == top(), "second value must be top");
1875 #endif
1876 phi->init_req(2, value_pow);
1877 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1878 }
1879
1880 C->set_has_split_ifs(true); // Has chance for split-if optimization
1881 set_control(_gvn.transform(region));
1882 record_for_igvn(region);
1883 set_result(_gvn.transform(phi));
1884
1885 return true;
1886 }
1887 }
1888
1889 return StubRoutines::dpow() != nullptr ?
1890 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1891 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1892 }
1893
1894 //------------------------------inline_math_native-----------------------------
1895 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1896 switch (id) {
1897 case vmIntrinsics::_dsin:
1898 return StubRoutines::dsin() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1900 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1901 case vmIntrinsics::_dcos:
1902 return StubRoutines::dcos() != nullptr ?
1903 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1904 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1905 case vmIntrinsics::_dtan:
1906 return StubRoutines::dtan() != nullptr ?
1907 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1908 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1909 case vmIntrinsics::_dsinh:
1910 return StubRoutines::dsinh() != nullptr ?
1911 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1912 case vmIntrinsics::_dtanh:
1913 return StubRoutines::dtanh() != nullptr ?
1914 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1915 case vmIntrinsics::_dcbrt:
1916 return StubRoutines::dcbrt() != nullptr ?
1917 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1918 case vmIntrinsics::_dexp:
1919 return StubRoutines::dexp() != nullptr ?
1920 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1921 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1922 case vmIntrinsics::_dlog:
1923 return StubRoutines::dlog() != nullptr ?
1924 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1925 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1926 case vmIntrinsics::_dlog10:
1927 return StubRoutines::dlog10() != nullptr ?
1928 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1929 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1930
1931 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1932 case vmIntrinsics::_ceil:
1933 case vmIntrinsics::_floor:
1934 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1935
1936 case vmIntrinsics::_dsqrt:
1937 case vmIntrinsics::_dsqrt_strict:
1938 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1939 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1940 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1941 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1942 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1943
1944 case vmIntrinsics::_dpow: return inline_math_pow();
1945 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1946 case vmIntrinsics::_fcopySign: return inline_math(id);
1947 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1948 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1949 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1950
1951 // These intrinsics are not yet correctly implemented
1952 case vmIntrinsics::_datan2:
1953 return false;
1954
1955 default:
1956 fatal_unexpected_iid(id);
1957 return false;
1958 }
1959 }
1960
1961 //----------------------------inline_notify-----------------------------------*
1962 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1963 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1964 address func;
1965 if (id == vmIntrinsics::_notify) {
1966 func = OptoRuntime::monitor_notify_Java();
1967 } else {
1968 func = OptoRuntime::monitor_notifyAll_Java();
1969 }
1970 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1971 make_slow_call_ex(call, env()->Throwable_klass(), false);
1972 return true;
1973 }
1974
1975
1976 //----------------------------inline_min_max-----------------------------------
1977 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1978 Node* a = nullptr;
1979 Node* b = nullptr;
1980 Node* n = nullptr;
1981 switch (id) {
1982 case vmIntrinsics::_min:
1983 case vmIntrinsics::_max:
1984 case vmIntrinsics::_minF:
1985 case vmIntrinsics::_maxF:
1986 case vmIntrinsics::_minF_strict:
1987 case vmIntrinsics::_maxF_strict:
1988 case vmIntrinsics::_min_strict:
1989 case vmIntrinsics::_max_strict:
1990 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1991 a = argument(0);
1992 b = argument(1);
1993 break;
1994 case vmIntrinsics::_minD:
1995 case vmIntrinsics::_maxD:
1996 case vmIntrinsics::_minD_strict:
1997 case vmIntrinsics::_maxD_strict:
1998 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1999 a = argument(0);
2000 b = argument(2);
2001 break;
2002 case vmIntrinsics::_minL:
2003 case vmIntrinsics::_maxL:
2004 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2005 a = argument(0);
2006 b = argument(2);
2007 break;
2008 default:
2009 fatal_unexpected_iid(id);
2010 break;
2011 }
2012
2013 switch (id) {
2014 case vmIntrinsics::_min:
2015 case vmIntrinsics::_min_strict:
2016 n = new MinINode(a, b);
2017 break;
2018 case vmIntrinsics::_max:
2019 case vmIntrinsics::_max_strict:
2020 n = new MaxINode(a, b);
2021 break;
2022 case vmIntrinsics::_minF:
2023 case vmIntrinsics::_minF_strict:
2024 n = new MinFNode(a, b);
2025 break;
2026 case vmIntrinsics::_maxF:
2027 case vmIntrinsics::_maxF_strict:
2028 n = new MaxFNode(a, b);
2029 break;
2030 case vmIntrinsics::_minD:
2031 case vmIntrinsics::_minD_strict:
2032 n = new MinDNode(a, b);
2033 break;
2034 case vmIntrinsics::_maxD:
2035 case vmIntrinsics::_maxD_strict:
2036 n = new MaxDNode(a, b);
2037 break;
2038 case vmIntrinsics::_minL:
2039 n = new MinLNode(_gvn.C, a, b);
2040 break;
2041 case vmIntrinsics::_maxL:
2042 n = new MaxLNode(_gvn.C, a, b);
2043 break;
2044 default:
2045 fatal_unexpected_iid(id);
2046 break;
2047 }
2048
2049 set_result(_gvn.transform(n));
2050 return true;
2051 }
2052
2053 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2054 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2055 env()->ArithmeticException_instance())) {
2056 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2057 // so let's bail out intrinsic rather than risking deopting again.
2058 return false;
2059 }
2060
2061 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2062 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2063 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2064 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2065
2066 {
2067 PreserveJVMState pjvms(this);
2068 PreserveReexecuteState preexecs(this);
2069 jvms()->set_should_reexecute(true);
2070
2071 set_control(slow_path);
2072 set_i_o(i_o());
2073
2074 builtin_throw(Deoptimization::Reason_intrinsic,
2075 env()->ArithmeticException_instance(),
2076 /*allow_too_many_traps*/ false);
2077 }
2078
2079 set_control(fast_path);
2080 set_result(math);
2081 return true;
2082 }
2083
2084 template <typename OverflowOp>
2085 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2086 typedef typename OverflowOp::MathOp MathOp;
2087
2088 MathOp* mathOp = new MathOp(arg1, arg2);
2089 Node* operation = _gvn.transform( mathOp );
2090 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2091 return inline_math_mathExact(operation, ofcheck);
2092 }
2093
2094 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2095 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2096 }
2097
2098 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2099 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2100 }
2101
2102 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2103 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2104 }
2105
2106 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2107 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2108 }
2109
2110 bool LibraryCallKit::inline_math_negateExactI() {
2111 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2112 }
2113
2114 bool LibraryCallKit::inline_math_negateExactL() {
2115 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2116 }
2117
2118 bool LibraryCallKit::inline_math_multiplyExactI() {
2119 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2120 }
2121
2122 bool LibraryCallKit::inline_math_multiplyExactL() {
2123 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2124 }
2125
2126 bool LibraryCallKit::inline_math_multiplyHigh() {
2127 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2128 return true;
2129 }
2130
2131 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2132 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2133 return true;
2134 }
2135
2136 inline int
2137 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2138 const TypePtr* base_type = TypePtr::NULL_PTR;
2139 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2140 if (base_type == nullptr) {
2141 // Unknown type.
2142 return Type::AnyPtr;
2143 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2144 // Since this is a null+long form, we have to switch to a rawptr.
2145 base = _gvn.transform(new CastX2PNode(offset));
2146 offset = MakeConX(0);
2147 return Type::RawPtr;
2148 } else if (base_type->base() == Type::RawPtr) {
2149 return Type::RawPtr;
2150 } else if (base_type->isa_oopptr()) {
2151 // Base is never null => always a heap address.
2152 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2153 return Type::OopPtr;
2154 }
2155 // Offset is small => always a heap address.
2156 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2157 if (offset_type != nullptr &&
2158 base_type->offset() == 0 && // (should always be?)
2159 offset_type->_lo >= 0 &&
2160 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2161 return Type::OopPtr;
2162 } else if (type == T_OBJECT) {
2163 // off heap access to an oop doesn't make any sense. Has to be on
2164 // heap.
2165 return Type::OopPtr;
2166 }
2167 // Otherwise, it might either be oop+off or null+addr.
2168 return Type::AnyPtr;
2169 } else {
2170 // No information:
2171 return Type::AnyPtr;
2172 }
2173 }
2174
2175 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2176 Node* uncasted_base = base;
2177 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2178 if (kind == Type::RawPtr) {
2179 return basic_plus_adr(top(), uncasted_base, offset);
2180 } else if (kind == Type::AnyPtr) {
2181 assert(base == uncasted_base, "unexpected base change");
2182 if (can_cast) {
2183 if (!_gvn.type(base)->speculative_maybe_null() &&
2184 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2185 // According to profiling, this access is always on
2186 // heap. Casting the base to not null and thus avoiding membars
2187 // around the access should allow better optimizations
2188 Node* null_ctl = top();
2189 base = null_check_oop(base, &null_ctl, true, true, true);
2190 assert(null_ctl->is_top(), "no null control here");
2191 return basic_plus_adr(base, offset);
2192 } else if (_gvn.type(base)->speculative_always_null() &&
2193 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2194 // According to profiling, this access is always off
2195 // heap.
2196 base = null_assert(base);
2197 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2198 offset = MakeConX(0);
2199 return basic_plus_adr(top(), raw_base, offset);
2200 }
2201 }
2202 // We don't know if it's an on heap or off heap access. Fall back
2203 // to raw memory access.
2204 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2205 return basic_plus_adr(top(), raw, offset);
2206 } else {
2207 assert(base == uncasted_base, "unexpected base change");
2208 // We know it's an on heap access so base can't be null
2209 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2210 base = must_be_not_null(base, true);
2211 }
2212 return basic_plus_adr(base, offset);
2213 }
2214 }
2215
2216 //--------------------------inline_number_methods-----------------------------
2217 // inline int Integer.numberOfLeadingZeros(int)
2218 // inline int Long.numberOfLeadingZeros(long)
2219 //
2220 // inline int Integer.numberOfTrailingZeros(int)
2221 // inline int Long.numberOfTrailingZeros(long)
2222 //
2223 // inline int Integer.bitCount(int)
2224 // inline int Long.bitCount(long)
2225 //
2226 // inline char Character.reverseBytes(char)
2227 // inline short Short.reverseBytes(short)
2228 // inline int Integer.reverseBytes(int)
2229 // inline long Long.reverseBytes(long)
2230 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2231 Node* arg = argument(0);
2232 Node* n = nullptr;
2233 switch (id) {
2234 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2235 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2236 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2237 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2238 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2239 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2240 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2241 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2242 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2243 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2244 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2245 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2246 default: fatal_unexpected_iid(id); break;
2247 }
2248 set_result(_gvn.transform(n));
2249 return true;
2250 }
2251
2252 //--------------------------inline_bitshuffle_methods-----------------------------
2253 // inline int Integer.compress(int, int)
2254 // inline int Integer.expand(int, int)
2255 // inline long Long.compress(long, long)
2256 // inline long Long.expand(long, long)
2257 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2258 Node* n = nullptr;
2259 switch (id) {
2260 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2261 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2262 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2263 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2264 default: fatal_unexpected_iid(id); break;
2265 }
2266 set_result(_gvn.transform(n));
2267 return true;
2268 }
2269
2270 //--------------------------inline_number_methods-----------------------------
2271 // inline int Integer.compareUnsigned(int, int)
2272 // inline int Long.compareUnsigned(long, long)
2273 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2274 Node* arg1 = argument(0);
2275 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2276 Node* n = nullptr;
2277 switch (id) {
2278 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2279 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2280 default: fatal_unexpected_iid(id); break;
2281 }
2282 set_result(_gvn.transform(n));
2283 return true;
2284 }
2285
2286 //--------------------------inline_unsigned_divmod_methods-----------------------------
2287 // inline int Integer.divideUnsigned(int, int)
2288 // inline int Integer.remainderUnsigned(int, int)
2289 // inline long Long.divideUnsigned(long, long)
2290 // inline long Long.remainderUnsigned(long, long)
2291 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2292 Node* n = nullptr;
2293 switch (id) {
2294 case vmIntrinsics::_divideUnsigned_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 UDivINode(control(), argument(0), argument(1));
2301 break;
2302 }
2303 case vmIntrinsics::_divideUnsigned_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 UDivLNode(control(), argument(0), argument(2));
2310 break;
2311 }
2312 case vmIntrinsics::_remainderUnsigned_i: {
2313 zero_check_int(argument(1));
2314 // Compile-time detect of null-exception
2315 if (stopped()) {
2316 return true; // keep the graph constructed so far
2317 }
2318 n = new UModINode(control(), argument(0), argument(1));
2319 break;
2320 }
2321 case vmIntrinsics::_remainderUnsigned_l: {
2322 zero_check_long(argument(2));
2323 // Compile-time detect of null-exception
2324 if (stopped()) {
2325 return true; // keep the graph constructed so far
2326 }
2327 n = new UModLNode(control(), argument(0), argument(2));
2328 break;
2329 }
2330 default: fatal_unexpected_iid(id); break;
2331 }
2332 set_result(_gvn.transform(n));
2333 return true;
2334 }
2335
2336 //----------------------------inline_unsafe_access----------------------------
2337
2338 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2339 // Attempt to infer a sharper value type from the offset and base type.
2340 ciKlass* sharpened_klass = nullptr;
2341
2342 // See if it is an instance field, with an object type.
2343 if (alias_type->field() != nullptr) {
2344 if (alias_type->field()->type()->is_klass()) {
2345 sharpened_klass = alias_type->field()->type()->as_klass();
2346 }
2347 }
2348
2349 const TypeOopPtr* result = nullptr;
2350 // See if it is a narrow oop array.
2351 if (adr_type->isa_aryptr()) {
2352 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2353 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2354 if (elem_type != nullptr && elem_type->is_loaded()) {
2355 // Sharpen the value type.
2356 result = elem_type;
2357 }
2358 }
2359 }
2360
2361 // The sharpened class might be unloaded if there is no class loader
2362 // contraint in place.
2363 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2364 // Sharpen the value type.
2365 result = TypeOopPtr::make_from_klass(sharpened_klass);
2366 }
2367 if (result != nullptr) {
2368 #ifndef PRODUCT
2369 if (C->print_intrinsics() || C->print_inlining()) {
2370 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2371 tty->print(" sharpened value: "); result->dump(); tty->cr();
2372 }
2373 #endif
2374 }
2375 return result;
2376 }
2377
2378 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2379 switch (kind) {
2380 case Relaxed:
2381 return MO_UNORDERED;
2382 case Opaque:
2383 return MO_RELAXED;
2384 case Acquire:
2385 return MO_ACQUIRE;
2386 case Release:
2387 return MO_RELEASE;
2388 case Volatile:
2389 return MO_SEQ_CST;
2390 default:
2391 ShouldNotReachHere();
2392 return 0;
2393 }
2394 }
2395
2396 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2397 _kit(kit),
2398 _sp(kit->sp()),
2399 _jvms(kit->jvms()),
2400 _map(kit->clone_map()),
2401 _discarded(false)
2402 {
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()) {
2406 _ctrl_succ.push(out);
2407 }
2408 }
2409 }
2410
2411 LibraryCallKit::SavedState::~SavedState() {
2412 if (_discarded) {
2413 _kit->destruct_map_clone(_map);
2414 return;
2415 }
2416 _kit->jvms()->set_map(_map);
2417 _kit->jvms()->set_sp(_sp);
2418 _map->set_jvms(_kit->jvms());
2419 _kit->set_map(_map);
2420 _kit->set_sp(_sp);
2421 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2422 Node* out = _kit->control()->fast_out(i);
2423 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2424 _kit->_gvn.hash_delete(out);
2425 out->set_req(0, _kit->C->top());
2426 _kit->C->record_for_igvn(out);
2427 --i; --imax;
2428 _kit->_gvn.hash_find_insert(out);
2429 }
2430 }
2431 }
2432
2433 void LibraryCallKit::SavedState::discard() {
2434 _discarded = true;
2435 }
2436
2437 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2438 if (callee()->is_static()) return false; // caller must have the capability!
2439 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2440 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2441 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2442 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2443
2444 if (is_reference_type(type)) {
2445 decorators |= ON_UNKNOWN_OOP_REF;
2446 }
2447
2448 if (unaligned) {
2449 decorators |= C2_UNALIGNED;
2450 }
2451
2452 #ifndef PRODUCT
2453 {
2454 ResourceMark rm;
2455 // Check the signatures.
2456 ciSignature* sig = callee()->signature();
2457 #ifdef ASSERT
2458 if (!is_store) {
2459 // Object getReference(Object base, int/long offset), etc.
2460 BasicType rtype = sig->return_type()->basic_type();
2461 assert(rtype == type, "getter must return the expected value");
2462 assert(sig->count() == 2, "oop getter has 2 arguments");
2463 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2464 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2465 } else {
2466 // void putReference(Object base, int/long offset, Object x), etc.
2467 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2468 assert(sig->count() == 3, "oop putter has 3 arguments");
2469 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2470 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2471 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2472 assert(vtype == type, "putter must accept the expected value");
2473 }
2474 #endif // ASSERT
2475 }
2476 #endif //PRODUCT
2477
2478 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2479
2480 Node* receiver = argument(0); // type: oop
2481
2482 // Build address expression.
2483 Node* heap_base_oop = top();
2484
2485 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2486 Node* base = argument(1); // type: oop
2487 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2488 Node* offset = argument(2); // type: long
2489 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2490 // to be plain byte offsets, which are also the same as those accepted
2491 // by oopDesc::field_addr.
2492 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2493 "fieldOffset must be byte-scaled");
2494 // 32-bit machines ignore the high half!
2495 offset = ConvL2X(offset);
2496
2497 // Save state and restore on bailout
2498 SavedState old_state(this);
2499
2500 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2501 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2502
2503 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2504 if (type != T_OBJECT) {
2505 decorators |= IN_NATIVE; // off-heap primitive access
2506 } else {
2507 return false; // off-heap oop accesses are not supported
2508 }
2509 } else {
2510 heap_base_oop = base; // on-heap or mixed access
2511 }
2512
2513 // Can base be null? Otherwise, always on-heap access.
2514 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2515
2516 if (!can_access_non_heap) {
2517 decorators |= IN_HEAP;
2518 }
2519
2520 Node* val = is_store ? argument(4) : nullptr;
2521
2522 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2523 if (adr_type == TypePtr::NULL_PTR) {
2524 return false; // off-heap access with zero address
2525 }
2526
2527 // Try to categorize the address.
2528 Compile::AliasType* alias_type = C->alias_type(adr_type);
2529 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2530
2531 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2532 alias_type->adr_type() == TypeAryPtr::RANGE) {
2533 return false; // not supported
2534 }
2535
2536 bool mismatched = false;
2537 BasicType bt = alias_type->basic_type();
2538 if (bt != T_ILLEGAL) {
2539 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2540 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2541 // Alias type doesn't differentiate between byte[] and boolean[]).
2542 // Use address type to get the element type.
2543 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2544 }
2545 if (is_reference_type(bt, true)) {
2546 // accessing an array field with getReference is not a mismatch
2547 bt = T_OBJECT;
2548 }
2549 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2550 // Don't intrinsify mismatched object accesses
2551 return false;
2552 }
2553 mismatched = (bt != type);
2554 } else if (alias_type->adr_type()->isa_oopptr()) {
2555 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2556 }
2557
2558 old_state.discard();
2559 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2560
2561 if (mismatched) {
2562 decorators |= C2_MISMATCHED;
2563 }
2564
2565 // First guess at the value type.
2566 const Type *value_type = Type::get_const_basic_type(type);
2567
2568 // Figure out the memory ordering.
2569 decorators |= mo_decorator_for_access_kind(kind);
2570
2571 if (!is_store && type == T_OBJECT) {
2572 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2573 if (tjp != nullptr) {
2574 value_type = tjp;
2575 }
2576 }
2577
2578 receiver = null_check(receiver);
2579 if (stopped()) {
2580 return true;
2581 }
2582 // Heap pointers get a null-check from the interpreter,
2583 // as a courtesy. However, this is not guaranteed by Unsafe,
2584 // and it is not possible to fully distinguish unintended nulls
2585 // from intended ones in this API.
2586
2587 if (!is_store) {
2588 Node* p = nullptr;
2589 // Try to constant fold a load from a constant field
2590 ciField* field = alias_type->field();
2591 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2592 // final or stable field
2593 p = make_constant_from_field(field, heap_base_oop);
2594 }
2595
2596 if (p == nullptr) { // Could not constant fold the load
2597 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2598 // Normalize the value returned by getBoolean in the following cases
2599 if (type == T_BOOLEAN &&
2600 (mismatched ||
2601 heap_base_oop == top() || // - heap_base_oop is null or
2602 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2603 // and the unsafe access is made to large offset
2604 // (i.e., larger than the maximum offset necessary for any
2605 // field access)
2606 ) {
2607 IdealKit ideal = IdealKit(this);
2608 #define __ ideal.
2609 IdealVariable normalized_result(ideal);
2610 __ declarations_done();
2611 __ set(normalized_result, p);
2612 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2613 __ set(normalized_result, ideal.ConI(1));
2614 ideal.end_if();
2615 final_sync(ideal);
2616 p = __ value(normalized_result);
2617 #undef __
2618 }
2619 }
2620 if (type == T_ADDRESS) {
2621 p = gvn().transform(new CastP2XNode(nullptr, p));
2622 p = ConvX2UL(p);
2623 }
2624 // The load node has the control of the preceding MemBarCPUOrder. All
2625 // following nodes will have the control of the MemBarCPUOrder inserted at
2626 // the end of this method. So, pushing the load onto the stack at a later
2627 // point is fine.
2628 set_result(p);
2629 } else {
2630 if (bt == T_ADDRESS) {
2631 // Repackage the long as a pointer.
2632 val = ConvL2X(val);
2633 val = gvn().transform(new CastX2PNode(val));
2634 }
2635 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2636 }
2637
2638 return true;
2639 }
2640
2641 //----------------------------inline_unsafe_load_store----------------------------
2642 // This method serves a couple of different customers (depending on LoadStoreKind):
2643 //
2644 // LS_cmp_swap:
2645 //
2646 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2647 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2648 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2649 //
2650 // LS_cmp_swap_weak:
2651 //
2652 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2653 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2654 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2655 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2656 //
2657 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2658 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2659 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2660 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2661 //
2662 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2663 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2664 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2665 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2666 //
2667 // LS_cmp_exchange:
2668 //
2669 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2670 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2671 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2672 //
2673 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2674 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2675 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2676 //
2677 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2678 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2679 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2680 //
2681 // LS_get_add:
2682 //
2683 // int getAndAddInt( Object o, long offset, int delta)
2684 // long getAndAddLong(Object o, long offset, long delta)
2685 //
2686 // LS_get_set:
2687 //
2688 // int getAndSet(Object o, long offset, int newValue)
2689 // long getAndSet(Object o, long offset, long newValue)
2690 // Object getAndSet(Object o, long offset, Object newValue)
2691 //
2692 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2693 // This basic scheme here is the same as inline_unsafe_access, but
2694 // differs in enough details that combining them would make the code
2695 // overly confusing. (This is a true fact! I originally combined
2696 // them, but even I was confused by it!) As much code/comments as
2697 // possible are retained from inline_unsafe_access though to make
2698 // the correspondences clearer. - dl
2699
2700 if (callee()->is_static()) return false; // caller must have the capability!
2701
2702 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2703 decorators |= mo_decorator_for_access_kind(access_kind);
2704
2705 #ifndef PRODUCT
2706 BasicType rtype;
2707 {
2708 ResourceMark rm;
2709 // Check the signatures.
2710 ciSignature* sig = callee()->signature();
2711 rtype = sig->return_type()->basic_type();
2712 switch(kind) {
2713 case LS_get_add:
2714 case LS_get_set: {
2715 // Check the signatures.
2716 #ifdef ASSERT
2717 assert(rtype == type, "get and set must return the expected type");
2718 assert(sig->count() == 3, "get and set has 3 arguments");
2719 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2720 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2721 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2722 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2723 #endif // ASSERT
2724 break;
2725 }
2726 case LS_cmp_swap:
2727 case LS_cmp_swap_weak: {
2728 // Check the signatures.
2729 #ifdef ASSERT
2730 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2731 assert(sig->count() == 4, "CAS has 4 arguments");
2732 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2733 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2734 #endif // ASSERT
2735 break;
2736 }
2737 case LS_cmp_exchange: {
2738 // Check the signatures.
2739 #ifdef ASSERT
2740 assert(rtype == type, "CAS must return the expected type");
2741 assert(sig->count() == 4, "CAS has 4 arguments");
2742 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2743 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2744 #endif // ASSERT
2745 break;
2746 }
2747 default:
2748 ShouldNotReachHere();
2749 }
2750 }
2751 #endif //PRODUCT
2752
2753 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2754
2755 // Get arguments:
2756 Node* receiver = nullptr;
2757 Node* base = nullptr;
2758 Node* offset = nullptr;
2759 Node* oldval = nullptr;
2760 Node* newval = nullptr;
2761 switch(kind) {
2762 case LS_cmp_swap:
2763 case LS_cmp_swap_weak:
2764 case LS_cmp_exchange: {
2765 const bool two_slot_type = type2size[type] == 2;
2766 receiver = argument(0); // type: oop
2767 base = argument(1); // type: oop
2768 offset = argument(2); // type: long
2769 oldval = argument(4); // type: oop, int, or long
2770 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2771 break;
2772 }
2773 case LS_get_add:
2774 case LS_get_set: {
2775 receiver = argument(0); // type: oop
2776 base = argument(1); // type: oop
2777 offset = argument(2); // type: long
2778 oldval = nullptr;
2779 newval = argument(4); // type: oop, int, or long
2780 break;
2781 }
2782 default:
2783 ShouldNotReachHere();
2784 }
2785
2786 // Build field offset expression.
2787 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2788 // to be plain byte offsets, which are also the same as those accepted
2789 // by oopDesc::field_addr.
2790 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2791 // 32-bit machines ignore the high half of long offsets
2792 offset = ConvL2X(offset);
2793 // Save state and restore on bailout
2794 SavedState old_state(this);
2795 Node* adr = make_unsafe_address(base, offset,type, false);
2796 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2797
2798 Compile::AliasType* alias_type = C->alias_type(adr_type);
2799 BasicType bt = alias_type->basic_type();
2800 if (bt != T_ILLEGAL &&
2801 (is_reference_type(bt) != (type == T_OBJECT))) {
2802 // Don't intrinsify mismatched object accesses.
2803 return false;
2804 }
2805
2806 old_state.discard();
2807
2808 // For CAS, unlike inline_unsafe_access, there seems no point in
2809 // trying to refine types. Just use the coarse types here.
2810 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2811 const Type *value_type = Type::get_const_basic_type(type);
2812
2813 switch (kind) {
2814 case LS_get_set:
2815 case LS_cmp_exchange: {
2816 if (type == T_OBJECT) {
2817 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2818 if (tjp != nullptr) {
2819 value_type = tjp;
2820 }
2821 }
2822 break;
2823 }
2824 case LS_cmp_swap:
2825 case LS_cmp_swap_weak:
2826 case LS_get_add:
2827 break;
2828 default:
2829 ShouldNotReachHere();
2830 }
2831
2832 // Null check receiver.
2833 receiver = null_check(receiver);
2834 if (stopped()) {
2835 return true;
2836 }
2837
2838 int alias_idx = C->get_alias_index(adr_type);
2839
2840 if (is_reference_type(type)) {
2841 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2842
2843 // Transformation of a value which could be null pointer (CastPP #null)
2844 // could be delayed during Parse (for example, in adjust_map_after_if()).
2845 // Execute transformation here to avoid barrier generation in such case.
2846 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2847 newval = _gvn.makecon(TypePtr::NULL_PTR);
2848
2849 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2850 // Refine the value to a null constant, when it is known to be null
2851 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2852 }
2853 }
2854
2855 Node* result = nullptr;
2856 switch (kind) {
2857 case LS_cmp_exchange: {
2858 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2859 oldval, newval, value_type, type, decorators);
2860 break;
2861 }
2862 case LS_cmp_swap_weak:
2863 decorators |= C2_WEAK_CMPXCHG;
2864 case LS_cmp_swap: {
2865 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2866 oldval, newval, value_type, type, decorators);
2867 break;
2868 }
2869 case LS_get_set: {
2870 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2871 newval, value_type, type, decorators);
2872 break;
2873 }
2874 case LS_get_add: {
2875 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2876 newval, value_type, type, decorators);
2877 break;
2878 }
2879 default:
2880 ShouldNotReachHere();
2881 }
2882
2883 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2884 set_result(result);
2885 return true;
2886 }
2887
2888 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2889 // Regardless of form, don't allow previous ld/st to move down,
2890 // then issue acquire, release, or volatile mem_bar.
2891 insert_mem_bar(Op_MemBarCPUOrder);
2892 switch(id) {
2893 case vmIntrinsics::_loadFence:
2894 insert_mem_bar(Op_LoadFence);
2895 return true;
2896 case vmIntrinsics::_storeFence:
2897 insert_mem_bar(Op_StoreFence);
2898 return true;
2899 case vmIntrinsics::_storeStoreFence:
2900 insert_mem_bar(Op_StoreStoreFence);
2901 return true;
2902 case vmIntrinsics::_fullFence:
2903 insert_mem_bar(Op_MemBarVolatile);
2904 return true;
2905 default:
2906 fatal_unexpected_iid(id);
2907 return false;
2908 }
2909 }
2910
2911 bool LibraryCallKit::inline_onspinwait() {
2912 insert_mem_bar(Op_OnSpinWait);
2913 return true;
2914 }
2915
2916 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2917 if (!kls->is_Con()) {
2918 return true;
2919 }
2920 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2921 if (klsptr == nullptr) {
2922 return true;
2923 }
2924 ciInstanceKlass* ik = klsptr->instance_klass();
2925 // don't need a guard for a klass that is already initialized
2926 return !ik->is_initialized();
2927 }
2928
2929 //----------------------------inline_unsafe_writeback0-------------------------
2930 // public native void Unsafe.writeback0(long address)
2931 bool LibraryCallKit::inline_unsafe_writeback0() {
2932 if (!Matcher::has_match_rule(Op_CacheWB)) {
2933 return false;
2934 }
2935 #ifndef PRODUCT
2936 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2937 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2938 ciSignature* sig = callee()->signature();
2939 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2940 #endif
2941 null_check_receiver(); // null-check, then ignore
2942 Node *addr = argument(1);
2943 addr = new CastX2PNode(addr);
2944 addr = _gvn.transform(addr);
2945 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2946 flush = _gvn.transform(flush);
2947 set_memory(flush, TypeRawPtr::BOTTOM);
2948 return true;
2949 }
2950
2951 //----------------------------inline_unsafe_writeback0-------------------------
2952 // public native void Unsafe.writeback0(long address)
2953 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2954 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2955 return false;
2956 }
2957 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2958 return false;
2959 }
2960 #ifndef PRODUCT
2961 assert(Matcher::has_match_rule(Op_CacheWB),
2962 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2963 : "found match rule for CacheWBPostSync but not CacheWB"));
2964
2965 #endif
2966 null_check_receiver(); // null-check, then ignore
2967 Node *sync;
2968 if (is_pre) {
2969 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2970 } else {
2971 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2972 }
2973 sync = _gvn.transform(sync);
2974 set_memory(sync, TypeRawPtr::BOTTOM);
2975 return true;
2976 }
2977
2978 //----------------------------inline_unsafe_allocate---------------------------
2979 // public native Object Unsafe.allocateInstance(Class<?> cls);
2980 bool LibraryCallKit::inline_unsafe_allocate() {
2981
2982 #if INCLUDE_JVMTI
2983 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2984 return false;
2985 }
2986 #endif //INCLUDE_JVMTI
2987
2988 if (callee()->is_static()) return false; // caller must have the capability!
2989
2990 null_check_receiver(); // null-check, then ignore
2991 Node* cls = null_check(argument(1));
2992 if (stopped()) return true;
2993
2994 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2995 kls = null_check(kls);
2996 if (stopped()) return true; // argument was like int.class
2997
2998 #if INCLUDE_JVMTI
2999 // Don't try to access new allocated obj in the intrinsic.
3000 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3001 // Deoptimize and allocate in interpreter instead.
3002 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3003 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3004 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3005 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3006 {
3007 BuildCutout unless(this, tst, PROB_MAX);
3008 uncommon_trap(Deoptimization::Reason_intrinsic,
3009 Deoptimization::Action_make_not_entrant);
3010 }
3011 if (stopped()) {
3012 return true;
3013 }
3014 #endif //INCLUDE_JVMTI
3015
3016 Node* test = nullptr;
3017 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3018 // Note: The argument might still be an illegal value like
3019 // Serializable.class or Object[].class. The runtime will handle it.
3020 // But we must make an explicit check for initialization.
3021 Node* insp = basic_plus_adr(top(), kls, in_bytes(InstanceKlass::init_state_offset()));
3022 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3023 // can generate code to load it as unsigned byte.
3024 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3025 Node* bits = intcon(InstanceKlass::fully_initialized);
3026 test = _gvn.transform(new SubINode(inst, bits));
3027 // The 'test' is non-zero if we need to take a slow path.
3028 }
3029
3030 Node* obj = new_instance(kls, test);
3031 set_result(obj);
3032 return true;
3033 }
3034
3035 //------------------------inline_native_time_funcs--------------
3036 // inline code for System.currentTimeMillis() and System.nanoTime()
3037 // these have the same type and signature
3038 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3039 const TypeFunc* tf = OptoRuntime::void_long_Type();
3040 const TypePtr* no_memory_effects = nullptr;
3041 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3042 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3043 #ifdef ASSERT
3044 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3045 assert(value_top == top(), "second value must be top");
3046 #endif
3047 set_result(value);
3048 return true;
3049 }
3050
3051 //--------------------inline_native_vthread_start_transition--------------------
3052 // inline void startTransition(boolean is_mount);
3053 // inline void startFinalTransition();
3054 // Pseudocode of implementation:
3055 //
3056 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3057 // carrier->set_is_in_vthread_transition(true);
3058 // OrderAccess::storeload();
3059 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3060 // + global_vthread_transition_disable_count();
3061 // if (disable_requests > 0) {
3062 // slow path: runtime call
3063 // }
3064 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3065 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3066 IdealKit ideal(this);
3067
3068 Node* thread = ideal.thread();
3069 Node* jt_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3070 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3071 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3072 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3073 insert_mem_bar(Op_MemBarVolatile);
3074 ideal.sync_kit(this);
3075
3076 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3077 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3078 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3079 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3080 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3081
3082 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3083 sync_kit(ideal);
3084 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3085 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3086 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3087 ideal.sync_kit(this);
3088 }
3089 ideal.end_if();
3090
3091 final_sync(ideal);
3092 return true;
3093 }
3094
3095 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3096 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3097 IdealKit ideal(this);
3098
3099 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3100 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3101
3102 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3103 sync_kit(ideal);
3104 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3105 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3106 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3107 ideal.sync_kit(this);
3108 } ideal.else_(); {
3109 Node* thread = ideal.thread();
3110 Node* jt_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3111 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3112
3113 sync_kit(ideal);
3114 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3115 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3116 ideal.sync_kit(this);
3117 } ideal.end_if();
3118
3119 final_sync(ideal);
3120 return true;
3121 }
3122
3123 #if INCLUDE_JVMTI
3124
3125 // Always update the is_disable_suspend bit.
3126 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3127 if (!DoJVMTIVirtualThreadTransitions) {
3128 return true;
3129 }
3130 IdealKit ideal(this);
3131
3132 {
3133 // unconditionally update the is_disable_suspend bit in current JavaThread
3134 Node* thread = ideal.thread();
3135 Node* arg = _gvn.transform(argument(0)); // argument for notification
3136 Node* addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3137 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3138
3139 sync_kit(ideal);
3140 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3141 ideal.sync_kit(this);
3142 }
3143 final_sync(ideal);
3144
3145 return true;
3146 }
3147
3148 #endif // INCLUDE_JVMTI
3149
3150 #ifdef JFR_HAVE_INTRINSICS
3151
3152 /**
3153 * if oop->klass != null
3154 * // normal class
3155 * epoch = _epoch_state ? 2 : 1
3156 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3157 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3158 * }
3159 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3160 * else
3161 * // primitive class
3162 * if oop->array_klass != null
3163 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3164 * else
3165 * id = LAST_TYPE_ID + 1 // void class path
3166 * if (!signaled)
3167 * signaled = true
3168 */
3169 bool LibraryCallKit::inline_native_classID() {
3170 Node* cls = argument(0);
3171
3172 IdealKit ideal(this);
3173 #define __ ideal.
3174 IdealVariable result(ideal); __ declarations_done();
3175 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3176 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3177 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3178
3179
3180 __ if_then(kls, BoolTest::ne, null()); {
3181 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3182 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3183
3184 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3185 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3186 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3187 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3188 mask = _gvn.transform(new OrLNode(mask, epoch));
3189 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3190
3191 float unlikely = PROB_UNLIKELY(0.999);
3192 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3193 sync_kit(ideal);
3194 make_runtime_call(RC_LEAF,
3195 OptoRuntime::class_id_load_barrier_Type(),
3196 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3197 "class id load barrier",
3198 TypePtr::BOTTOM,
3199 kls);
3200 ideal.sync_kit(this);
3201 } __ end_if();
3202
3203 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3204 } __ else_(); {
3205 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3206 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3207 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3208 __ if_then(array_kls, BoolTest::ne, null()); {
3209 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3210 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3211 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3212 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3213 } __ else_(); {
3214 // void class case
3215 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3216 } __ end_if();
3217
3218 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3219 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3220 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3221 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3222 } __ end_if();
3223 } __ end_if();
3224
3225 final_sync(ideal);
3226 set_result(ideal.value(result));
3227 #undef __
3228 return true;
3229 }
3230
3231 //------------------------inline_native_jvm_commit------------------
3232 bool LibraryCallKit::inline_native_jvm_commit() {
3233 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3234
3235 // Save input memory and i_o state.
3236 Node* input_memory_state = reset_memory();
3237 set_all_memory(input_memory_state);
3238 Node* input_io_state = i_o();
3239
3240 // TLS.
3241 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3242 // Jfr java buffer.
3243 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3244 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3245 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3246
3247 // Load the current value of the notified field in the JfrThreadLocal.
3248 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3249 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3250
3251 // Test for notification.
3252 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3253 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3254 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3255
3256 // True branch, is notified.
3257 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3258 set_control(is_notified);
3259
3260 // Reset notified state.
3261 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3262 Node* notified_reset_memory = reset_memory();
3263
3264 // 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.
3265 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3266 // Convert the machine-word to a long.
3267 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3268
3269 // False branch, not notified.
3270 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3271 set_control(not_notified);
3272 set_all_memory(input_memory_state);
3273
3274 // Arg is the next position as a long.
3275 Node* arg = argument(0);
3276 // Convert long to machine-word.
3277 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3278
3279 // Store the next_position to the underlying jfr java buffer.
3280 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3281
3282 Node* commit_memory = reset_memory();
3283 set_all_memory(commit_memory);
3284
3285 // 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.
3286 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3287 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3288 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3289
3290 // And flags with lease constant.
3291 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3292
3293 // Branch on lease to conditionalize returning the leased java buffer.
3294 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3295 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3296 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3297
3298 // False branch, not a lease.
3299 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3300
3301 // True branch, is lease.
3302 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3303 set_control(is_lease);
3304
3305 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3306 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3307 OptoRuntime::void_void_Type(),
3308 SharedRuntime::jfr_return_lease(),
3309 "return_lease", TypePtr::BOTTOM);
3310 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3311
3312 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3313 record_for_igvn(lease_compare_rgn);
3314 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3315 record_for_igvn(lease_compare_mem);
3316 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3317 record_for_igvn(lease_compare_io);
3318 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3319 record_for_igvn(lease_result_value);
3320
3321 // Update control and phi nodes.
3322 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3323 lease_compare_rgn->init_req(_false_path, not_lease);
3324
3325 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3326 lease_compare_mem->init_req(_false_path, commit_memory);
3327
3328 lease_compare_io->init_req(_true_path, i_o());
3329 lease_compare_io->init_req(_false_path, input_io_state);
3330
3331 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3332 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3333
3334 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3335 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3336 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3337 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3338
3339 // Update control and phi nodes.
3340 result_rgn->init_req(_true_path, is_notified);
3341 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3342
3343 result_mem->init_req(_true_path, notified_reset_memory);
3344 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3345
3346 result_io->init_req(_true_path, input_io_state);
3347 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3348
3349 result_value->init_req(_true_path, current_pos);
3350 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3351
3352 // Set output state.
3353 set_control(_gvn.transform(result_rgn));
3354 set_all_memory(_gvn.transform(result_mem));
3355 set_i_o(_gvn.transform(result_io));
3356 set_result(result_rgn, result_value);
3357 return true;
3358 }
3359
3360 /*
3361 * The intrinsic is a model of this pseudo-code:
3362 *
3363 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3364 * jobject h_event_writer = tl->java_event_writer();
3365 * if (h_event_writer == nullptr) {
3366 * return nullptr;
3367 * }
3368 * oop threadObj = Thread::threadObj();
3369 * oop vthread = java_lang_Thread::vthread(threadObj);
3370 * traceid tid;
3371 * bool pinVirtualThread;
3372 * bool excluded;
3373 * if (vthread != threadObj) { // i.e. current thread is virtual
3374 * tid = java_lang_Thread::tid(vthread);
3375 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3376 * pinVirtualThread = VMContinuations;
3377 * excluded = vthread_epoch_raw & excluded_mask;
3378 * if (!excluded) {
3379 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3380 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3381 * if (vthread_epoch != current_epoch) {
3382 * write_checkpoint();
3383 * }
3384 * }
3385 * } else {
3386 * tid = java_lang_Thread::tid(threadObj);
3387 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3388 * pinVirtualThread = false;
3389 * excluded = thread_epoch_raw & excluded_mask;
3390 * }
3391 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3392 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3393 * if (tid_in_event_writer != tid) {
3394 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3395 * setField(event_writer, "excluded", excluded);
3396 * setField(event_writer, "threadID", tid);
3397 * }
3398 * return event_writer
3399 */
3400 bool LibraryCallKit::inline_native_getEventWriter() {
3401 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3402
3403 // Save input memory and i_o state.
3404 Node* input_memory_state = reset_memory();
3405 set_all_memory(input_memory_state);
3406 Node* input_io_state = i_o();
3407
3408 // The most significant bit of the u2 is used to denote thread exclusion
3409 Node* excluded_shift = _gvn.intcon(15);
3410 Node* excluded_mask = _gvn.intcon(1 << 15);
3411 // The epoch generation is the range [1-32767]
3412 Node* epoch_mask = _gvn.intcon(32767);
3413
3414 // TLS
3415 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3416
3417 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3418 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3419
3420 // Load the eventwriter jobject handle.
3421 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3422
3423 // Null check the jobject handle.
3424 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3425 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3426 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3427
3428 // False path, jobj is null.
3429 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3430
3431 // True path, jobj is not null.
3432 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3433
3434 set_control(jobj_is_not_null);
3435
3436 // Load the threadObj for the CarrierThread.
3437 Node* threadObj = generate_current_thread(tls_ptr);
3438
3439 // Load the vthread.
3440 Node* vthread = generate_virtual_thread(tls_ptr);
3441
3442 // If vthread != threadObj, this is a virtual thread.
3443 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3444 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3445 IfNode* iff_vthread_not_equal_threadObj =
3446 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3447
3448 // False branch, fallback to threadObj.
3449 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3450 set_control(vthread_equal_threadObj);
3451
3452 // Load the tid field from the vthread object.
3453 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3454
3455 // Load the raw epoch value from the threadObj.
3456 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3457 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3458 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3459 TypeInt::CHAR, T_CHAR,
3460 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3461
3462 // Mask off the excluded information from the epoch.
3463 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3464
3465 // True branch, this is a virtual thread.
3466 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3467 set_control(vthread_not_equal_threadObj);
3468
3469 // Load the tid field from the vthread object.
3470 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3471
3472 // Continuation support determines if a virtual thread should be pinned.
3473 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3474 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3475
3476 // Load the raw epoch value from the vthread.
3477 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3478 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3479 TypeInt::CHAR, T_CHAR,
3480 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3481
3482 // Mask off the excluded information from the epoch.
3483 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3484
3485 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3486 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3487 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3488 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3489
3490 // False branch, vthread is excluded, no need to write epoch info.
3491 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3492
3493 // True branch, vthread is included, update epoch info.
3494 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3495 set_control(included);
3496
3497 // Get epoch value.
3498 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3499
3500 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3501 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3502 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3503
3504 // Compare the epoch in the vthread to the current epoch generation.
3505 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3506 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3507 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3508
3509 // False path, epoch is equal, checkpoint information is valid.
3510 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3511
3512 // True path, epoch is not equal, write a checkpoint for the vthread.
3513 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3514
3515 set_control(epoch_is_not_equal);
3516
3517 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3518 // The call also updates the native thread local thread id and the vthread with the current epoch.
3519 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3520 OptoRuntime::jfr_write_checkpoint_Type(),
3521 SharedRuntime::jfr_write_checkpoint(),
3522 "write_checkpoint", TypePtr::BOTTOM);
3523 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3524
3525 // vthread epoch != current epoch
3526 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3527 record_for_igvn(epoch_compare_rgn);
3528 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3529 record_for_igvn(epoch_compare_mem);
3530 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3531 record_for_igvn(epoch_compare_io);
3532
3533 // Update control and phi nodes.
3534 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3535 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3536 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3537 epoch_compare_mem->init_req(_false_path, input_memory_state);
3538 epoch_compare_io->init_req(_true_path, i_o());
3539 epoch_compare_io->init_req(_false_path, input_io_state);
3540
3541 // excluded != true
3542 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3543 record_for_igvn(exclude_compare_rgn);
3544 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3545 record_for_igvn(exclude_compare_mem);
3546 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3547 record_for_igvn(exclude_compare_io);
3548
3549 // Update control and phi nodes.
3550 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3551 exclude_compare_rgn->init_req(_false_path, excluded);
3552 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3553 exclude_compare_mem->init_req(_false_path, input_memory_state);
3554 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3555 exclude_compare_io->init_req(_false_path, input_io_state);
3556
3557 // vthread != threadObj
3558 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3559 record_for_igvn(vthread_compare_rgn);
3560 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3561 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3562 record_for_igvn(vthread_compare_io);
3563 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3564 record_for_igvn(tid);
3565 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3566 record_for_igvn(exclusion);
3567 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3568 record_for_igvn(pinVirtualThread);
3569
3570 // Update control and phi nodes.
3571 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3572 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3573 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3574 vthread_compare_mem->init_req(_false_path, input_memory_state);
3575 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3576 vthread_compare_io->init_req(_false_path, input_io_state);
3577 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3578 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3579 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3580 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3581 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3582 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3583
3584 // Update branch state.
3585 set_control(_gvn.transform(vthread_compare_rgn));
3586 set_all_memory(_gvn.transform(vthread_compare_mem));
3587 set_i_o(_gvn.transform(vthread_compare_io));
3588
3589 // Load the event writer oop by dereferencing the jobject handle.
3590 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3591 assert(klass_EventWriter->is_loaded(), "invariant");
3592 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3593 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3594 const TypeOopPtr* const xtype = aklass->as_instance_type();
3595 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3596 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3597
3598 // Load the current thread id from the event writer object.
3599 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3600 // Get the field offset to, conditionally, store an updated tid value later.
3601 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3602 // Get the field offset to, conditionally, store an updated exclusion value later.
3603 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3604 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3605 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3606
3607 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3608 record_for_igvn(event_writer_tid_compare_rgn);
3609 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3610 record_for_igvn(event_writer_tid_compare_mem);
3611 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3612 record_for_igvn(event_writer_tid_compare_io);
3613
3614 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3615 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3616 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3617 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3618
3619 // False path, tids are the same.
3620 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3621
3622 // True path, tid is not equal, need to update the tid in the event writer.
3623 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3624 record_for_igvn(tid_is_not_equal);
3625
3626 // Store the pin state to the event writer.
3627 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3628
3629 // Store the exclusion state to the event writer.
3630 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3631 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3632
3633 // Store the tid to the event writer.
3634 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3635
3636 // Update control and phi nodes.
3637 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3638 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3639 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3640 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3641 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3642 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3643
3644 // Result of top level CFG, Memory, IO and Value.
3645 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3646 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3647 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3648 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3649
3650 // Result control.
3651 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3652 result_rgn->init_req(_false_path, jobj_is_null);
3653
3654 // Result memory.
3655 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3656 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3657
3658 // Result IO.
3659 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3660 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3661
3662 // Result value.
3663 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3664 result_value->init_req(_false_path, null()); // return null
3665
3666 // Set output state.
3667 set_control(_gvn.transform(result_rgn));
3668 set_all_memory(_gvn.transform(result_mem));
3669 set_i_o(_gvn.transform(result_io));
3670 set_result(result_rgn, result_value);
3671 return true;
3672 }
3673
3674 /*
3675 * The intrinsic is a model of this pseudo-code:
3676 *
3677 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3678 * if (carrierThread != thread) { // is virtual thread
3679 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3680 * bool excluded = vthread_epoch_raw & excluded_mask;
3681 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3682 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3683 * if (!excluded) {
3684 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3685 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3686 * }
3687 * AtomicAccess::release_store(&tl->_vthread, true);
3688 * return;
3689 * }
3690 * AtomicAccess::release_store(&tl->_vthread, false);
3691 */
3692 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3693 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3694
3695 Node* input_memory_state = reset_memory();
3696 set_all_memory(input_memory_state);
3697
3698 // The most significant bit of the u2 is used to denote thread exclusion
3699 Node* excluded_mask = _gvn.intcon(1 << 15);
3700 // The epoch generation is the range [1-32767]
3701 Node* epoch_mask = _gvn.intcon(32767);
3702
3703 Node* const carrierThread = generate_current_thread(jt);
3704 // If thread != carrierThread, this is a virtual thread.
3705 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3706 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3707 IfNode* iff_thread_not_equal_carrierThread =
3708 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3709
3710 Node* vthread_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3711
3712 // False branch, is carrierThread.
3713 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3714 // Store release
3715 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3716
3717 set_all_memory(input_memory_state);
3718
3719 // True branch, is virtual thread.
3720 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3721 set_control(thread_not_equal_carrierThread);
3722
3723 // Load the raw epoch value from the vthread.
3724 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3725 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3726 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3727
3728 // Mask off the excluded information from the epoch.
3729 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3730
3731 // Load the tid field from the thread.
3732 Node* tid = load_field_from_object(thread, "tid", "J");
3733
3734 // Store the vthread tid to the jfr thread local.
3735 Node* thread_id_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3736 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3737
3738 // Branch is_excluded to conditionalize updating the epoch .
3739 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3740 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3741 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3742
3743 // True branch, vthread is excluded, no need to write epoch info.
3744 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3745 set_control(excluded);
3746 Node* vthread_is_excluded = _gvn.intcon(1);
3747
3748 // False branch, vthread is included, update epoch info.
3749 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3750 set_control(included);
3751 Node* vthread_is_included = _gvn.intcon(0);
3752
3753 // Get epoch value.
3754 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3755
3756 // Store the vthread epoch to the jfr thread local.
3757 Node* vthread_epoch_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3758 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3759
3760 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3761 record_for_igvn(excluded_rgn);
3762 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3763 record_for_igvn(excluded_mem);
3764 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3765 record_for_igvn(exclusion);
3766
3767 // Merge the excluded control and memory.
3768 excluded_rgn->init_req(_true_path, excluded);
3769 excluded_rgn->init_req(_false_path, included);
3770 excluded_mem->init_req(_true_path, tid_memory);
3771 excluded_mem->init_req(_false_path, included_memory);
3772 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3773 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3774
3775 // Set intermediate state.
3776 set_control(_gvn.transform(excluded_rgn));
3777 set_all_memory(excluded_mem);
3778
3779 // Store the vthread exclusion state to the jfr thread local.
3780 Node* thread_local_excluded_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3781 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3782
3783 // Store release
3784 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3785
3786 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3787 record_for_igvn(thread_compare_rgn);
3788 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3789 record_for_igvn(thread_compare_mem);
3790 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3791 record_for_igvn(vthread);
3792
3793 // Merge the thread_compare control and memory.
3794 thread_compare_rgn->init_req(_true_path, control());
3795 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3796 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3797 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3798
3799 // Set output state.
3800 set_control(_gvn.transform(thread_compare_rgn));
3801 set_all_memory(_gvn.transform(thread_compare_mem));
3802 }
3803
3804 #endif // JFR_HAVE_INTRINSICS
3805
3806 //------------------------inline_native_currentCarrierThread------------------
3807 bool LibraryCallKit::inline_native_currentCarrierThread() {
3808 Node* junk = nullptr;
3809 set_result(generate_current_thread(junk));
3810 return true;
3811 }
3812
3813 //------------------------inline_native_currentThread------------------
3814 bool LibraryCallKit::inline_native_currentThread() {
3815 Node* junk = nullptr;
3816 set_result(generate_virtual_thread(junk));
3817 return true;
3818 }
3819
3820 //------------------------inline_native_setVthread------------------
3821 bool LibraryCallKit::inline_native_setCurrentThread() {
3822 assert(C->method()->changes_current_thread(),
3823 "method changes current Thread but is not annotated ChangesCurrentThread");
3824 Node* arr = argument(1);
3825 Node* thread = _gvn.transform(new ThreadLocalNode());
3826 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3827 Node* thread_obj_handle
3828 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3829 thread_obj_handle = _gvn.transform(thread_obj_handle);
3830 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3831 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3832
3833 // Change the _monitor_owner_id of the JavaThread
3834 Node* tid = load_field_from_object(arr, "tid", "J");
3835 Node* monitor_owner_id_offset = basic_plus_adr(top(), thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3836 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3837
3838 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3839 return true;
3840 }
3841
3842 const Type* LibraryCallKit::scopedValueCache_type() {
3843 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3844 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3845 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3846
3847 // Because we create the scopedValue cache lazily we have to make the
3848 // type of the result BotPTR.
3849 bool xk = etype->klass_is_exact();
3850 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3851 return objects_type;
3852 }
3853
3854 Node* LibraryCallKit::scopedValueCache_helper() {
3855 Node* thread = _gvn.transform(new ThreadLocalNode());
3856 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3857 // We cannot use immutable_memory() because we might flip onto a
3858 // different carrier thread, at which point we'll need to use that
3859 // carrier thread's cache.
3860 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3861 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3862 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3863 }
3864
3865 //------------------------inline_native_scopedValueCache------------------
3866 bool LibraryCallKit::inline_native_scopedValueCache() {
3867 Node* cache_obj_handle = scopedValueCache_helper();
3868 const Type* objects_type = scopedValueCache_type();
3869 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3870
3871 return true;
3872 }
3873
3874 //------------------------inline_native_setScopedValueCache------------------
3875 bool LibraryCallKit::inline_native_setScopedValueCache() {
3876 Node* arr = argument(0);
3877 Node* cache_obj_handle = scopedValueCache_helper();
3878 const Type* objects_type = scopedValueCache_type();
3879
3880 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3881 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3882
3883 return true;
3884 }
3885
3886 //------------------------inline_native_Continuation_pin and unpin-----------
3887
3888 // Shared implementation routine for both pin and unpin.
3889 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3890 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3891
3892 // Save input memory.
3893 Node* input_memory_state = reset_memory();
3894 set_all_memory(input_memory_state);
3895
3896 // TLS
3897 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3898 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3899 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3900
3901 // Null check the last continuation object.
3902 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3903 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3904 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3905
3906 // False path, last continuation is null.
3907 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3908
3909 // True path, last continuation is not null.
3910 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3911
3912 set_control(continuation_is_not_null);
3913
3914 // Load the pin count from the last continuation.
3915 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3916 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3917
3918 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3919 Node* pin_count_rhs;
3920 if (unpin) {
3921 pin_count_rhs = _gvn.intcon(0);
3922 } else {
3923 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3924 }
3925 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3926 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3927 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3928
3929 // True branch, pin count over/underflow.
3930 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3931 {
3932 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3933 // which will throw IllegalStateException for pin count over/underflow.
3934 // No memory changed so far - we can use memory create by reset_memory()
3935 // at the beginning of this intrinsic. No need to call reset_memory() again.
3936 PreserveJVMState pjvms(this);
3937 set_control(pin_count_over_underflow);
3938 uncommon_trap(Deoptimization::Reason_intrinsic,
3939 Deoptimization::Action_none);
3940 assert(stopped(), "invariant");
3941 }
3942
3943 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3944 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3945 set_control(valid_pin_count);
3946
3947 Node* next_pin_count;
3948 if (unpin) {
3949 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3950 } else {
3951 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3952 }
3953
3954 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3955
3956 // Result of top level CFG and Memory.
3957 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3958 record_for_igvn(result_rgn);
3959 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3960 record_for_igvn(result_mem);
3961
3962 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3963 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3964 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3965 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3966
3967 // Set output state.
3968 set_control(_gvn.transform(result_rgn));
3969 set_all_memory(_gvn.transform(result_mem));
3970
3971 return true;
3972 }
3973
3974 //---------------------------load_mirror_from_klass----------------------------
3975 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3976 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3977 Node* p = basic_plus_adr(top(), klass, in_bytes(Klass::java_mirror_offset()));
3978 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3979 // mirror = ((OopHandle)mirror)->resolve();
3980 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3981 }
3982
3983 //-----------------------load_klass_from_mirror_common-------------------------
3984 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3985 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3986 // and branch to the given path on the region.
3987 // If never_see_null, take an uncommon trap on null, so we can optimistically
3988 // compile for the non-null case.
3989 // If the region is null, force never_see_null = true.
3990 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3991 bool never_see_null,
3992 RegionNode* region,
3993 int null_path,
3994 int offset) {
3995 if (region == nullptr) never_see_null = true;
3996 Node* p = basic_plus_adr(mirror, offset);
3997 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3998 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3999 Node* null_ctl = top();
4000 kls = null_check_oop(kls, &null_ctl, never_see_null);
4001 if (region != nullptr) {
4002 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4003 region->init_req(null_path, null_ctl);
4004 } else {
4005 assert(null_ctl == top(), "no loose ends");
4006 }
4007 return kls;
4008 }
4009
4010 //--------------------(inline_native_Class_query helpers)---------------------
4011 // Use this for JVM_ACC_INTERFACE.
4012 // Fall through if (mods & mask) == bits, take the guard otherwise.
4013 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4014 ByteSize offset, const Type* type, BasicType bt) {
4015 // Branch around if the given klass has the given modifier bit set.
4016 // Like generate_guard, adds a new path onto the region.
4017 Node* modp = basic_plus_adr(top(), kls, in_bytes(offset));
4018 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4019 Node* mask = intcon(modifier_mask);
4020 Node* bits = intcon(modifier_bits);
4021 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4022 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4023 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4024 return generate_fair_guard(bol, region);
4025 }
4026 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4027 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4028 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4029 }
4030
4031 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4032 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4033 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4034 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4035 }
4036
4037 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4038 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4039 }
4040
4041 //-------------------------inline_native_Class_query-------------------
4042 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4043 const Type* return_type = TypeInt::BOOL;
4044 Node* prim_return_value = top(); // what happens if it's a primitive class?
4045 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4046 bool expect_prim = false; // most of these guys expect to work on refs
4047
4048 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4049
4050 Node* mirror = argument(0);
4051 Node* obj = top();
4052
4053 switch (id) {
4054 case vmIntrinsics::_isInstance:
4055 // nothing is an instance of a primitive type
4056 prim_return_value = intcon(0);
4057 obj = argument(1);
4058 break;
4059 case vmIntrinsics::_isHidden:
4060 prim_return_value = intcon(0);
4061 break;
4062 case vmIntrinsics::_getSuperclass:
4063 prim_return_value = null();
4064 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4065 break;
4066 default:
4067 fatal_unexpected_iid(id);
4068 break;
4069 }
4070
4071 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4072 if (mirror_con == nullptr) return false; // cannot happen?
4073
4074 #ifndef PRODUCT
4075 if (C->print_intrinsics() || C->print_inlining()) {
4076 ciType* k = mirror_con->java_mirror_type();
4077 if (k) {
4078 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4079 k->print_name();
4080 tty->cr();
4081 }
4082 }
4083 #endif
4084
4085 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4086 RegionNode* region = new RegionNode(PATH_LIMIT);
4087 record_for_igvn(region);
4088 PhiNode* phi = new PhiNode(region, return_type);
4089
4090 // The mirror will never be null of Reflection.getClassAccessFlags, however
4091 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4092 // if it is. See bug 4774291.
4093
4094 // For Reflection.getClassAccessFlags(), the null check occurs in
4095 // the wrong place; see inline_unsafe_access(), above, for a similar
4096 // situation.
4097 mirror = null_check(mirror);
4098 // If mirror or obj is dead, only null-path is taken.
4099 if (stopped()) return true;
4100
4101 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4102
4103 // Now load the mirror's klass metaobject, and null-check it.
4104 // Side-effects region with the control path if the klass is null.
4105 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4106 // If kls is null, we have a primitive mirror.
4107 phi->init_req(_prim_path, prim_return_value);
4108 if (stopped()) { set_result(region, phi); return true; }
4109 bool safe_for_replace = (region->in(_prim_path) == top());
4110
4111 Node* p; // handy temp
4112 Node* null_ctl;
4113
4114 // Now that we have the non-null klass, we can perform the real query.
4115 // For constant classes, the query will constant-fold in LoadNode::Value.
4116 Node* query_value = top();
4117 switch (id) {
4118 case vmIntrinsics::_isInstance:
4119 // nothing is an instance of a primitive type
4120 query_value = gen_instanceof(obj, kls, safe_for_replace);
4121 break;
4122
4123 case vmIntrinsics::_isHidden:
4124 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4125 if (generate_hidden_class_guard(kls, region) != nullptr)
4126 // A guard was added. If the guard is taken, it was an hidden class.
4127 phi->add_req(intcon(1));
4128 // If we fall through, it's a plain class.
4129 query_value = intcon(0);
4130 break;
4131
4132
4133 case vmIntrinsics::_getSuperclass:
4134 // The rules here are somewhat unfortunate, but we can still do better
4135 // with random logic than with a JNI call.
4136 // Interfaces store null or Object as _super, but must report null.
4137 // Arrays store an intermediate super as _super, but must report Object.
4138 // Other types can report the actual _super.
4139 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4140 if (generate_array_guard(kls, region) != nullptr) {
4141 // A guard was added. If the guard is taken, it was an array.
4142 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4143 }
4144 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4145 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4146 if (generate_interface_guard(kls, region) != nullptr) {
4147 // A guard was added. If the guard is taken, it was an interface.
4148 phi->add_req(null());
4149 }
4150 // If we fall through, it's a plain class. Get its _super.
4151 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4152 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4153 null_ctl = top();
4154 kls = null_check_oop(kls, &null_ctl);
4155 if (null_ctl != top()) {
4156 // If the guard is taken, Object.superClass is null (both klass and mirror).
4157 region->add_req(null_ctl);
4158 phi ->add_req(null());
4159 }
4160 if (!stopped()) {
4161 query_value = load_mirror_from_klass(kls);
4162 }
4163 break;
4164
4165 default:
4166 fatal_unexpected_iid(id);
4167 break;
4168 }
4169
4170 // Fall-through is the normal case of a query to a real class.
4171 phi->init_req(1, query_value);
4172 region->init_req(1, control());
4173
4174 C->set_has_split_ifs(true); // Has chance for split-if optimization
4175 set_result(region, phi);
4176 return true;
4177 }
4178
4179 //-------------------------inline_Class_cast-------------------
4180 bool LibraryCallKit::inline_Class_cast() {
4181 Node* mirror = argument(0); // Class
4182 Node* obj = argument(1);
4183 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4184 if (mirror_con == nullptr) {
4185 return false; // dead path (mirror->is_top()).
4186 }
4187 if (obj == nullptr || obj->is_top()) {
4188 return false; // dead path
4189 }
4190 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4191
4192 // First, see if Class.cast() can be folded statically.
4193 // java_mirror_type() returns non-null for compile-time Class constants.
4194 ciType* tm = mirror_con->java_mirror_type();
4195 if (tm != nullptr && tm->is_klass() &&
4196 tp != nullptr) {
4197 if (!tp->is_loaded()) {
4198 // Don't use intrinsic when class is not loaded.
4199 return false;
4200 } else {
4201 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4202 if (static_res == Compile::SSC_always_true) {
4203 // isInstance() is true - fold the code.
4204 set_result(obj);
4205 return true;
4206 } else if (static_res == Compile::SSC_always_false) {
4207 // Don't use intrinsic, have to throw ClassCastException.
4208 // If the reference is null, the non-intrinsic bytecode will
4209 // be optimized appropriately.
4210 return false;
4211 }
4212 }
4213 }
4214
4215 // Bailout intrinsic and do normal inlining if exception path is frequent.
4216 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4217 return false;
4218 }
4219
4220 // Generate dynamic checks.
4221 // Class.cast() is java implementation of _checkcast bytecode.
4222 // Do checkcast (Parse::do_checkcast()) optimizations here.
4223
4224 mirror = null_check(mirror);
4225 // If mirror is dead, only null-path is taken.
4226 if (stopped()) {
4227 return true;
4228 }
4229
4230 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4231 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4232 RegionNode* region = new RegionNode(PATH_LIMIT);
4233 record_for_igvn(region);
4234
4235 // Now load the mirror's klass metaobject, and null-check it.
4236 // If kls is null, we have a primitive mirror and
4237 // nothing is an instance of a primitive type.
4238 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4239
4240 Node* res = top();
4241 if (!stopped()) {
4242 Node* bad_type_ctrl = top();
4243 // Do checkcast optimizations.
4244 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4245 region->init_req(_bad_type_path, bad_type_ctrl);
4246 }
4247 if (region->in(_prim_path) != top() ||
4248 region->in(_bad_type_path) != top()) {
4249 // Let Interpreter throw ClassCastException.
4250 PreserveJVMState pjvms(this);
4251 set_control(_gvn.transform(region));
4252 uncommon_trap(Deoptimization::Reason_intrinsic,
4253 Deoptimization::Action_maybe_recompile);
4254 }
4255 if (!stopped()) {
4256 set_result(res);
4257 }
4258 return true;
4259 }
4260
4261
4262 //--------------------------inline_native_subtype_check------------------------
4263 // This intrinsic takes the JNI calls out of the heart of
4264 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4265 bool LibraryCallKit::inline_native_subtype_check() {
4266 // Pull both arguments off the stack.
4267 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4268 args[0] = argument(0);
4269 args[1] = argument(1);
4270 Node* klasses[2]; // corresponding Klasses: superk, subk
4271 klasses[0] = klasses[1] = top();
4272
4273 enum {
4274 // A full decision tree on {superc is prim, subc is prim}:
4275 _prim_0_path = 1, // {P,N} => false
4276 // {P,P} & superc!=subc => false
4277 _prim_same_path, // {P,P} & superc==subc => true
4278 _prim_1_path, // {N,P} => false
4279 _ref_subtype_path, // {N,N} & subtype check wins => true
4280 _both_ref_path, // {N,N} & subtype check loses => false
4281 PATH_LIMIT
4282 };
4283
4284 RegionNode* region = new RegionNode(PATH_LIMIT);
4285 Node* phi = new PhiNode(region, TypeInt::BOOL);
4286 record_for_igvn(region);
4287
4288 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4289 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4290 int class_klass_offset = java_lang_Class::klass_offset();
4291
4292 // First null-check both mirrors and load each mirror's klass metaobject.
4293 int which_arg;
4294 for (which_arg = 0; which_arg <= 1; which_arg++) {
4295 Node* arg = args[which_arg];
4296 arg = null_check(arg);
4297 if (stopped()) break;
4298 args[which_arg] = arg;
4299
4300 Node* p = basic_plus_adr(arg, class_klass_offset);
4301 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4302 klasses[which_arg] = _gvn.transform(kls);
4303 }
4304
4305 // Having loaded both klasses, test each for null.
4306 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4307 for (which_arg = 0; which_arg <= 1; which_arg++) {
4308 Node* kls = klasses[which_arg];
4309 Node* null_ctl = top();
4310 kls = null_check_oop(kls, &null_ctl, never_see_null);
4311 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4312 region->init_req(prim_path, null_ctl);
4313 if (stopped()) break;
4314 klasses[which_arg] = kls;
4315 }
4316
4317 if (!stopped()) {
4318 // now we have two reference types, in klasses[0..1]
4319 Node* subk = klasses[1]; // the argument to isAssignableFrom
4320 Node* superk = klasses[0]; // the receiver
4321 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4322 // now we have a successful reference subtype check
4323 region->set_req(_ref_subtype_path, control());
4324 }
4325
4326 // If both operands are primitive (both klasses null), then
4327 // we must return true when they are identical primitives.
4328 // It is convenient to test this after the first null klass check.
4329 set_control(region->in(_prim_0_path)); // go back to first null check
4330 if (!stopped()) {
4331 // Since superc is primitive, make a guard for the superc==subc case.
4332 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4333 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4334 generate_guard(bol_eq, region, PROB_FAIR);
4335 if (region->req() == PATH_LIMIT+1) {
4336 // A guard was added. If the added guard is taken, superc==subc.
4337 region->swap_edges(PATH_LIMIT, _prim_same_path);
4338 region->del_req(PATH_LIMIT);
4339 }
4340 region->set_req(_prim_0_path, control()); // Not equal after all.
4341 }
4342
4343 // these are the only paths that produce 'true':
4344 phi->set_req(_prim_same_path, intcon(1));
4345 phi->set_req(_ref_subtype_path, intcon(1));
4346
4347 // pull together the cases:
4348 assert(region->req() == PATH_LIMIT, "sane region");
4349 for (uint i = 1; i < region->req(); i++) {
4350 Node* ctl = region->in(i);
4351 if (ctl == nullptr || ctl == top()) {
4352 region->set_req(i, top());
4353 phi ->set_req(i, top());
4354 } else if (phi->in(i) == nullptr) {
4355 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4356 }
4357 }
4358
4359 set_control(_gvn.transform(region));
4360 set_result(_gvn.transform(phi));
4361 return true;
4362 }
4363
4364 //---------------------generate_array_guard_common------------------------
4365 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4366 bool obj_array, bool not_array, Node** obj) {
4367
4368 if (stopped()) {
4369 return nullptr;
4370 }
4371
4372 // If obj_array/non_array==false/false:
4373 // Branch around if the given klass is in fact an array (either obj or prim).
4374 // If obj_array/non_array==false/true:
4375 // Branch around if the given klass is not an array klass of any kind.
4376 // If obj_array/non_array==true/true:
4377 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4378 // If obj_array/non_array==true/false:
4379 // Branch around if the kls is an oop array (Object[] or subtype)
4380 //
4381 // Like generate_guard, adds a new path onto the region.
4382 jint layout_con = 0;
4383 Node* layout_val = get_layout_helper(kls, layout_con);
4384 if (layout_val == nullptr) {
4385 bool query = (obj_array
4386 ? Klass::layout_helper_is_objArray(layout_con)
4387 : Klass::layout_helper_is_array(layout_con));
4388 if (query == not_array) {
4389 return nullptr; // never a branch
4390 } else { // always a branch
4391 Node* always_branch = control();
4392 if (region != nullptr)
4393 region->add_req(always_branch);
4394 set_control(top());
4395 return always_branch;
4396 }
4397 }
4398 // Now test the correct condition.
4399 jint nval = (obj_array
4400 ? (jint)(Klass::_lh_array_tag_type_value
4401 << Klass::_lh_array_tag_shift)
4402 : Klass::_lh_neutral_value);
4403 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4404 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4405 // invert the test if we are looking for a non-array
4406 if (not_array) btest = BoolTest(btest).negate();
4407 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4408 Node* ctrl = generate_fair_guard(bol, region);
4409 Node* is_array_ctrl = not_array ? control() : ctrl;
4410 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4411 // Keep track of the fact that 'obj' is an array to prevent
4412 // array specific accesses from floating above the guard.
4413 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4414 }
4415 return ctrl;
4416 }
4417
4418
4419 //-----------------------inline_native_newArray--------------------------
4420 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4421 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4422 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4423 Node* mirror;
4424 Node* count_val;
4425 if (uninitialized) {
4426 null_check_receiver();
4427 mirror = argument(1);
4428 count_val = argument(2);
4429 } else {
4430 mirror = argument(0);
4431 count_val = argument(1);
4432 }
4433
4434 mirror = null_check(mirror);
4435 // If mirror or obj is dead, only null-path is taken.
4436 if (stopped()) return true;
4437
4438 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4439 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4440 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4441 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4442 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4443
4444 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4445 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4446 result_reg, _slow_path);
4447 Node* normal_ctl = control();
4448 Node* no_array_ctl = result_reg->in(_slow_path);
4449
4450 // Generate code for the slow case. We make a call to newArray().
4451 set_control(no_array_ctl);
4452 if (!stopped()) {
4453 // Either the input type is void.class, or else the
4454 // array klass has not yet been cached. Either the
4455 // ensuing call will throw an exception, or else it
4456 // will cache the array klass for next time.
4457 PreserveJVMState pjvms(this);
4458 CallJavaNode* slow_call = nullptr;
4459 if (uninitialized) {
4460 // Generate optimized virtual call (holder class 'Unsafe' is final)
4461 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4462 } else {
4463 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4464 }
4465 Node* slow_result = set_results_for_java_call(slow_call);
4466 // this->control() comes from set_results_for_java_call
4467 result_reg->set_req(_slow_path, control());
4468 result_val->set_req(_slow_path, slow_result);
4469 result_io ->set_req(_slow_path, i_o());
4470 result_mem->set_req(_slow_path, reset_memory());
4471 }
4472
4473 set_control(normal_ctl);
4474 if (!stopped()) {
4475 // Normal case: The array type has been cached in the java.lang.Class.
4476 // The following call works fine even if the array type is polymorphic.
4477 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4478 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4479 result_reg->init_req(_normal_path, control());
4480 result_val->init_req(_normal_path, obj);
4481 result_io ->init_req(_normal_path, i_o());
4482 result_mem->init_req(_normal_path, reset_memory());
4483
4484 if (uninitialized) {
4485 // Mark the allocation so that zeroing is skipped
4486 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4487 alloc->maybe_set_complete(&_gvn);
4488 }
4489 }
4490
4491 // Return the combined state.
4492 set_i_o( _gvn.transform(result_io) );
4493 set_all_memory( _gvn.transform(result_mem));
4494
4495 C->set_has_split_ifs(true); // Has chance for split-if optimization
4496 set_result(result_reg, result_val);
4497 return true;
4498 }
4499
4500 //----------------------inline_native_getLength--------------------------
4501 // public static native int java.lang.reflect.Array.getLength(Object array);
4502 bool LibraryCallKit::inline_native_getLength() {
4503 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4504
4505 Node* array = null_check(argument(0));
4506 // If array is dead, only null-path is taken.
4507 if (stopped()) return true;
4508
4509 // Deoptimize if it is a non-array.
4510 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4511
4512 if (non_array != nullptr) {
4513 PreserveJVMState pjvms(this);
4514 set_control(non_array);
4515 uncommon_trap(Deoptimization::Reason_intrinsic,
4516 Deoptimization::Action_maybe_recompile);
4517 }
4518
4519 // If control is dead, only non-array-path is taken.
4520 if (stopped()) return true;
4521
4522 // The works fine even if the array type is polymorphic.
4523 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4524 Node* result = load_array_length(array);
4525
4526 C->set_has_split_ifs(true); // Has chance for split-if optimization
4527 set_result(result);
4528 return true;
4529 }
4530
4531 //------------------------inline_array_copyOf----------------------------
4532 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4533 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4534 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4535 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4536
4537 // Get the arguments.
4538 Node* original = argument(0);
4539 Node* start = is_copyOfRange? argument(1): intcon(0);
4540 Node* end = is_copyOfRange? argument(2): argument(1);
4541 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4542
4543 Node* newcopy = nullptr;
4544
4545 // Set the original stack and the reexecute bit for the interpreter to reexecute
4546 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4547 { PreserveReexecuteState preexecs(this);
4548 jvms()->set_should_reexecute(true);
4549
4550 array_type_mirror = null_check(array_type_mirror);
4551 original = null_check(original);
4552
4553 // Check if a null path was taken unconditionally.
4554 if (stopped()) return true;
4555
4556 Node* orig_length = load_array_length(original);
4557
4558 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4559 klass_node = null_check(klass_node);
4560
4561 RegionNode* bailout = new RegionNode(1);
4562 record_for_igvn(bailout);
4563
4564 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4565 // Bail out if that is so.
4566 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4567 if (not_objArray != nullptr) {
4568 // Improve the klass node's type from the new optimistic assumption:
4569 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4570 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4571 Node* cast = new CastPPNode(control(), klass_node, akls);
4572 klass_node = _gvn.transform(cast);
4573 }
4574
4575 // Bail out if either start or end is negative.
4576 generate_negative_guard(start, bailout, &start);
4577 generate_negative_guard(end, bailout, &end);
4578
4579 Node* length = end;
4580 if (_gvn.type(start) != TypeInt::ZERO) {
4581 length = _gvn.transform(new SubINode(end, start));
4582 }
4583
4584 // Bail out if length is negative (i.e., if start > end).
4585 // Without this the new_array would throw
4586 // NegativeArraySizeException but IllegalArgumentException is what
4587 // should be thrown
4588 generate_negative_guard(length, bailout, &length);
4589
4590 // Bail out if start is larger than the original length
4591 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4592 generate_negative_guard(orig_tail, bailout, &orig_tail);
4593
4594 if (bailout->req() > 1) {
4595 PreserveJVMState pjvms(this);
4596 set_control(_gvn.transform(bailout));
4597 uncommon_trap(Deoptimization::Reason_intrinsic,
4598 Deoptimization::Action_maybe_recompile);
4599 }
4600
4601 if (!stopped()) {
4602 // How many elements will we copy from the original?
4603 // The answer is MinI(orig_tail, length).
4604 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4605
4606 // Generate a direct call to the right arraycopy function(s).
4607 // We know the copy is disjoint but we might not know if the
4608 // oop stores need checking.
4609 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4610 // This will fail a store-check if x contains any non-nulls.
4611
4612 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4613 // loads/stores but it is legal only if we're sure the
4614 // Arrays.copyOf would succeed. So we need all input arguments
4615 // to the copyOf to be validated, including that the copy to the
4616 // new array won't trigger an ArrayStoreException. That subtype
4617 // check can be optimized if we know something on the type of
4618 // the input array from type speculation.
4619 if (_gvn.type(klass_node)->singleton()) {
4620 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4621 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4622
4623 int test = C->static_subtype_check(superk, subk);
4624 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4625 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4626 if (t_original->speculative_type() != nullptr) {
4627 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4628 }
4629 }
4630 }
4631
4632 bool validated = false;
4633 // Reason_class_check rather than Reason_intrinsic because we
4634 // want to intrinsify even if this traps.
4635 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4636 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4637
4638 if (not_subtype_ctrl != top()) {
4639 PreserveJVMState pjvms(this);
4640 set_control(not_subtype_ctrl);
4641 uncommon_trap(Deoptimization::Reason_class_check,
4642 Deoptimization::Action_make_not_entrant);
4643 assert(stopped(), "Should be stopped");
4644 }
4645 validated = true;
4646 }
4647
4648 if (!stopped()) {
4649 newcopy = new_array(klass_node, length, 0); // no arguments to push
4650
4651 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4652 load_object_klass(original), klass_node);
4653 if (!is_copyOfRange) {
4654 ac->set_copyof(validated);
4655 } else {
4656 ac->set_copyofrange(validated);
4657 }
4658 Node* n = _gvn.transform(ac);
4659 if (n == ac) {
4660 ac->connect_outputs(this);
4661 } else {
4662 assert(validated, "shouldn't transform if all arguments not validated");
4663 set_all_memory(n);
4664 }
4665 }
4666 }
4667 } // original reexecute is set back here
4668
4669 C->set_has_split_ifs(true); // Has chance for split-if optimization
4670 if (!stopped()) {
4671 set_result(newcopy);
4672 }
4673 return true;
4674 }
4675
4676
4677 //----------------------generate_virtual_guard---------------------------
4678 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4679 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4680 RegionNode* slow_region) {
4681 ciMethod* method = callee();
4682 int vtable_index = method->vtable_index();
4683 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4684 "bad index %d", vtable_index);
4685 // Get the Method* out of the appropriate vtable entry.
4686 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4687 vtable_index*vtableEntry::size_in_bytes() +
4688 in_bytes(vtableEntry::method_offset());
4689 Node* entry_addr = basic_plus_adr(top(), obj_klass, entry_offset);
4690 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4691
4692 // Compare the target method with the expected method (e.g., Object.hashCode).
4693 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4694
4695 Node* native_call = makecon(native_call_addr);
4696 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4697 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4698
4699 return generate_slow_guard(test_native, slow_region);
4700 }
4701
4702 //-----------------------generate_method_call----------------------------
4703 // Use generate_method_call to make a slow-call to the real
4704 // method if the fast path fails. An alternative would be to
4705 // use a stub like OptoRuntime::slow_arraycopy_Java.
4706 // This only works for expanding the current library call,
4707 // not another intrinsic. (E.g., don't use this for making an
4708 // arraycopy call inside of the copyOf intrinsic.)
4709 CallJavaNode*
4710 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4711 // When compiling the intrinsic method itself, do not use this technique.
4712 guarantee(callee() != C->method(), "cannot make slow-call to self");
4713
4714 ciMethod* method = callee();
4715 // ensure the JVMS we have will be correct for this call
4716 guarantee(method_id == method->intrinsic_id(), "must match");
4717
4718 const TypeFunc* tf = TypeFunc::make(method);
4719 if (res_not_null) {
4720 assert(tf->return_type() == T_OBJECT, "");
4721 const TypeTuple* range = tf->range();
4722 const Type** fields = TypeTuple::fields(range->cnt());
4723 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4724 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4725 tf = TypeFunc::make(tf->domain(), new_range);
4726 }
4727 CallJavaNode* slow_call;
4728 if (is_static) {
4729 assert(!is_virtual, "");
4730 slow_call = new CallStaticJavaNode(C, tf,
4731 SharedRuntime::get_resolve_static_call_stub(), method);
4732 } else if (is_virtual) {
4733 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4734 int vtable_index = Method::invalid_vtable_index;
4735 if (UseInlineCaches) {
4736 // Suppress the vtable call
4737 } else {
4738 // hashCode and clone are not a miranda methods,
4739 // so the vtable index is fixed.
4740 // No need to use the linkResolver to get it.
4741 vtable_index = method->vtable_index();
4742 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4743 "bad index %d", vtable_index);
4744 }
4745 slow_call = new CallDynamicJavaNode(tf,
4746 SharedRuntime::get_resolve_virtual_call_stub(),
4747 method, vtable_index);
4748 } else { // neither virtual nor static: opt_virtual
4749 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4750 slow_call = new CallStaticJavaNode(C, tf,
4751 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4752 slow_call->set_optimized_virtual(true);
4753 }
4754 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4755 // To be able to issue a direct call (optimized virtual or virtual)
4756 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4757 // about the method being invoked should be attached to the call site to
4758 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4759 slow_call->set_override_symbolic_info(true);
4760 }
4761 set_arguments_for_java_call(slow_call);
4762 set_edges_for_java_call(slow_call);
4763 return slow_call;
4764 }
4765
4766
4767 /**
4768 * Build special case code for calls to hashCode on an object. This call may
4769 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4770 * slightly different code.
4771 */
4772 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4773 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4774 assert(!(is_virtual && is_static), "either virtual, special, or static");
4775
4776 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4777
4778 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4779 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4780 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4781 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4782 Node* obj = nullptr;
4783 if (!is_static) {
4784 // Check for hashing null object
4785 obj = null_check_receiver();
4786 if (stopped()) return true; // unconditionally null
4787 result_reg->init_req(_null_path, top());
4788 result_val->init_req(_null_path, top());
4789 } else {
4790 // Do a null check, and return zero if null.
4791 // System.identityHashCode(null) == 0
4792 obj = argument(0);
4793 Node* null_ctl = top();
4794 obj = null_check_oop(obj, &null_ctl);
4795 result_reg->init_req(_null_path, null_ctl);
4796 result_val->init_req(_null_path, _gvn.intcon(0));
4797 }
4798
4799 // Unconditionally null? Then return right away.
4800 if (stopped()) {
4801 set_control( result_reg->in(_null_path));
4802 if (!stopped())
4803 set_result(result_val->in(_null_path));
4804 return true;
4805 }
4806
4807 // We only go to the fast case code if we pass a number of guards. The
4808 // paths which do not pass are accumulated in the slow_region.
4809 RegionNode* slow_region = new RegionNode(1);
4810 record_for_igvn(slow_region);
4811
4812 // If this is a virtual call, we generate a funny guard. We pull out
4813 // the vtable entry corresponding to hashCode() from the target object.
4814 // If the target method which we are calling happens to be the native
4815 // Object hashCode() method, we pass the guard. We do not need this
4816 // guard for non-virtual calls -- the caller is known to be the native
4817 // Object hashCode().
4818 if (is_virtual) {
4819 // After null check, get the object's klass.
4820 Node* obj_klass = load_object_klass(obj);
4821 generate_virtual_guard(obj_klass, slow_region);
4822 }
4823
4824 // Get the header out of the object, use LoadMarkNode when available
4825 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4826 // The control of the load must be null. Otherwise, the load can move before
4827 // the null check after castPP removal.
4828 Node* no_ctrl = nullptr;
4829 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4830
4831 if (!UseObjectMonitorTable) {
4832 // Test the header to see if it is safe to read w.r.t. locking.
4833 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4834 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4835 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4836 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4837 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4838
4839 generate_slow_guard(test_monitor, slow_region);
4840 }
4841
4842 // Get the hash value and check to see that it has been properly assigned.
4843 // We depend on hash_mask being at most 32 bits and avoid the use of
4844 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4845 // vm: see markWord.hpp.
4846 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4847 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4848 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4849 // This hack lets the hash bits live anywhere in the mark object now, as long
4850 // as the shift drops the relevant bits into the low 32 bits. Note that
4851 // Java spec says that HashCode is an int so there's no point in capturing
4852 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4853 hshifted_header = ConvX2I(hshifted_header);
4854 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4855
4856 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4857 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4858 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4859
4860 generate_slow_guard(test_assigned, slow_region);
4861
4862 Node* init_mem = reset_memory();
4863 // fill in the rest of the null path:
4864 result_io ->init_req(_null_path, i_o());
4865 result_mem->init_req(_null_path, init_mem);
4866
4867 result_val->init_req(_fast_path, hash_val);
4868 result_reg->init_req(_fast_path, control());
4869 result_io ->init_req(_fast_path, i_o());
4870 result_mem->init_req(_fast_path, init_mem);
4871
4872 // Generate code for the slow case. We make a call to hashCode().
4873 set_control(_gvn.transform(slow_region));
4874 if (!stopped()) {
4875 // No need for PreserveJVMState, because we're using up the present state.
4876 set_all_memory(init_mem);
4877 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4878 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4879 Node* slow_result = set_results_for_java_call(slow_call);
4880 // this->control() comes from set_results_for_java_call
4881 result_reg->init_req(_slow_path, control());
4882 result_val->init_req(_slow_path, slow_result);
4883 result_io ->set_req(_slow_path, i_o());
4884 result_mem ->set_req(_slow_path, reset_memory());
4885 }
4886
4887 // Return the combined state.
4888 set_i_o( _gvn.transform(result_io) );
4889 set_all_memory( _gvn.transform(result_mem));
4890
4891 set_result(result_reg, result_val);
4892 return true;
4893 }
4894
4895 //---------------------------inline_native_getClass----------------------------
4896 // public final native Class<?> java.lang.Object.getClass();
4897 //
4898 // Build special case code for calls to getClass on an object.
4899 bool LibraryCallKit::inline_native_getClass() {
4900 Node* obj = null_check_receiver();
4901 if (stopped()) return true;
4902 set_result(load_mirror_from_klass(load_object_klass(obj)));
4903 return true;
4904 }
4905
4906 //-----------------inline_native_Reflection_getCallerClass---------------------
4907 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4908 //
4909 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4910 //
4911 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4912 // in that it must skip particular security frames and checks for
4913 // caller sensitive methods.
4914 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4915 #ifndef PRODUCT
4916 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4917 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4918 }
4919 #endif
4920
4921 if (!jvms()->has_method()) {
4922 #ifndef PRODUCT
4923 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4924 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4925 }
4926 #endif
4927 return false;
4928 }
4929
4930 // Walk back up the JVM state to find the caller at the required
4931 // depth.
4932 JVMState* caller_jvms = jvms();
4933
4934 // Cf. JVM_GetCallerClass
4935 // NOTE: Start the loop at depth 1 because the current JVM state does
4936 // not include the Reflection.getCallerClass() frame.
4937 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4938 ciMethod* m = caller_jvms->method();
4939 switch (n) {
4940 case 0:
4941 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4942 break;
4943 case 1:
4944 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4945 if (!m->caller_sensitive()) {
4946 #ifndef PRODUCT
4947 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4948 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4949 }
4950 #endif
4951 return false; // bail-out; let JVM_GetCallerClass do the work
4952 }
4953 break;
4954 default:
4955 if (!m->is_ignored_by_security_stack_walk()) {
4956 // We have reached the desired frame; return the holder class.
4957 // Acquire method holder as java.lang.Class and push as constant.
4958 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4959 ciInstance* caller_mirror = caller_klass->java_mirror();
4960 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4961
4962 #ifndef PRODUCT
4963 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4964 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());
4965 tty->print_cr(" JVM state at this point:");
4966 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4967 ciMethod* m = jvms()->of_depth(i)->method();
4968 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4969 }
4970 }
4971 #endif
4972 return true;
4973 }
4974 break;
4975 }
4976 }
4977
4978 #ifndef PRODUCT
4979 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4980 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4981 tty->print_cr(" JVM state at this point:");
4982 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4983 ciMethod* m = jvms()->of_depth(i)->method();
4984 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4985 }
4986 }
4987 #endif
4988
4989 return false; // bail-out; let JVM_GetCallerClass do the work
4990 }
4991
4992 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4993 Node* arg = argument(0);
4994 Node* result = nullptr;
4995
4996 switch (id) {
4997 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4998 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4999 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5000 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5001 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5002 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5003
5004 case vmIntrinsics::_doubleToLongBits: {
5005 // two paths (plus control) merge in a wood
5006 RegionNode *r = new RegionNode(3);
5007 Node *phi = new PhiNode(r, TypeLong::LONG);
5008
5009 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5010 // Build the boolean node
5011 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5012
5013 // Branch either way.
5014 // NaN case is less traveled, which makes all the difference.
5015 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5016 Node *opt_isnan = _gvn.transform(ifisnan);
5017 assert( opt_isnan->is_If(), "Expect an IfNode");
5018 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5019 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5020
5021 set_control(iftrue);
5022
5023 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5024 Node *slow_result = longcon(nan_bits); // return NaN
5025 phi->init_req(1, _gvn.transform( slow_result ));
5026 r->init_req(1, iftrue);
5027
5028 // Else fall through
5029 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5030 set_control(iffalse);
5031
5032 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5033 r->init_req(2, iffalse);
5034
5035 // Post merge
5036 set_control(_gvn.transform(r));
5037 record_for_igvn(r);
5038
5039 C->set_has_split_ifs(true); // Has chance for split-if optimization
5040 result = phi;
5041 assert(result->bottom_type()->isa_long(), "must be");
5042 break;
5043 }
5044
5045 case vmIntrinsics::_floatToIntBits: {
5046 // two paths (plus control) merge in a wood
5047 RegionNode *r = new RegionNode(3);
5048 Node *phi = new PhiNode(r, TypeInt::INT);
5049
5050 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5051 // Build the boolean node
5052 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5053
5054 // Branch either way.
5055 // NaN case is less traveled, which makes all the difference.
5056 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5057 Node *opt_isnan = _gvn.transform(ifisnan);
5058 assert( opt_isnan->is_If(), "Expect an IfNode");
5059 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5060 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5061
5062 set_control(iftrue);
5063
5064 static const jint nan_bits = 0x7fc00000;
5065 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5066 phi->init_req(1, _gvn.transform( slow_result ));
5067 r->init_req(1, iftrue);
5068
5069 // Else fall through
5070 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5071 set_control(iffalse);
5072
5073 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5074 r->init_req(2, iffalse);
5075
5076 // Post merge
5077 set_control(_gvn.transform(r));
5078 record_for_igvn(r);
5079
5080 C->set_has_split_ifs(true); // Has chance for split-if optimization
5081 result = phi;
5082 assert(result->bottom_type()->isa_int(), "must be");
5083 break;
5084 }
5085
5086 default:
5087 fatal_unexpected_iid(id);
5088 break;
5089 }
5090 set_result(_gvn.transform(result));
5091 return true;
5092 }
5093
5094 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5095 Node* arg = argument(0);
5096 Node* result = nullptr;
5097
5098 switch (id) {
5099 case vmIntrinsics::_floatIsInfinite:
5100 result = new IsInfiniteFNode(arg);
5101 break;
5102 case vmIntrinsics::_floatIsFinite:
5103 result = new IsFiniteFNode(arg);
5104 break;
5105 case vmIntrinsics::_doubleIsInfinite:
5106 result = new IsInfiniteDNode(arg);
5107 break;
5108 case vmIntrinsics::_doubleIsFinite:
5109 result = new IsFiniteDNode(arg);
5110 break;
5111 default:
5112 fatal_unexpected_iid(id);
5113 break;
5114 }
5115 set_result(_gvn.transform(result));
5116 return true;
5117 }
5118
5119 //----------------------inline_unsafe_copyMemory-------------------------
5120 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5121
5122 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5123 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5124 const Type* base_t = gvn.type(base);
5125
5126 bool in_native = (base_t == TypePtr::NULL_PTR);
5127 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5128 bool is_mixed = !in_heap && !in_native;
5129
5130 if (is_mixed) {
5131 return true; // mixed accesses can touch both on-heap and off-heap memory
5132 }
5133 if (in_heap) {
5134 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5135 if (!is_prim_array) {
5136 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5137 // there's not enough type information available to determine proper memory slice for it.
5138 return true;
5139 }
5140 }
5141 return false;
5142 }
5143
5144 bool LibraryCallKit::inline_unsafe_copyMemory() {
5145 if (callee()->is_static()) return false; // caller must have the capability!
5146 null_check_receiver(); // null-check receiver
5147 if (stopped()) return true;
5148
5149 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5150
5151 Node* src_base = argument(1); // type: oop
5152 Node* src_off = ConvL2X(argument(2)); // type: long
5153 Node* dst_base = argument(4); // type: oop
5154 Node* dst_off = ConvL2X(argument(5)); // type: long
5155 Node* size = ConvL2X(argument(7)); // type: long
5156
5157 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5158 "fieldOffset must be byte-scaled");
5159
5160 Node* src_addr = make_unsafe_address(src_base, src_off);
5161 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5162
5163 Node* thread = _gvn.transform(new ThreadLocalNode());
5164 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5165 BasicType doing_unsafe_access_bt = T_BYTE;
5166 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5167
5168 // update volatile field
5169 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5170
5171 int flags = RC_LEAF | RC_NO_FP;
5172
5173 const TypePtr* dst_type = TypePtr::BOTTOM;
5174
5175 // Adjust memory effects of the runtime call based on input values.
5176 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5177 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5178 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5179
5180 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5181 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5182 flags |= RC_NARROW_MEM; // narrow in memory
5183 }
5184 }
5185
5186 // Call it. Note that the length argument is not scaled.
5187 make_runtime_call(flags,
5188 OptoRuntime::fast_arraycopy_Type(),
5189 StubRoutines::unsafe_arraycopy(),
5190 "unsafe_arraycopy",
5191 dst_type,
5192 src_addr, dst_addr, size XTOP);
5193
5194 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5195
5196 return true;
5197 }
5198
5199 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5200 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5201 bool LibraryCallKit::inline_unsafe_setMemory() {
5202 if (callee()->is_static()) return false; // caller must have the capability!
5203 null_check_receiver(); // null-check receiver
5204 if (stopped()) return true;
5205
5206 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5207
5208 Node* dst_base = argument(1); // type: oop
5209 Node* dst_off = ConvL2X(argument(2)); // type: long
5210 Node* size = ConvL2X(argument(4)); // type: long
5211 Node* byte = argument(6); // type: byte
5212
5213 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5214 "fieldOffset must be byte-scaled");
5215
5216 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5217
5218 Node* thread = _gvn.transform(new ThreadLocalNode());
5219 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5220 BasicType doing_unsafe_access_bt = T_BYTE;
5221 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5222
5223 // update volatile field
5224 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5225
5226 int flags = RC_LEAF | RC_NO_FP;
5227
5228 const TypePtr* dst_type = TypePtr::BOTTOM;
5229
5230 // Adjust memory effects of the runtime call based on input values.
5231 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5232 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5233
5234 flags |= RC_NARROW_MEM; // narrow in memory
5235 }
5236
5237 // Call it. Note that the length argument is not scaled.
5238 make_runtime_call(flags,
5239 OptoRuntime::unsafe_setmemory_Type(),
5240 StubRoutines::unsafe_setmemory(),
5241 "unsafe_setmemory",
5242 dst_type,
5243 dst_addr, size XTOP, byte);
5244
5245 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5246
5247 return true;
5248 }
5249
5250 #undef XTOP
5251
5252 //------------------------clone_coping-----------------------------------
5253 // Helper function for inline_native_clone.
5254 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5255 assert(obj_size != nullptr, "");
5256 Node* raw_obj = alloc_obj->in(1);
5257 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5258
5259 AllocateNode* alloc = nullptr;
5260 if (ReduceBulkZeroing &&
5261 // If we are implementing an array clone without knowing its source type
5262 // (can happen when compiling the array-guarded branch of a reflective
5263 // Object.clone() invocation), initialize the array within the allocation.
5264 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5265 // to a runtime clone call that assumes fully initialized source arrays.
5266 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5267 // We will be completely responsible for initializing this object -
5268 // mark Initialize node as complete.
5269 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5270 // The object was just allocated - there should be no any stores!
5271 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5272 // Mark as complete_with_arraycopy so that on AllocateNode
5273 // expansion, we know this AllocateNode is initialized by an array
5274 // copy and a StoreStore barrier exists after the array copy.
5275 alloc->initialization()->set_complete_with_arraycopy();
5276 }
5277
5278 Node* size = _gvn.transform(obj_size);
5279 access_clone(obj, alloc_obj, size, is_array);
5280
5281 // Do not let reads from the cloned object float above the arraycopy.
5282 if (alloc != nullptr) {
5283 // Do not let stores that initialize this object be reordered with
5284 // a subsequent store that would make this object accessible by
5285 // other threads.
5286 // Record what AllocateNode this StoreStore protects so that
5287 // escape analysis can go from the MemBarStoreStoreNode to the
5288 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5289 // based on the escape status of the AllocateNode.
5290 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5291 } else {
5292 insert_mem_bar(Op_MemBarCPUOrder);
5293 }
5294 }
5295
5296 //------------------------inline_native_clone----------------------------
5297 // protected native Object java.lang.Object.clone();
5298 //
5299 // Here are the simple edge cases:
5300 // null receiver => normal trap
5301 // virtual and clone was overridden => slow path to out-of-line clone
5302 // not cloneable or finalizer => slow path to out-of-line Object.clone
5303 //
5304 // The general case has two steps, allocation and copying.
5305 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5306 //
5307 // Copying also has two cases, oop arrays and everything else.
5308 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5309 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5310 //
5311 // These steps fold up nicely if and when the cloned object's klass
5312 // can be sharply typed as an object array, a type array, or an instance.
5313 //
5314 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5315 PhiNode* result_val;
5316
5317 // Set the reexecute bit for the interpreter to reexecute
5318 // the bytecode that invokes Object.clone if deoptimization happens.
5319 { PreserveReexecuteState preexecs(this);
5320 jvms()->set_should_reexecute(true);
5321
5322 Node* obj = null_check_receiver();
5323 if (stopped()) return true;
5324
5325 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5326
5327 // If we are going to clone an instance, we need its exact type to
5328 // know the number and types of fields to convert the clone to
5329 // loads/stores. Maybe a speculative type can help us.
5330 if (!obj_type->klass_is_exact() &&
5331 obj_type->speculative_type() != nullptr &&
5332 obj_type->speculative_type()->is_instance_klass()) {
5333 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5334 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5335 !spec_ik->has_injected_fields()) {
5336 if (!obj_type->isa_instptr() ||
5337 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5338 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5339 }
5340 }
5341 }
5342
5343 // Conservatively insert a memory barrier on all memory slices.
5344 // Do not let writes into the original float below the clone.
5345 insert_mem_bar(Op_MemBarCPUOrder);
5346
5347 // paths into result_reg:
5348 enum {
5349 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5350 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5351 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5352 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5353 PATH_LIMIT
5354 };
5355 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5356 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5357 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5358 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5359 record_for_igvn(result_reg);
5360
5361 Node* obj_klass = load_object_klass(obj);
5362 Node* array_obj = obj;
5363 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5364 if (array_ctl != nullptr) {
5365 // It's an array.
5366 PreserveJVMState pjvms(this);
5367 set_control(array_ctl);
5368 Node* obj_length = load_array_length(array_obj);
5369 Node* array_size = nullptr; // Size of the array without object alignment padding.
5370 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5371
5372 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5373 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5374 // If it is an oop array, it requires very special treatment,
5375 // because gc barriers are required when accessing the array.
5376 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5377 if (is_obja != nullptr) {
5378 PreserveJVMState pjvms2(this);
5379 set_control(is_obja);
5380 // Generate a direct call to the right arraycopy function(s).
5381 // Clones are always tightly coupled.
5382 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5383 ac->set_clone_oop_array();
5384 Node* n = _gvn.transform(ac);
5385 assert(n == ac, "cannot disappear");
5386 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5387
5388 result_reg->init_req(_objArray_path, control());
5389 result_val->init_req(_objArray_path, alloc_obj);
5390 result_i_o ->set_req(_objArray_path, i_o());
5391 result_mem ->set_req(_objArray_path, reset_memory());
5392 }
5393 }
5394 // Otherwise, there are no barriers to worry about.
5395 // (We can dispense with card marks if we know the allocation
5396 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5397 // causes the non-eden paths to take compensating steps to
5398 // simulate a fresh allocation, so that no further
5399 // card marks are required in compiled code to initialize
5400 // the object.)
5401
5402 if (!stopped()) {
5403 copy_to_clone(array_obj, alloc_obj, array_size, true);
5404
5405 // Present the results of the copy.
5406 result_reg->init_req(_array_path, control());
5407 result_val->init_req(_array_path, alloc_obj);
5408 result_i_o ->set_req(_array_path, i_o());
5409 result_mem ->set_req(_array_path, reset_memory());
5410 }
5411 }
5412
5413 // We only go to the instance fast case code if we pass a number of guards.
5414 // The paths which do not pass are accumulated in the slow_region.
5415 RegionNode* slow_region = new RegionNode(1);
5416 record_for_igvn(slow_region);
5417 if (!stopped()) {
5418 // It's an instance (we did array above). Make the slow-path tests.
5419 // If this is a virtual call, we generate a funny guard. We grab
5420 // the vtable entry corresponding to clone() from the target object.
5421 // If the target method which we are calling happens to be the
5422 // Object clone() method, we pass the guard. We do not need this
5423 // guard for non-virtual calls; the caller is known to be the native
5424 // Object clone().
5425 if (is_virtual) {
5426 generate_virtual_guard(obj_klass, slow_region);
5427 }
5428
5429 // The object must be easily cloneable and must not have a finalizer.
5430 // Both of these conditions may be checked in a single test.
5431 // We could optimize the test further, but we don't care.
5432 generate_misc_flags_guard(obj_klass,
5433 // Test both conditions:
5434 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5435 // Must be cloneable but not finalizer:
5436 KlassFlags::_misc_is_cloneable_fast,
5437 slow_region);
5438 }
5439
5440 if (!stopped()) {
5441 // It's an instance, and it passed the slow-path tests.
5442 PreserveJVMState pjvms(this);
5443 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5444 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5445 // is reexecuted if deoptimization occurs and there could be problems when merging
5446 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5447 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5448
5449 copy_to_clone(obj, alloc_obj, obj_size, false);
5450
5451 // Present the results of the slow call.
5452 result_reg->init_req(_instance_path, control());
5453 result_val->init_req(_instance_path, alloc_obj);
5454 result_i_o ->set_req(_instance_path, i_o());
5455 result_mem ->set_req(_instance_path, reset_memory());
5456 }
5457
5458 // Generate code for the slow case. We make a call to clone().
5459 set_control(_gvn.transform(slow_region));
5460 if (!stopped()) {
5461 PreserveJVMState pjvms(this);
5462 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5463 // We need to deoptimize on exception (see comment above)
5464 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5465 // this->control() comes from set_results_for_java_call
5466 result_reg->init_req(_slow_path, control());
5467 result_val->init_req(_slow_path, slow_result);
5468 result_i_o ->set_req(_slow_path, i_o());
5469 result_mem ->set_req(_slow_path, reset_memory());
5470 }
5471
5472 // Return the combined state.
5473 set_control( _gvn.transform(result_reg));
5474 set_i_o( _gvn.transform(result_i_o));
5475 set_all_memory( _gvn.transform(result_mem));
5476 } // original reexecute is set back here
5477
5478 set_result(_gvn.transform(result_val));
5479 return true;
5480 }
5481
5482 // If we have a tightly coupled allocation, the arraycopy may take care
5483 // of the array initialization. If one of the guards we insert between
5484 // the allocation and the arraycopy causes a deoptimization, an
5485 // uninitialized array will escape the compiled method. To prevent that
5486 // we set the JVM state for uncommon traps between the allocation and
5487 // the arraycopy to the state before the allocation so, in case of
5488 // deoptimization, we'll reexecute the allocation and the
5489 // initialization.
5490 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5491 if (alloc != nullptr) {
5492 ciMethod* trap_method = alloc->jvms()->method();
5493 int trap_bci = alloc->jvms()->bci();
5494
5495 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5496 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5497 // Make sure there's no store between the allocation and the
5498 // arraycopy otherwise visible side effects could be rexecuted
5499 // in case of deoptimization and cause incorrect execution.
5500 bool no_interfering_store = true;
5501 Node* mem = alloc->in(TypeFunc::Memory);
5502 if (mem->is_MergeMem()) {
5503 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5504 Node* n = mms.memory();
5505 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5506 assert(n->is_Store(), "what else?");
5507 no_interfering_store = false;
5508 break;
5509 }
5510 }
5511 } else {
5512 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5513 Node* n = mms.memory();
5514 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5515 assert(n->is_Store(), "what else?");
5516 no_interfering_store = false;
5517 break;
5518 }
5519 }
5520 }
5521
5522 if (no_interfering_store) {
5523 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5524
5525 JVMState* saved_jvms = jvms();
5526 saved_reexecute_sp = _reexecute_sp;
5527
5528 set_jvms(sfpt->jvms());
5529 _reexecute_sp = jvms()->sp();
5530
5531 return saved_jvms;
5532 }
5533 }
5534 }
5535 return nullptr;
5536 }
5537
5538 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5539 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5540 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5541 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5542 uint size = alloc->req();
5543 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5544 old_jvms->set_map(sfpt);
5545 for (uint i = 0; i < size; i++) {
5546 sfpt->init_req(i, alloc->in(i));
5547 }
5548 // re-push array length for deoptimization
5549 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5550 old_jvms->set_sp(old_jvms->sp()+1);
5551 old_jvms->set_monoff(old_jvms->monoff()+1);
5552 old_jvms->set_scloff(old_jvms->scloff()+1);
5553 old_jvms->set_endoff(old_jvms->endoff()+1);
5554 old_jvms->set_should_reexecute(true);
5555
5556 sfpt->set_i_o(map()->i_o());
5557 sfpt->set_memory(map()->memory());
5558 sfpt->set_control(map()->control());
5559 return sfpt;
5560 }
5561
5562 // In case of a deoptimization, we restart execution at the
5563 // allocation, allocating a new array. We would leave an uninitialized
5564 // array in the heap that GCs wouldn't expect. Move the allocation
5565 // after the traps so we don't allocate the array if we
5566 // deoptimize. This is possible because tightly_coupled_allocation()
5567 // guarantees there's no observer of the allocated array at this point
5568 // and the control flow is simple enough.
5569 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5570 int saved_reexecute_sp, uint new_idx) {
5571 if (saved_jvms_before_guards != nullptr && !stopped()) {
5572 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5573
5574 assert(alloc != nullptr, "only with a tightly coupled allocation");
5575 // restore JVM state to the state at the arraycopy
5576 saved_jvms_before_guards->map()->set_control(map()->control());
5577 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5578 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5579 // If we've improved the types of some nodes (null check) while
5580 // emitting the guards, propagate them to the current state
5581 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5582 set_jvms(saved_jvms_before_guards);
5583 _reexecute_sp = saved_reexecute_sp;
5584
5585 // Remove the allocation from above the guards
5586 CallProjections callprojs;
5587 alloc->extract_projections(&callprojs, true);
5588 InitializeNode* init = alloc->initialization();
5589 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5590 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5591 init->replace_mem_projs_by(alloc_mem, C);
5592
5593 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5594 // the allocation (i.e. is only valid if the allocation succeeds):
5595 // 1) replace CastIINode with AllocateArrayNode's length here
5596 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5597 //
5598 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5599 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5600 Node* init_control = init->proj_out(TypeFunc::Control);
5601 Node* alloc_length = alloc->Ideal_length();
5602 #ifdef ASSERT
5603 Node* prev_cast = nullptr;
5604 #endif
5605 for (uint i = 0; i < init_control->outcnt(); i++) {
5606 Node* init_out = init_control->raw_out(i);
5607 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5608 #ifdef ASSERT
5609 if (prev_cast == nullptr) {
5610 prev_cast = init_out;
5611 } else {
5612 if (prev_cast->cmp(*init_out) == false) {
5613 prev_cast->dump();
5614 init_out->dump();
5615 assert(false, "not equal CastIINode");
5616 }
5617 }
5618 #endif
5619 C->gvn_replace_by(init_out, alloc_length);
5620 }
5621 }
5622 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5623
5624 // move the allocation here (after the guards)
5625 _gvn.hash_delete(alloc);
5626 alloc->set_req(TypeFunc::Control, control());
5627 alloc->set_req(TypeFunc::I_O, i_o());
5628 Node *mem = reset_memory();
5629 set_all_memory(mem);
5630 alloc->set_req(TypeFunc::Memory, mem);
5631 set_control(init->proj_out_or_null(TypeFunc::Control));
5632 set_i_o(callprojs.fallthrough_ioproj);
5633
5634 // Update memory as done in GraphKit::set_output_for_allocation()
5635 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5636 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5637 if (ary_type->isa_aryptr() && length_type != nullptr) {
5638 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5639 }
5640 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5641 int elemidx = C->get_alias_index(telemref);
5642 // Need to properly move every memory projection for the Initialize
5643 #ifdef ASSERT
5644 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5645 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
5646 #endif
5647 auto move_proj = [&](ProjNode* proj) {
5648 int alias_idx = C->get_alias_index(proj->adr_type());
5649 assert(alias_idx == Compile::AliasIdxRaw ||
5650 alias_idx == elemidx ||
5651 alias_idx == mark_idx ||
5652 alias_idx == klass_idx, "should be raw memory or array element type");
5653 set_memory(proj, alias_idx);
5654 };
5655 init->for_each_proj(move_proj, TypeFunc::Memory);
5656
5657 Node* allocx = _gvn.transform(alloc);
5658 assert(allocx == alloc, "where has the allocation gone?");
5659 assert(dest->is_CheckCastPP(), "not an allocation result?");
5660
5661 _gvn.hash_delete(dest);
5662 dest->set_req(0, control());
5663 Node* destx = _gvn.transform(dest);
5664 assert(destx == dest, "where has the allocation result gone?");
5665
5666 array_ideal_length(alloc, ary_type, true);
5667 }
5668 }
5669
5670 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5671 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5672 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5673 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5674 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5675 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5676 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5677 JVMState* saved_jvms_before_guards) {
5678 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5679 // There is at least one unrelated uncommon trap which needs to be replaced.
5680 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5681
5682 JVMState* saved_jvms = jvms();
5683 const int saved_reexecute_sp = _reexecute_sp;
5684 set_jvms(sfpt->jvms());
5685 _reexecute_sp = jvms()->sp();
5686
5687 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5688
5689 // Restore state
5690 set_jvms(saved_jvms);
5691 _reexecute_sp = saved_reexecute_sp;
5692 }
5693 }
5694
5695 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5696 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5697 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5698 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5699 while (if_proj->is_IfProj()) {
5700 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5701 if (uncommon_trap != nullptr) {
5702 create_new_uncommon_trap(uncommon_trap);
5703 }
5704 assert(if_proj->in(0)->is_If(), "must be If");
5705 if_proj = if_proj->in(0)->in(0);
5706 }
5707 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5708 "must have reached control projection of init node");
5709 }
5710
5711 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5712 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5713 assert(trap_request != 0, "no valid UCT trap request");
5714 PreserveJVMState pjvms(this);
5715 set_control(uncommon_trap_call->in(0));
5716 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5717 Deoptimization::trap_request_action(trap_request));
5718 assert(stopped(), "Should be stopped");
5719 _gvn.hash_delete(uncommon_trap_call);
5720 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5721 }
5722
5723 // Common checks for array sorting intrinsics arguments.
5724 // Returns `true` if checks passed.
5725 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5726 // check address of the class
5727 if (elementType == nullptr || elementType->is_top()) {
5728 return false; // dead path
5729 }
5730 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5731 if (elem_klass == nullptr) {
5732 return false; // dead path
5733 }
5734 // java_mirror_type() returns non-null for compile-time Class constants only
5735 ciType* elem_type = elem_klass->java_mirror_type();
5736 if (elem_type == nullptr) {
5737 return false;
5738 }
5739 bt = elem_type->basic_type();
5740 // Disable the intrinsic if the CPU does not support SIMD sort
5741 if (!Matcher::supports_simd_sort(bt)) {
5742 return false;
5743 }
5744 // check address of the array
5745 if (obj == nullptr || obj->is_top()) {
5746 return false; // dead path
5747 }
5748 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5749 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5750 return false; // failed input validation
5751 }
5752 return true;
5753 }
5754
5755 //------------------------------inline_array_partition-----------------------
5756 bool LibraryCallKit::inline_array_partition() {
5757 address stubAddr = StubRoutines::select_array_partition_function();
5758 if (stubAddr == nullptr) {
5759 return false; // Intrinsic's stub is not implemented on this platform
5760 }
5761 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5762
5763 // no receiver because it is a static method
5764 Node* elementType = argument(0);
5765 Node* obj = argument(1);
5766 Node* offset = argument(2); // long
5767 Node* fromIndex = argument(4);
5768 Node* toIndex = argument(5);
5769 Node* indexPivot1 = argument(6);
5770 Node* indexPivot2 = argument(7);
5771 // PartitionOperation: argument(8) is ignored
5772
5773 Node* pivotIndices = nullptr;
5774 BasicType bt = T_ILLEGAL;
5775
5776 if (!check_array_sort_arguments(elementType, obj, bt)) {
5777 return false;
5778 }
5779 null_check(obj);
5780 // If obj is dead, only null-path is taken.
5781 if (stopped()) {
5782 return true;
5783 }
5784 // Set the original stack and the reexecute bit for the interpreter to reexecute
5785 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5786 { PreserveReexecuteState preexecs(this);
5787 jvms()->set_should_reexecute(true);
5788
5789 Node* obj_adr = make_unsafe_address(obj, offset);
5790
5791 // create the pivotIndices array of type int and size = 2
5792 Node* size = intcon(2);
5793 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5794 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5795 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5796 guarantee(alloc != nullptr, "created above");
5797 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5798
5799 // pass the basic type enum to the stub
5800 Node* elemType = intcon(bt);
5801
5802 // Call the stub
5803 const char *stubName = "array_partition_stub";
5804 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5805 stubAddr, stubName, TypePtr::BOTTOM,
5806 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5807 indexPivot1, indexPivot2);
5808
5809 } // original reexecute is set back here
5810
5811 if (!stopped()) {
5812 set_result(pivotIndices);
5813 }
5814
5815 return true;
5816 }
5817
5818
5819 //------------------------------inline_array_sort-----------------------
5820 bool LibraryCallKit::inline_array_sort() {
5821 address stubAddr = StubRoutines::select_arraysort_function();
5822 if (stubAddr == nullptr) {
5823 return false; // Intrinsic's stub is not implemented on this platform
5824 }
5825 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5826
5827 // no receiver because it is a static method
5828 Node* elementType = argument(0);
5829 Node* obj = argument(1);
5830 Node* offset = argument(2); // long
5831 Node* fromIndex = argument(4);
5832 Node* toIndex = argument(5);
5833 // SortOperation: argument(6) is ignored
5834
5835 BasicType bt = T_ILLEGAL;
5836
5837 if (!check_array_sort_arguments(elementType, obj, bt)) {
5838 return false;
5839 }
5840 null_check(obj);
5841 // If obj is dead, only null-path is taken.
5842 if (stopped()) {
5843 return true;
5844 }
5845 Node* obj_adr = make_unsafe_address(obj, offset);
5846
5847 // pass the basic type enum to the stub
5848 Node* elemType = intcon(bt);
5849
5850 // Call the stub.
5851 const char *stubName = "arraysort_stub";
5852 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5853 stubAddr, stubName, TypePtr::BOTTOM,
5854 obj_adr, elemType, fromIndex, toIndex);
5855
5856 return true;
5857 }
5858
5859
5860 //------------------------------inline_arraycopy-----------------------
5861 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5862 // Object dest, int destPos,
5863 // int length);
5864 bool LibraryCallKit::inline_arraycopy() {
5865 // Get the arguments.
5866 Node* src = argument(0); // type: oop
5867 Node* src_offset = argument(1); // type: int
5868 Node* dest = argument(2); // type: oop
5869 Node* dest_offset = argument(3); // type: int
5870 Node* length = argument(4); // type: int
5871
5872 uint new_idx = C->unique();
5873
5874 // Check for allocation before we add nodes that would confuse
5875 // tightly_coupled_allocation()
5876 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5877
5878 int saved_reexecute_sp = -1;
5879 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5880 // See arraycopy_restore_alloc_state() comment
5881 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5882 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5883 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5884 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5885
5886 // The following tests must be performed
5887 // (1) src and dest are arrays.
5888 // (2) src and dest arrays must have elements of the same BasicType
5889 // (3) src and dest must not be null.
5890 // (4) src_offset must not be negative.
5891 // (5) dest_offset must not be negative.
5892 // (6) length must not be negative.
5893 // (7) src_offset + length must not exceed length of src.
5894 // (8) dest_offset + length must not exceed length of dest.
5895 // (9) each element of an oop array must be assignable
5896
5897 // (3) src and dest must not be null.
5898 // always do this here because we need the JVM state for uncommon traps
5899 Node* null_ctl = top();
5900 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5901 assert(null_ctl->is_top(), "no null control here");
5902 dest = null_check(dest, T_ARRAY);
5903
5904 if (!can_emit_guards) {
5905 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5906 // guards but the arraycopy node could still take advantage of a
5907 // tightly allocated allocation. tightly_coupled_allocation() is
5908 // called again to make sure it takes the null check above into
5909 // account: the null check is mandatory and if it caused an
5910 // uncommon trap to be emitted then the allocation can't be
5911 // considered tightly coupled in this context.
5912 alloc = tightly_coupled_allocation(dest);
5913 }
5914
5915 bool validated = false;
5916
5917 const Type* src_type = _gvn.type(src);
5918 const Type* dest_type = _gvn.type(dest);
5919 const TypeAryPtr* top_src = src_type->isa_aryptr();
5920 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5921
5922 // Do we have the type of src?
5923 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5924 // Do we have the type of dest?
5925 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5926 // Is the type for src from speculation?
5927 bool src_spec = false;
5928 // Is the type for dest from speculation?
5929 bool dest_spec = false;
5930
5931 if ((!has_src || !has_dest) && can_emit_guards) {
5932 // We don't have sufficient type information, let's see if
5933 // speculative types can help. We need to have types for both src
5934 // and dest so that it pays off.
5935
5936 // Do we already have or could we have type information for src
5937 bool could_have_src = has_src;
5938 // Do we already have or could we have type information for dest
5939 bool could_have_dest = has_dest;
5940
5941 ciKlass* src_k = nullptr;
5942 if (!has_src) {
5943 src_k = src_type->speculative_type_not_null();
5944 if (src_k != nullptr && src_k->is_array_klass()) {
5945 could_have_src = true;
5946 }
5947 }
5948
5949 ciKlass* dest_k = nullptr;
5950 if (!has_dest) {
5951 dest_k = dest_type->speculative_type_not_null();
5952 if (dest_k != nullptr && dest_k->is_array_klass()) {
5953 could_have_dest = true;
5954 }
5955 }
5956
5957 if (could_have_src && could_have_dest) {
5958 // This is going to pay off so emit the required guards
5959 if (!has_src) {
5960 src = maybe_cast_profiled_obj(src, src_k, true);
5961 src_type = _gvn.type(src);
5962 top_src = src_type->isa_aryptr();
5963 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5964 src_spec = true;
5965 }
5966 if (!has_dest) {
5967 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5968 dest_type = _gvn.type(dest);
5969 top_dest = dest_type->isa_aryptr();
5970 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5971 dest_spec = true;
5972 }
5973 }
5974 }
5975
5976 if (has_src && has_dest && can_emit_guards) {
5977 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5978 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5979 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5980 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5981
5982 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5983 // If both arrays are object arrays then having the exact types
5984 // for both will remove the need for a subtype check at runtime
5985 // before the call and may make it possible to pick a faster copy
5986 // routine (without a subtype check on every element)
5987 // Do we have the exact type of src?
5988 bool could_have_src = src_spec;
5989 // Do we have the exact type of dest?
5990 bool could_have_dest = dest_spec;
5991 ciKlass* src_k = nullptr;
5992 ciKlass* dest_k = nullptr;
5993 if (!src_spec) {
5994 src_k = src_type->speculative_type_not_null();
5995 if (src_k != nullptr && src_k->is_array_klass()) {
5996 could_have_src = true;
5997 }
5998 }
5999 if (!dest_spec) {
6000 dest_k = dest_type->speculative_type_not_null();
6001 if (dest_k != nullptr && dest_k->is_array_klass()) {
6002 could_have_dest = true;
6003 }
6004 }
6005 if (could_have_src && could_have_dest) {
6006 // If we can have both exact types, emit the missing guards
6007 if (could_have_src && !src_spec) {
6008 src = maybe_cast_profiled_obj(src, src_k, true);
6009 }
6010 if (could_have_dest && !dest_spec) {
6011 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6012 }
6013 }
6014 }
6015 }
6016
6017 ciMethod* trap_method = method();
6018 int trap_bci = bci();
6019 if (saved_jvms_before_guards != nullptr) {
6020 trap_method = alloc->jvms()->method();
6021 trap_bci = alloc->jvms()->bci();
6022 }
6023
6024 bool negative_length_guard_generated = false;
6025
6026 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6027 can_emit_guards &&
6028 !src->is_top() && !dest->is_top()) {
6029 // validate arguments: enables transformation the ArrayCopyNode
6030 validated = true;
6031
6032 RegionNode* slow_region = new RegionNode(1);
6033 record_for_igvn(slow_region);
6034
6035 // (1) src and dest are arrays.
6036 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6037 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6038
6039 // (2) src and dest arrays must have elements of the same BasicType
6040 // done at macro expansion or at Ideal transformation time
6041
6042 // (4) src_offset must not be negative.
6043 generate_negative_guard(src_offset, slow_region);
6044
6045 // (5) dest_offset must not be negative.
6046 generate_negative_guard(dest_offset, slow_region);
6047
6048 // (7) src_offset + length must not exceed length of src.
6049 generate_limit_guard(src_offset, length,
6050 load_array_length(src),
6051 slow_region);
6052
6053 // (8) dest_offset + length must not exceed length of dest.
6054 generate_limit_guard(dest_offset, length,
6055 load_array_length(dest),
6056 slow_region);
6057
6058 // (6) length must not be negative.
6059 // This is also checked in generate_arraycopy() during macro expansion, but
6060 // we also have to check it here for the case where the ArrayCopyNode will
6061 // be eliminated by Escape Analysis.
6062 if (EliminateAllocations) {
6063 generate_negative_guard(length, slow_region);
6064 negative_length_guard_generated = true;
6065 }
6066
6067 // (9) each element of an oop array must be assignable
6068 Node* dest_klass = load_object_klass(dest);
6069 if (src != dest) {
6070 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6071
6072 if (not_subtype_ctrl != top()) {
6073 PreserveJVMState pjvms(this);
6074 set_control(not_subtype_ctrl);
6075 uncommon_trap(Deoptimization::Reason_intrinsic,
6076 Deoptimization::Action_make_not_entrant);
6077 assert(stopped(), "Should be stopped");
6078 }
6079 }
6080 {
6081 PreserveJVMState pjvms(this);
6082 set_control(_gvn.transform(slow_region));
6083 uncommon_trap(Deoptimization::Reason_intrinsic,
6084 Deoptimization::Action_make_not_entrant);
6085 assert(stopped(), "Should be stopped");
6086 }
6087
6088 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6089 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6090 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6091 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6092 }
6093
6094 if (stopped()) {
6095 return true;
6096 }
6097
6098 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6099 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6100 // so the compiler has a chance to eliminate them: during macro expansion,
6101 // we have to set their control (CastPP nodes are eliminated).
6102 load_object_klass(src), load_object_klass(dest),
6103 load_array_length(src), load_array_length(dest));
6104
6105 ac->set_arraycopy(validated);
6106
6107 Node* n = _gvn.transform(ac);
6108 if (n == ac) {
6109 ac->connect_outputs(this);
6110 } else {
6111 assert(validated, "shouldn't transform if all arguments not validated");
6112 set_all_memory(n);
6113 }
6114 clear_upper_avx();
6115
6116
6117 return true;
6118 }
6119
6120
6121 // Helper function which determines if an arraycopy immediately follows
6122 // an allocation, with no intervening tests or other escapes for the object.
6123 AllocateArrayNode*
6124 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6125 if (stopped()) return nullptr; // no fast path
6126 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6127
6128 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6129 if (alloc == nullptr) return nullptr;
6130
6131 Node* rawmem = memory(Compile::AliasIdxRaw);
6132 // Is the allocation's memory state untouched?
6133 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6134 // Bail out if there have been raw-memory effects since the allocation.
6135 // (Example: There might have been a call or safepoint.)
6136 return nullptr;
6137 }
6138 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6139 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6140 return nullptr;
6141 }
6142
6143 // There must be no unexpected observers of this allocation.
6144 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6145 Node* obs = ptr->fast_out(i);
6146 if (obs != this->map()) {
6147 return nullptr;
6148 }
6149 }
6150
6151 // This arraycopy must unconditionally follow the allocation of the ptr.
6152 Node* alloc_ctl = ptr->in(0);
6153 Node* ctl = control();
6154 while (ctl != alloc_ctl) {
6155 // There may be guards which feed into the slow_region.
6156 // Any other control flow means that we might not get a chance
6157 // to finish initializing the allocated object.
6158 // Various low-level checks bottom out in uncommon traps. These
6159 // are considered safe since we've already checked above that
6160 // there is no unexpected observer of this allocation.
6161 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6162 assert(ctl->in(0)->is_If(), "must be If");
6163 ctl = ctl->in(0)->in(0);
6164 } else {
6165 return nullptr;
6166 }
6167 }
6168
6169 // If we get this far, we have an allocation which immediately
6170 // precedes the arraycopy, and we can take over zeroing the new object.
6171 // The arraycopy will finish the initialization, and provide
6172 // a new control state to which we will anchor the destination pointer.
6173
6174 return alloc;
6175 }
6176
6177 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6178 if (node->is_IfProj()) {
6179 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6180 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6181 Node* obs = other_proj->fast_out(j);
6182 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6183 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6184 return obs->as_CallStaticJava();
6185 }
6186 }
6187 }
6188 return nullptr;
6189 }
6190
6191 //-------------inline_encodeISOArray-----------------------------------
6192 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6193 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6194 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6195 // encode char[] to byte[] in ISO_8859_1 or ASCII
6196 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6197 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6198 // no receiver since it is static method
6199 Node *src = argument(0);
6200 Node *src_offset = argument(1);
6201 Node *dst = argument(2);
6202 Node *dst_offset = argument(3);
6203 Node *length = argument(4);
6204
6205 // Cast source & target arrays to not-null
6206 src = must_be_not_null(src, true);
6207 dst = must_be_not_null(dst, true);
6208 if (stopped()) {
6209 return true;
6210 }
6211
6212 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6213 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6214 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6215 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6216 // failed array check
6217 return false;
6218 }
6219
6220 // Figure out the size and type of the elements we will be copying.
6221 BasicType src_elem = src_type->elem()->array_element_basic_type();
6222 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6223 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6224 return false;
6225 }
6226
6227 // Check source & target bounds
6228 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6229 generate_string_range_check(dst, dst_offset, length, false, true);
6230 if (stopped()) {
6231 return true;
6232 }
6233
6234 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6235 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6236 // 'src_start' points to src array + scaled offset
6237 // 'dst_start' points to dst array + scaled offset
6238
6239 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6240 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6241 enc = _gvn.transform(enc);
6242 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6243 set_memory(res_mem, mtype);
6244 set_result(enc);
6245 clear_upper_avx();
6246
6247 return true;
6248 }
6249
6250 //-------------inline_multiplyToLen-----------------------------------
6251 bool LibraryCallKit::inline_multiplyToLen() {
6252 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6253
6254 address stubAddr = StubRoutines::multiplyToLen();
6255 if (stubAddr == nullptr) {
6256 return false; // Intrinsic's stub is not implemented on this platform
6257 }
6258 const char* stubName = "multiplyToLen";
6259
6260 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6261
6262 // no receiver because it is a static method
6263 Node* x = argument(0);
6264 Node* xlen = argument(1);
6265 Node* y = argument(2);
6266 Node* ylen = argument(3);
6267 Node* z = argument(4);
6268
6269 x = must_be_not_null(x, true);
6270 y = must_be_not_null(y, true);
6271
6272 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6273 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6274 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6275 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6276 // failed array check
6277 return false;
6278 }
6279
6280 BasicType x_elem = x_type->elem()->array_element_basic_type();
6281 BasicType y_elem = y_type->elem()->array_element_basic_type();
6282 if (x_elem != T_INT || y_elem != T_INT) {
6283 return false;
6284 }
6285
6286 Node* x_start = array_element_address(x, intcon(0), x_elem);
6287 Node* y_start = array_element_address(y, intcon(0), y_elem);
6288 // 'x_start' points to x array + scaled xlen
6289 // 'y_start' points to y array + scaled ylen
6290
6291 Node* z_start = array_element_address(z, intcon(0), T_INT);
6292
6293 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6294 OptoRuntime::multiplyToLen_Type(),
6295 stubAddr, stubName, TypePtr::BOTTOM,
6296 x_start, xlen, y_start, ylen, z_start);
6297
6298 C->set_has_split_ifs(true); // Has chance for split-if optimization
6299 set_result(z);
6300 return true;
6301 }
6302
6303 //-------------inline_squareToLen------------------------------------
6304 bool LibraryCallKit::inline_squareToLen() {
6305 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6306
6307 address stubAddr = StubRoutines::squareToLen();
6308 if (stubAddr == nullptr) {
6309 return false; // Intrinsic's stub is not implemented on this platform
6310 }
6311 const char* stubName = "squareToLen";
6312
6313 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6314
6315 Node* x = argument(0);
6316 Node* len = argument(1);
6317 Node* z = argument(2);
6318 Node* zlen = argument(3);
6319
6320 x = must_be_not_null(x, true);
6321 z = must_be_not_null(z, true);
6322
6323 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6324 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6325 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6326 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6327 // failed array check
6328 return false;
6329 }
6330
6331 BasicType x_elem = x_type->elem()->array_element_basic_type();
6332 BasicType z_elem = z_type->elem()->array_element_basic_type();
6333 if (x_elem != T_INT || z_elem != T_INT) {
6334 return false;
6335 }
6336
6337
6338 Node* x_start = array_element_address(x, intcon(0), x_elem);
6339 Node* z_start = array_element_address(z, intcon(0), z_elem);
6340
6341 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6342 OptoRuntime::squareToLen_Type(),
6343 stubAddr, stubName, TypePtr::BOTTOM,
6344 x_start, len, z_start, zlen);
6345
6346 set_result(z);
6347 return true;
6348 }
6349
6350 //-------------inline_mulAdd------------------------------------------
6351 bool LibraryCallKit::inline_mulAdd() {
6352 assert(UseMulAddIntrinsic, "not implemented on this platform");
6353
6354 address stubAddr = StubRoutines::mulAdd();
6355 if (stubAddr == nullptr) {
6356 return false; // Intrinsic's stub is not implemented on this platform
6357 }
6358 const char* stubName = "mulAdd";
6359
6360 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6361
6362 Node* out = argument(0);
6363 Node* in = argument(1);
6364 Node* offset = argument(2);
6365 Node* len = argument(3);
6366 Node* k = argument(4);
6367
6368 in = must_be_not_null(in, true);
6369 out = must_be_not_null(out, true);
6370
6371 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6372 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6373 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6374 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6375 // failed array check
6376 return false;
6377 }
6378
6379 BasicType out_elem = out_type->elem()->array_element_basic_type();
6380 BasicType in_elem = in_type->elem()->array_element_basic_type();
6381 if (out_elem != T_INT || in_elem != T_INT) {
6382 return false;
6383 }
6384
6385 Node* outlen = load_array_length(out);
6386 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6387 Node* out_start = array_element_address(out, intcon(0), out_elem);
6388 Node* in_start = array_element_address(in, intcon(0), in_elem);
6389
6390 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6391 OptoRuntime::mulAdd_Type(),
6392 stubAddr, stubName, TypePtr::BOTTOM,
6393 out_start,in_start, new_offset, len, k);
6394 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6395 set_result(result);
6396 return true;
6397 }
6398
6399 //-------------inline_montgomeryMultiply-----------------------------------
6400 bool LibraryCallKit::inline_montgomeryMultiply() {
6401 address stubAddr = StubRoutines::montgomeryMultiply();
6402 if (stubAddr == nullptr) {
6403 return false; // Intrinsic's stub is not implemented on this platform
6404 }
6405
6406 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6407 const char* stubName = "montgomery_multiply";
6408
6409 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6410
6411 Node* a = argument(0);
6412 Node* b = argument(1);
6413 Node* n = argument(2);
6414 Node* len = argument(3);
6415 Node* inv = argument(4);
6416 Node* m = argument(6);
6417
6418 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6419 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6420 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6421 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6422 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6423 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6424 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6425 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6426 // failed array check
6427 return false;
6428 }
6429
6430 BasicType a_elem = a_type->elem()->array_element_basic_type();
6431 BasicType b_elem = b_type->elem()->array_element_basic_type();
6432 BasicType n_elem = n_type->elem()->array_element_basic_type();
6433 BasicType m_elem = m_type->elem()->array_element_basic_type();
6434 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6435 return false;
6436 }
6437
6438 // Make the call
6439 {
6440 Node* a_start = array_element_address(a, intcon(0), a_elem);
6441 Node* b_start = array_element_address(b, intcon(0), b_elem);
6442 Node* n_start = array_element_address(n, intcon(0), n_elem);
6443 Node* m_start = array_element_address(m, intcon(0), m_elem);
6444
6445 Node* call = make_runtime_call(RC_LEAF,
6446 OptoRuntime::montgomeryMultiply_Type(),
6447 stubAddr, stubName, TypePtr::BOTTOM,
6448 a_start, b_start, n_start, len, inv, top(),
6449 m_start);
6450 set_result(m);
6451 }
6452
6453 return true;
6454 }
6455
6456 bool LibraryCallKit::inline_montgomerySquare() {
6457 address stubAddr = StubRoutines::montgomerySquare();
6458 if (stubAddr == nullptr) {
6459 return false; // Intrinsic's stub is not implemented on this platform
6460 }
6461
6462 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6463 const char* stubName = "montgomery_square";
6464
6465 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6466
6467 Node* a = argument(0);
6468 Node* n = argument(1);
6469 Node* len = argument(2);
6470 Node* inv = argument(3);
6471 Node* m = argument(5);
6472
6473 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6474 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6475 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6476 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6477 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6478 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6479 // failed array check
6480 return false;
6481 }
6482
6483 BasicType a_elem = a_type->elem()->array_element_basic_type();
6484 BasicType n_elem = n_type->elem()->array_element_basic_type();
6485 BasicType m_elem = m_type->elem()->array_element_basic_type();
6486 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6487 return false;
6488 }
6489
6490 // Make the call
6491 {
6492 Node* a_start = array_element_address(a, intcon(0), a_elem);
6493 Node* n_start = array_element_address(n, intcon(0), n_elem);
6494 Node* m_start = array_element_address(m, intcon(0), m_elem);
6495
6496 Node* call = make_runtime_call(RC_LEAF,
6497 OptoRuntime::montgomerySquare_Type(),
6498 stubAddr, stubName, TypePtr::BOTTOM,
6499 a_start, n_start, len, inv, top(),
6500 m_start);
6501 set_result(m);
6502 }
6503
6504 return true;
6505 }
6506
6507 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6508 address stubAddr = nullptr;
6509 const char* stubName = nullptr;
6510
6511 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6512 if (stubAddr == nullptr) {
6513 return false; // Intrinsic's stub is not implemented on this platform
6514 }
6515
6516 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6517
6518 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6519
6520 Node* newArr = argument(0);
6521 Node* oldArr = argument(1);
6522 Node* newIdx = argument(2);
6523 Node* shiftCount = argument(3);
6524 Node* numIter = argument(4);
6525
6526 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6527 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6528 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6529 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6530 return false;
6531 }
6532
6533 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6534 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6535 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6536 return false;
6537 }
6538
6539 // Make the call
6540 {
6541 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6542 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6543
6544 Node* call = make_runtime_call(RC_LEAF,
6545 OptoRuntime::bigIntegerShift_Type(),
6546 stubAddr,
6547 stubName,
6548 TypePtr::BOTTOM,
6549 newArr_start,
6550 oldArr_start,
6551 newIdx,
6552 shiftCount,
6553 numIter);
6554 }
6555
6556 return true;
6557 }
6558
6559 //-------------inline_vectorizedMismatch------------------------------
6560 bool LibraryCallKit::inline_vectorizedMismatch() {
6561 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6562
6563 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6564 Node* obja = argument(0); // Object
6565 Node* aoffset = argument(1); // long
6566 Node* objb = argument(3); // Object
6567 Node* boffset = argument(4); // long
6568 Node* length = argument(6); // int
6569 Node* scale = argument(7); // int
6570
6571 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6572 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6573 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6574 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6575 scale == top()) {
6576 return false; // failed input validation
6577 }
6578
6579 Node* obja_adr = make_unsafe_address(obja, aoffset);
6580 Node* objb_adr = make_unsafe_address(objb, boffset);
6581
6582 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6583 //
6584 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6585 // if (length <= inline_limit) {
6586 // inline_path:
6587 // vmask = VectorMaskGen length
6588 // vload1 = LoadVectorMasked obja, vmask
6589 // vload2 = LoadVectorMasked objb, vmask
6590 // result1 = VectorCmpMasked vload1, vload2, vmask
6591 // } else {
6592 // call_stub_path:
6593 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6594 // }
6595 // exit_block:
6596 // return Phi(result1, result2);
6597 //
6598 enum { inline_path = 1, // input is small enough to process it all at once
6599 stub_path = 2, // input is too large; call into the VM
6600 PATH_LIMIT = 3
6601 };
6602
6603 Node* exit_block = new RegionNode(PATH_LIMIT);
6604 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6605 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6606
6607 Node* call_stub_path = control();
6608
6609 BasicType elem_bt = T_ILLEGAL;
6610
6611 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6612 if (scale_t->is_con()) {
6613 switch (scale_t->get_con()) {
6614 case 0: elem_bt = T_BYTE; break;
6615 case 1: elem_bt = T_SHORT; break;
6616 case 2: elem_bt = T_INT; break;
6617 case 3: elem_bt = T_LONG; break;
6618
6619 default: elem_bt = T_ILLEGAL; break; // not supported
6620 }
6621 }
6622
6623 int inline_limit = 0;
6624 bool do_partial_inline = false;
6625
6626 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6627 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6628 do_partial_inline = inline_limit >= 16;
6629 }
6630
6631 if (do_partial_inline) {
6632 assert(elem_bt != T_ILLEGAL, "sanity");
6633
6634 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6635 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6636 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6637
6638 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6639 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6640 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6641
6642 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6643
6644 if (!stopped()) {
6645 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6646
6647 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6648 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6649 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6650 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6651
6652 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6653 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6654 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6655 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6656
6657 exit_block->init_req(inline_path, control());
6658 memory_phi->init_req(inline_path, map()->memory());
6659 result_phi->init_req(inline_path, result);
6660
6661 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6662 clear_upper_avx();
6663 }
6664 }
6665 }
6666
6667 if (call_stub_path != nullptr) {
6668 set_control(call_stub_path);
6669
6670 Node* call = make_runtime_call(RC_LEAF,
6671 OptoRuntime::vectorizedMismatch_Type(),
6672 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6673 obja_adr, objb_adr, length, scale);
6674
6675 exit_block->init_req(stub_path, control());
6676 memory_phi->init_req(stub_path, map()->memory());
6677 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6678 }
6679
6680 exit_block = _gvn.transform(exit_block);
6681 memory_phi = _gvn.transform(memory_phi);
6682 result_phi = _gvn.transform(result_phi);
6683
6684 record_for_igvn(exit_block);
6685 record_for_igvn(memory_phi);
6686 record_for_igvn(result_phi);
6687
6688 set_control(exit_block);
6689 set_all_memory(memory_phi);
6690 set_result(result_phi);
6691
6692 return true;
6693 }
6694
6695 //------------------------------inline_vectorizedHashcode----------------------------
6696 bool LibraryCallKit::inline_vectorizedHashCode() {
6697 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6698
6699 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6700 Node* array = argument(0);
6701 Node* offset = argument(1);
6702 Node* length = argument(2);
6703 Node* initialValue = argument(3);
6704 Node* basic_type = argument(4);
6705
6706 if (basic_type == top()) {
6707 return false; // failed input validation
6708 }
6709
6710 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6711 if (!basic_type_t->is_con()) {
6712 return false; // Only intrinsify if mode argument is constant
6713 }
6714
6715 array = must_be_not_null(array, true);
6716
6717 BasicType bt = (BasicType)basic_type_t->get_con();
6718
6719 // Resolve address of first element
6720 Node* array_start = array_element_address(array, offset, bt);
6721
6722 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6723 array_start, length, initialValue, basic_type)));
6724 clear_upper_avx();
6725
6726 return true;
6727 }
6728
6729 /**
6730 * Calculate CRC32 for byte.
6731 * int java.util.zip.CRC32.update(int crc, int b)
6732 */
6733 bool LibraryCallKit::inline_updateCRC32() {
6734 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6735 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6736 // no receiver since it is static method
6737 Node* crc = argument(0); // type: int
6738 Node* b = argument(1); // type: int
6739
6740 /*
6741 * int c = ~ crc;
6742 * b = timesXtoThe32[(b ^ c) & 0xFF];
6743 * b = b ^ (c >>> 8);
6744 * crc = ~b;
6745 */
6746
6747 Node* M1 = intcon(-1);
6748 crc = _gvn.transform(new XorINode(crc, M1));
6749 Node* result = _gvn.transform(new XorINode(crc, b));
6750 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6751
6752 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6753 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6754 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6755 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6756
6757 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6758 result = _gvn.transform(new XorINode(crc, result));
6759 result = _gvn.transform(new XorINode(result, M1));
6760 set_result(result);
6761 return true;
6762 }
6763
6764 /**
6765 * Calculate CRC32 for byte[] array.
6766 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6767 */
6768 bool LibraryCallKit::inline_updateBytesCRC32() {
6769 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6770 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6771 // no receiver since it is static method
6772 Node* crc = argument(0); // type: int
6773 Node* src = argument(1); // type: oop
6774 Node* offset = argument(2); // type: int
6775 Node* length = argument(3); // type: int
6776
6777 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6778 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6779 // failed array check
6780 return false;
6781 }
6782
6783 // Figure out the size and type of the elements we will be copying.
6784 BasicType src_elem = src_type->elem()->array_element_basic_type();
6785 if (src_elem != T_BYTE) {
6786 return false;
6787 }
6788
6789 // 'src_start' points to src array + scaled offset
6790 src = must_be_not_null(src, true);
6791 Node* src_start = array_element_address(src, offset, src_elem);
6792
6793 // We assume that range check is done by caller.
6794 // TODO: generate range check (offset+length < src.length) in debug VM.
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 /**
6809 * Calculate CRC32 for ByteBuffer.
6810 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6811 */
6812 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6813 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6814 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6815 // no receiver since it is static method
6816 Node* crc = argument(0); // type: int
6817 Node* src = argument(1); // type: long
6818 Node* offset = argument(3); // type: int
6819 Node* length = argument(4); // type: int
6820
6821 src = ConvL2X(src); // adjust Java long to machine word
6822 Node* base = _gvn.transform(new CastX2PNode(src));
6823 offset = ConvI2X(offset);
6824
6825 // 'src_start' points to src array + scaled offset
6826 Node* src_start = basic_plus_adr(top(), base, offset);
6827
6828 // Call the stub.
6829 address stubAddr = StubRoutines::updateBytesCRC32();
6830 const char *stubName = "updateBytesCRC32";
6831
6832 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6833 stubAddr, stubName, TypePtr::BOTTOM,
6834 crc, src_start, length);
6835 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6836 set_result(result);
6837 return true;
6838 }
6839
6840 //------------------------------get_table_from_crc32c_class-----------------------
6841 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6842 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6843 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6844
6845 return table;
6846 }
6847
6848 //------------------------------inline_updateBytesCRC32C-----------------------
6849 //
6850 // Calculate CRC32C for byte[] array.
6851 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6852 //
6853 bool LibraryCallKit::inline_updateBytesCRC32C() {
6854 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6855 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6856 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6857 // no receiver since it is a static method
6858 Node* crc = argument(0); // type: int
6859 Node* src = argument(1); // type: oop
6860 Node* offset = argument(2); // type: int
6861 Node* end = argument(3); // type: int
6862
6863 Node* length = _gvn.transform(new SubINode(end, offset));
6864
6865 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6866 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6867 // failed array check
6868 return false;
6869 }
6870
6871 // Figure out the size and type of the elements we will be copying.
6872 BasicType src_elem = src_type->elem()->array_element_basic_type();
6873 if (src_elem != T_BYTE) {
6874 return false;
6875 }
6876
6877 // 'src_start' points to src array + scaled offset
6878 src = must_be_not_null(src, true);
6879 Node* src_start = array_element_address(src, offset, src_elem);
6880
6881 // static final int[] byteTable in class CRC32C
6882 Node* table = get_table_from_crc32c_class(callee()->holder());
6883 table = must_be_not_null(table, true);
6884 Node* table_start = array_element_address(table, intcon(0), T_INT);
6885
6886 // We assume that range check is done by caller.
6887 // TODO: generate range check (offset+length < src.length) in debug VM.
6888
6889 // Call the stub.
6890 address stubAddr = StubRoutines::updateBytesCRC32C();
6891 const char *stubName = "updateBytesCRC32C";
6892
6893 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6894 stubAddr, stubName, TypePtr::BOTTOM,
6895 crc, src_start, length, table_start);
6896 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6897 set_result(result);
6898 return true;
6899 }
6900
6901 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6902 //
6903 // Calculate CRC32C for DirectByteBuffer.
6904 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6905 //
6906 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6907 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6908 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6909 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6910 // no receiver since it is a static method
6911 Node* crc = argument(0); // type: int
6912 Node* src = argument(1); // type: long
6913 Node* offset = argument(3); // type: int
6914 Node* end = argument(4); // type: int
6915
6916 Node* length = _gvn.transform(new SubINode(end, offset));
6917
6918 src = ConvL2X(src); // adjust Java long to machine word
6919 Node* base = _gvn.transform(new CastX2PNode(src));
6920 offset = ConvI2X(offset);
6921
6922 // 'src_start' points to src array + scaled offset
6923 Node* src_start = basic_plus_adr(top(), base, offset);
6924
6925 // static final int[] byteTable in class CRC32C
6926 Node* table = get_table_from_crc32c_class(callee()->holder());
6927 table = must_be_not_null(table, true);
6928 Node* table_start = array_element_address(table, intcon(0), T_INT);
6929
6930 // Call the stub.
6931 address stubAddr = StubRoutines::updateBytesCRC32C();
6932 const char *stubName = "updateBytesCRC32C";
6933
6934 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6935 stubAddr, stubName, TypePtr::BOTTOM,
6936 crc, src_start, length, table_start);
6937 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6938 set_result(result);
6939 return true;
6940 }
6941
6942 //------------------------------inline_updateBytesAdler32----------------------
6943 //
6944 // Calculate Adler32 checksum for byte[] array.
6945 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6946 //
6947 bool LibraryCallKit::inline_updateBytesAdler32() {
6948 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6949 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6950 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6951 // no receiver since it is static method
6952 Node* crc = argument(0); // type: int
6953 Node* src = argument(1); // type: oop
6954 Node* offset = argument(2); // type: int
6955 Node* length = argument(3); // type: int
6956
6957 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6958 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6959 // failed array check
6960 return false;
6961 }
6962
6963 // Figure out the size and type of the elements we will be copying.
6964 BasicType src_elem = src_type->elem()->array_element_basic_type();
6965 if (src_elem != T_BYTE) {
6966 return false;
6967 }
6968
6969 // 'src_start' points to src array + scaled offset
6970 Node* src_start = array_element_address(src, offset, src_elem);
6971
6972 // We assume that range check is done by caller.
6973 // TODO: generate range check (offset+length < src.length) in debug VM.
6974
6975 // Call the stub.
6976 address stubAddr = StubRoutines::updateBytesAdler32();
6977 const char *stubName = "updateBytesAdler32";
6978
6979 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6980 stubAddr, stubName, TypePtr::BOTTOM,
6981 crc, src_start, length);
6982 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6983 set_result(result);
6984 return true;
6985 }
6986
6987 //------------------------------inline_updateByteBufferAdler32---------------
6988 //
6989 // Calculate Adler32 checksum for DirectByteBuffer.
6990 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6991 //
6992 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6993 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6994 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6995 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6996 // no receiver since it is static method
6997 Node* crc = argument(0); // type: int
6998 Node* src = argument(1); // type: long
6999 Node* offset = argument(3); // type: int
7000 Node* length = argument(4); // type: int
7001
7002 src = ConvL2X(src); // adjust Java long to machine word
7003 Node* base = _gvn.transform(new CastX2PNode(src));
7004 offset = ConvI2X(offset);
7005
7006 // 'src_start' points to src array + scaled offset
7007 Node* src_start = basic_plus_adr(top(), base, offset);
7008
7009 // Call the stub.
7010 address stubAddr = StubRoutines::updateBytesAdler32();
7011 const char *stubName = "updateBytesAdler32";
7012
7013 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7014 stubAddr, stubName, TypePtr::BOTTOM,
7015 crc, src_start, length);
7016
7017 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7018 set_result(result);
7019 return true;
7020 }
7021
7022 //----------------------------inline_reference_get0----------------------------
7023 // public T java.lang.ref.Reference.get();
7024 bool LibraryCallKit::inline_reference_get0() {
7025 const int referent_offset = java_lang_ref_Reference::referent_offset();
7026
7027 // Get the argument:
7028 Node* reference_obj = null_check_receiver();
7029 if (stopped()) return true;
7030
7031 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7032 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7033 decorators, /*is_static*/ false, nullptr);
7034 if (result == nullptr) return false;
7035
7036 // Add memory barrier to prevent commoning reads from this field
7037 // across safepoint since GC can change its value.
7038 insert_mem_bar(Op_MemBarCPUOrder);
7039
7040 set_result(result);
7041 return true;
7042 }
7043
7044 //----------------------------inline_reference_refersTo0----------------------------
7045 // bool java.lang.ref.Reference.refersTo0();
7046 // bool java.lang.ref.PhantomReference.refersTo0();
7047 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7048 // Get arguments:
7049 Node* reference_obj = null_check_receiver();
7050 Node* other_obj = argument(1);
7051 if (stopped()) return true;
7052
7053 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7054 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7055 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7056 decorators, /*is_static*/ false, nullptr);
7057 if (referent == nullptr) return false;
7058
7059 // Add memory barrier to prevent commoning reads from this field
7060 // across safepoint since GC can change its value.
7061 insert_mem_bar(Op_MemBarCPUOrder);
7062
7063 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7064 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7065 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7066
7067 RegionNode* region = new RegionNode(3);
7068 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7069
7070 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7071 region->init_req(1, if_true);
7072 phi->init_req(1, intcon(1));
7073
7074 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7075 region->init_req(2, if_false);
7076 phi->init_req(2, intcon(0));
7077
7078 set_control(_gvn.transform(region));
7079 record_for_igvn(region);
7080 set_result(_gvn.transform(phi));
7081 return true;
7082 }
7083
7084 //----------------------------inline_reference_clear0----------------------------
7085 // void java.lang.ref.Reference.clear0();
7086 // void java.lang.ref.PhantomReference.clear0();
7087 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7088 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7089
7090 // Get arguments
7091 Node* reference_obj = null_check_receiver();
7092 if (stopped()) return true;
7093
7094 // Common access parameters
7095 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7096 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7097 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7098 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7099 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7100
7101 Node* referent = access_load_at(reference_obj,
7102 referent_field_addr,
7103 referent_field_addr_type,
7104 val_type,
7105 T_OBJECT,
7106 decorators);
7107
7108 IdealKit ideal(this);
7109 #define __ ideal.
7110 __ if_then(referent, BoolTest::ne, null());
7111 sync_kit(ideal);
7112 access_store_at(reference_obj,
7113 referent_field_addr,
7114 referent_field_addr_type,
7115 null(),
7116 val_type,
7117 T_OBJECT,
7118 decorators);
7119 __ sync_kit(this);
7120 __ end_if();
7121 final_sync(ideal);
7122 #undef __
7123
7124 return true;
7125 }
7126
7127 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7128 DecoratorSet decorators, bool is_static,
7129 ciInstanceKlass* fromKls) {
7130 if (fromKls == nullptr) {
7131 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7132 assert(tinst != nullptr, "obj is null");
7133 assert(tinst->is_loaded(), "obj is not loaded");
7134 fromKls = tinst->instance_klass();
7135 } else {
7136 assert(is_static, "only for static field access");
7137 }
7138 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7139 ciSymbol::make(fieldTypeString),
7140 is_static);
7141
7142 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7143 if (field == nullptr) return (Node *) nullptr;
7144
7145 if (is_static) {
7146 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7147 fromObj = makecon(tip);
7148 }
7149
7150 // Next code copied from Parse::do_get_xxx():
7151
7152 // Compute address and memory type.
7153 int offset = field->offset_in_bytes();
7154 bool is_vol = field->is_volatile();
7155 ciType* field_klass = field->type();
7156 assert(field_klass->is_loaded(), "should be loaded");
7157 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7158 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7159 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7160 "slice of address and input slice don't match");
7161 BasicType bt = field->layout_type();
7162
7163 // Build the resultant type of the load
7164 const Type *type;
7165 if (bt == T_OBJECT) {
7166 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7167 } else {
7168 type = Type::get_const_basic_type(bt);
7169 }
7170
7171 if (is_vol) {
7172 decorators |= MO_SEQ_CST;
7173 }
7174
7175 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7176 }
7177
7178 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7179 bool is_exact /* true */, bool is_static /* false */,
7180 ciInstanceKlass * fromKls /* nullptr */) {
7181 if (fromKls == nullptr) {
7182 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7183 assert(tinst != nullptr, "obj is null");
7184 assert(tinst->is_loaded(), "obj is not loaded");
7185 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7186 fromKls = tinst->instance_klass();
7187 }
7188 else {
7189 assert(is_static, "only for static field access");
7190 }
7191 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7192 ciSymbol::make(fieldTypeString),
7193 is_static);
7194
7195 assert(field != nullptr, "undefined field");
7196 assert(!field->is_volatile(), "not defined for volatile fields");
7197
7198 if (is_static) {
7199 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7200 fromObj = makecon(tip);
7201 }
7202
7203 // Next code copied from Parse::do_get_xxx():
7204
7205 // Compute address and memory type.
7206 int offset = field->offset_in_bytes();
7207 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7208
7209 return adr;
7210 }
7211
7212 //------------------------------inline_aescrypt_Block-----------------------
7213 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7214 address stubAddr = nullptr;
7215 const char *stubName;
7216 bool is_decrypt = false;
7217 assert(UseAES, "need AES instruction support");
7218
7219 switch(id) {
7220 case vmIntrinsics::_aescrypt_encryptBlock:
7221 stubAddr = StubRoutines::aescrypt_encryptBlock();
7222 stubName = "aescrypt_encryptBlock";
7223 break;
7224 case vmIntrinsics::_aescrypt_decryptBlock:
7225 stubAddr = StubRoutines::aescrypt_decryptBlock();
7226 stubName = "aescrypt_decryptBlock";
7227 is_decrypt = true;
7228 break;
7229 default:
7230 break;
7231 }
7232 if (stubAddr == nullptr) return false;
7233
7234 Node* aescrypt_object = argument(0);
7235 Node* src = argument(1);
7236 Node* src_offset = argument(2);
7237 Node* dest = argument(3);
7238 Node* dest_offset = argument(4);
7239
7240 src = must_be_not_null(src, true);
7241 dest = must_be_not_null(dest, true);
7242
7243 // (1) src and dest are arrays.
7244 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7245 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7246 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7247 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7248
7249 // for the quick and dirty code we will skip all the checks.
7250 // we are just trying to get the call to be generated.
7251 Node* src_start = src;
7252 Node* dest_start = dest;
7253 if (src_offset != nullptr || dest_offset != nullptr) {
7254 assert(src_offset != nullptr && dest_offset != nullptr, "");
7255 src_start = array_element_address(src, src_offset, T_BYTE);
7256 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7257 }
7258
7259 // now need to get the start of its expanded key array
7260 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7261 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7262 if (k_start == nullptr) return false;
7263
7264 // Call the stub.
7265 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7266 stubAddr, stubName, TypePtr::BOTTOM,
7267 src_start, dest_start, k_start);
7268
7269 return true;
7270 }
7271
7272 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7273 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7274 address stubAddr = nullptr;
7275 const char *stubName = nullptr;
7276 bool is_decrypt = false;
7277 assert(UseAES, "need AES instruction support");
7278
7279 switch(id) {
7280 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7281 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7282 stubName = "cipherBlockChaining_encryptAESCrypt";
7283 break;
7284 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7285 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7286 stubName = "cipherBlockChaining_decryptAESCrypt";
7287 is_decrypt = true;
7288 break;
7289 default:
7290 break;
7291 }
7292 if (stubAddr == nullptr) return false;
7293
7294 Node* cipherBlockChaining_object = argument(0);
7295 Node* src = argument(1);
7296 Node* src_offset = argument(2);
7297 Node* len = argument(3);
7298 Node* dest = argument(4);
7299 Node* dest_offset = argument(5);
7300
7301 src = must_be_not_null(src, false);
7302 dest = must_be_not_null(dest, false);
7303
7304 // (1) src and dest are arrays.
7305 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7306 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7307 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7308 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7309
7310 // checks are the responsibility of the caller
7311 Node* src_start = src;
7312 Node* dest_start = dest;
7313 if (src_offset != nullptr || dest_offset != nullptr) {
7314 assert(src_offset != nullptr && dest_offset != nullptr, "");
7315 src_start = array_element_address(src, src_offset, T_BYTE);
7316 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7317 }
7318
7319 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7320 // (because of the predicated logic executed earlier).
7321 // so we cast it here safely.
7322 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7323
7324 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7325 if (embeddedCipherObj == nullptr) return false;
7326
7327 // cast it to what we know it will be at runtime
7328 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7329 assert(tinst != nullptr, "CBC obj is null");
7330 assert(tinst->is_loaded(), "CBC obj is not loaded");
7331 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7332 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7333
7334 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7335 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7336 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7337 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7338 aescrypt_object = _gvn.transform(aescrypt_object);
7339
7340 // we need to get the start of the aescrypt_object's expanded key array
7341 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7342 if (k_start == nullptr) return false;
7343
7344 // similarly, get the start address of the r vector
7345 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7346 if (objRvec == nullptr) return false;
7347 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7348
7349 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7350 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7351 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7352 stubAddr, stubName, TypePtr::BOTTOM,
7353 src_start, dest_start, k_start, r_start, len);
7354
7355 // return cipher length (int)
7356 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7357 set_result(retvalue);
7358 return true;
7359 }
7360
7361 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7362 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7363 address stubAddr = nullptr;
7364 const char *stubName = nullptr;
7365 bool is_decrypt = false;
7366 assert(UseAES, "need AES instruction support");
7367
7368 switch (id) {
7369 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7370 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7371 stubName = "electronicCodeBook_encryptAESCrypt";
7372 break;
7373 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7374 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7375 stubName = "electronicCodeBook_decryptAESCrypt";
7376 is_decrypt = true;
7377 break;
7378 default:
7379 break;
7380 }
7381
7382 if (stubAddr == nullptr) return false;
7383
7384 Node* electronicCodeBook_object = argument(0);
7385 Node* src = argument(1);
7386 Node* src_offset = argument(2);
7387 Node* len = argument(3);
7388 Node* dest = argument(4);
7389 Node* dest_offset = argument(5);
7390
7391 // (1) src and dest are arrays.
7392 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7393 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7394 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7395 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7396
7397 // checks are the responsibility of the caller
7398 Node* src_start = src;
7399 Node* dest_start = dest;
7400 if (src_offset != nullptr || dest_offset != nullptr) {
7401 assert(src_offset != nullptr && dest_offset != nullptr, "");
7402 src_start = array_element_address(src, src_offset, T_BYTE);
7403 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7404 }
7405
7406 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7407 // (because of the predicated logic executed earlier).
7408 // so we cast it here safely.
7409 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7410
7411 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7412 if (embeddedCipherObj == nullptr) return false;
7413
7414 // cast it to what we know it will be at runtime
7415 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7416 assert(tinst != nullptr, "ECB obj is null");
7417 assert(tinst->is_loaded(), "ECB obj is not loaded");
7418 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7419 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7420
7421 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7422 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7423 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7424 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7425 aescrypt_object = _gvn.transform(aescrypt_object);
7426
7427 // we need to get the start of the aescrypt_object's expanded key array
7428 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7429 if (k_start == nullptr) return false;
7430
7431 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7432 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7433 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7434 stubAddr, stubName, TypePtr::BOTTOM,
7435 src_start, dest_start, k_start, len);
7436
7437 // return cipher length (int)
7438 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7439 set_result(retvalue);
7440 return true;
7441 }
7442
7443 //------------------------------inline_counterMode_AESCrypt-----------------------
7444 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7445 assert(UseAES, "need AES instruction support");
7446 if (!UseAESCTRIntrinsics) return false;
7447
7448 address stubAddr = nullptr;
7449 const char *stubName = nullptr;
7450 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7451 stubAddr = StubRoutines::counterMode_AESCrypt();
7452 stubName = "counterMode_AESCrypt";
7453 }
7454 if (stubAddr == nullptr) return false;
7455
7456 Node* counterMode_object = argument(0);
7457 Node* src = argument(1);
7458 Node* src_offset = argument(2);
7459 Node* len = argument(3);
7460 Node* dest = argument(4);
7461 Node* dest_offset = argument(5);
7462
7463 // (1) src and dest are arrays.
7464 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7465 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7466 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7467 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7468
7469 // checks are the responsibility of the caller
7470 Node* src_start = src;
7471 Node* dest_start = dest;
7472 if (src_offset != nullptr || dest_offset != nullptr) {
7473 assert(src_offset != nullptr && dest_offset != nullptr, "");
7474 src_start = array_element_address(src, src_offset, T_BYTE);
7475 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7476 }
7477
7478 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7479 // (because of the predicated logic executed earlier).
7480 // so we cast it here safely.
7481 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7482 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7483 if (embeddedCipherObj == nullptr) return false;
7484 // cast it to what we know it will be at runtime
7485 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7486 assert(tinst != nullptr, "CTR obj is null");
7487 assert(tinst->is_loaded(), "CTR obj is not loaded");
7488 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7489 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7490 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7491 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7492 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7493 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7494 aescrypt_object = _gvn.transform(aescrypt_object);
7495 // we need to get the start of the aescrypt_object's expanded key array
7496 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7497 if (k_start == nullptr) return false;
7498 // similarly, get the start address of the r vector
7499 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7500 if (obj_counter == nullptr) return false;
7501 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7502
7503 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7504 if (saved_encCounter == nullptr) return false;
7505 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7506 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7507
7508 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7509 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7510 OptoRuntime::counterMode_aescrypt_Type(),
7511 stubAddr, stubName, TypePtr::BOTTOM,
7512 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7513
7514 // return cipher length (int)
7515 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7516 set_result(retvalue);
7517 return true;
7518 }
7519
7520 //------------------------------get_key_start_from_aescrypt_object-----------------------
7521 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7522 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7523 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7524 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7525 // The following platform specific stubs of encryption and decryption use the same round keys.
7526 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7527 bool use_decryption_key = false;
7528 #else
7529 bool use_decryption_key = is_decrypt;
7530 #endif
7531 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7532 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7533 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7534
7535 // now have the array, need to get the start address of the selected key array
7536 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7537 return k_start;
7538 }
7539
7540 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7541 // Return node representing slow path of predicate check.
7542 // the pseudo code we want to emulate with this predicate is:
7543 // for encryption:
7544 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7545 // for decryption:
7546 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7547 // note cipher==plain is more conservative than the original java code but that's OK
7548 //
7549 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7550 // The receiver was checked for null already.
7551 Node* objCBC = argument(0);
7552
7553 Node* src = argument(1);
7554 Node* dest = argument(4);
7555
7556 // Load embeddedCipher field of CipherBlockChaining object.
7557 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7558
7559 // get AESCrypt klass for instanceOf check
7560 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7561 // will have same classloader as CipherBlockChaining object
7562 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7563 assert(tinst != nullptr, "CBCobj is null");
7564 assert(tinst->is_loaded(), "CBCobj is not loaded");
7565
7566 // we want to do an instanceof comparison against the AESCrypt class
7567 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7568 if (!klass_AESCrypt->is_loaded()) {
7569 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7570 Node* ctrl = control();
7571 set_control(top()); // no regular fast path
7572 return ctrl;
7573 }
7574
7575 src = must_be_not_null(src, true);
7576 dest = must_be_not_null(dest, true);
7577
7578 // Resolve oops to stable for CmpP below.
7579 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7580
7581 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7582 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7583 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7584
7585 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7586
7587 // for encryption, we are done
7588 if (!decrypting)
7589 return instof_false; // even if it is null
7590
7591 // for decryption, we need to add a further check to avoid
7592 // taking the intrinsic path when cipher and plain are the same
7593 // see the original java code for why.
7594 RegionNode* region = new RegionNode(3);
7595 region->init_req(1, instof_false);
7596
7597 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7598 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7599 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7600 region->init_req(2, src_dest_conjoint);
7601
7602 record_for_igvn(region);
7603 return _gvn.transform(region);
7604 }
7605
7606 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7607 // Return node representing slow path of predicate check.
7608 // the pseudo code we want to emulate with this predicate is:
7609 // for encryption:
7610 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7611 // for decryption:
7612 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7613 // note cipher==plain is more conservative than the original java code but that's OK
7614 //
7615 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7616 // The receiver was checked for null already.
7617 Node* objECB = argument(0);
7618
7619 // Load embeddedCipher field of ElectronicCodeBook object.
7620 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7621
7622 // get AESCrypt klass for instanceOf check
7623 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7624 // will have same classloader as ElectronicCodeBook object
7625 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7626 assert(tinst != nullptr, "ECBobj is null");
7627 assert(tinst->is_loaded(), "ECBobj is not loaded");
7628
7629 // we want to do an instanceof comparison against the AESCrypt class
7630 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7631 if (!klass_AESCrypt->is_loaded()) {
7632 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7633 Node* ctrl = control();
7634 set_control(top()); // no regular fast path
7635 return ctrl;
7636 }
7637 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7638
7639 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7640 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7641 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7642
7643 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7644
7645 // for encryption, we are done
7646 if (!decrypting)
7647 return instof_false; // even if it is null
7648
7649 // for decryption, we need to add a further check to avoid
7650 // taking the intrinsic path when cipher and plain are the same
7651 // see the original java code for why.
7652 RegionNode* region = new RegionNode(3);
7653 region->init_req(1, instof_false);
7654 Node* src = argument(1);
7655 Node* dest = argument(4);
7656 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7657 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7658 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7659 region->init_req(2, src_dest_conjoint);
7660
7661 record_for_igvn(region);
7662 return _gvn.transform(region);
7663 }
7664
7665 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7666 // Return node representing slow path of predicate check.
7667 // the pseudo code we want to emulate with this predicate is:
7668 // for encryption:
7669 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7670 // for decryption:
7671 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7672 // note cipher==plain is more conservative than the original java code but that's OK
7673 //
7674
7675 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7676 // The receiver was checked for null already.
7677 Node* objCTR = argument(0);
7678
7679 // Load embeddedCipher field of CipherBlockChaining object.
7680 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7681
7682 // get AESCrypt klass for instanceOf check
7683 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7684 // will have same classloader as CipherBlockChaining object
7685 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7686 assert(tinst != nullptr, "CTRobj is null");
7687 assert(tinst->is_loaded(), "CTRobj is not loaded");
7688
7689 // we want to do an instanceof comparison against the AESCrypt class
7690 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7691 if (!klass_AESCrypt->is_loaded()) {
7692 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7693 Node* ctrl = control();
7694 set_control(top()); // no regular fast path
7695 return ctrl;
7696 }
7697
7698 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7699 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7700 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7701 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7702 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7703
7704 return instof_false; // even if it is null
7705 }
7706
7707 //------------------------------inline_ghash_processBlocks
7708 bool LibraryCallKit::inline_ghash_processBlocks() {
7709 address stubAddr;
7710 const char *stubName;
7711 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7712
7713 stubAddr = StubRoutines::ghash_processBlocks();
7714 stubName = "ghash_processBlocks";
7715
7716 Node* data = argument(0);
7717 Node* offset = argument(1);
7718 Node* len = argument(2);
7719 Node* state = argument(3);
7720 Node* subkeyH = argument(4);
7721
7722 state = must_be_not_null(state, true);
7723 subkeyH = must_be_not_null(subkeyH, true);
7724 data = must_be_not_null(data, true);
7725
7726 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7727 assert(state_start, "state is null");
7728 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7729 assert(subkeyH_start, "subkeyH is null");
7730 Node* data_start = array_element_address(data, offset, T_BYTE);
7731 assert(data_start, "data is null");
7732
7733 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7734 OptoRuntime::ghash_processBlocks_Type(),
7735 stubAddr, stubName, TypePtr::BOTTOM,
7736 state_start, subkeyH_start, data_start, len);
7737 return true;
7738 }
7739
7740 //------------------------------inline_chacha20Block
7741 bool LibraryCallKit::inline_chacha20Block() {
7742 address stubAddr;
7743 const char *stubName;
7744 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7745
7746 stubAddr = StubRoutines::chacha20Block();
7747 stubName = "chacha20Block";
7748
7749 Node* state = argument(0);
7750 Node* result = argument(1);
7751
7752 state = must_be_not_null(state, true);
7753 result = must_be_not_null(result, true);
7754
7755 Node* state_start = array_element_address(state, intcon(0), T_INT);
7756 assert(state_start, "state is null");
7757 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7758 assert(result_start, "result is null");
7759
7760 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7761 OptoRuntime::chacha20Block_Type(),
7762 stubAddr, stubName, TypePtr::BOTTOM,
7763 state_start, result_start);
7764 // return key stream length (int)
7765 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7766 set_result(retvalue);
7767 return true;
7768 }
7769
7770 //------------------------------inline_kyberNtt
7771 bool LibraryCallKit::inline_kyberNtt() {
7772 address stubAddr;
7773 const char *stubName;
7774 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7775 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7776
7777 stubAddr = StubRoutines::kyberNtt();
7778 stubName = "kyberNtt";
7779 if (!stubAddr) return false;
7780
7781 Node* coeffs = argument(0);
7782 Node* ntt_zetas = argument(1);
7783
7784 coeffs = must_be_not_null(coeffs, true);
7785 ntt_zetas = must_be_not_null(ntt_zetas, true);
7786
7787 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7788 assert(coeffs_start, "coeffs is null");
7789 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7790 assert(ntt_zetas_start, "ntt_zetas is null");
7791 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7792 OptoRuntime::kyberNtt_Type(),
7793 stubAddr, stubName, TypePtr::BOTTOM,
7794 coeffs_start, ntt_zetas_start);
7795 // return an int
7796 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7797 set_result(retvalue);
7798 return true;
7799 }
7800
7801 //------------------------------inline_kyberInverseNtt
7802 bool LibraryCallKit::inline_kyberInverseNtt() {
7803 address stubAddr;
7804 const char *stubName;
7805 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7806 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7807
7808 stubAddr = StubRoutines::kyberInverseNtt();
7809 stubName = "kyberInverseNtt";
7810 if (!stubAddr) return false;
7811
7812 Node* coeffs = argument(0);
7813 Node* zetas = argument(1);
7814
7815 coeffs = must_be_not_null(coeffs, true);
7816 zetas = must_be_not_null(zetas, true);
7817
7818 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7819 assert(coeffs_start, "coeffs is null");
7820 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7821 assert(zetas_start, "inverseNtt_zetas is null");
7822 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7823 OptoRuntime::kyberInverseNtt_Type(),
7824 stubAddr, stubName, TypePtr::BOTTOM,
7825 coeffs_start, zetas_start);
7826
7827 // return an int
7828 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7829 set_result(retvalue);
7830 return true;
7831 }
7832
7833 //------------------------------inline_kyberNttMult
7834 bool LibraryCallKit::inline_kyberNttMult() {
7835 address stubAddr;
7836 const char *stubName;
7837 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7838 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7839
7840 stubAddr = StubRoutines::kyberNttMult();
7841 stubName = "kyberNttMult";
7842 if (!stubAddr) return false;
7843
7844 Node* result = argument(0);
7845 Node* ntta = argument(1);
7846 Node* nttb = argument(2);
7847 Node* zetas = argument(3);
7848
7849 result = must_be_not_null(result, true);
7850 ntta = must_be_not_null(ntta, true);
7851 nttb = must_be_not_null(nttb, true);
7852 zetas = must_be_not_null(zetas, true);
7853
7854 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7855 assert(result_start, "result is null");
7856 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7857 assert(ntta_start, "ntta is null");
7858 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7859 assert(nttb_start, "nttb is null");
7860 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7861 assert(zetas_start, "nttMult_zetas is null");
7862 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7863 OptoRuntime::kyberNttMult_Type(),
7864 stubAddr, stubName, TypePtr::BOTTOM,
7865 result_start, ntta_start, nttb_start,
7866 zetas_start);
7867
7868 // return an int
7869 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7870 set_result(retvalue);
7871
7872 return true;
7873 }
7874
7875 //------------------------------inline_kyberAddPoly_2
7876 bool LibraryCallKit::inline_kyberAddPoly_2() {
7877 address stubAddr;
7878 const char *stubName;
7879 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7880 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7881
7882 stubAddr = StubRoutines::kyberAddPoly_2();
7883 stubName = "kyberAddPoly_2";
7884 if (!stubAddr) return false;
7885
7886 Node* result = argument(0);
7887 Node* a = argument(1);
7888 Node* b = argument(2);
7889
7890 result = must_be_not_null(result, true);
7891 a = must_be_not_null(a, true);
7892 b = must_be_not_null(b, true);
7893
7894 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7895 assert(result_start, "result is null");
7896 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7897 assert(a_start, "a is null");
7898 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7899 assert(b_start, "b is null");
7900 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7901 OptoRuntime::kyberAddPoly_2_Type(),
7902 stubAddr, stubName, TypePtr::BOTTOM,
7903 result_start, a_start, b_start);
7904 // return an int
7905 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7906 set_result(retvalue);
7907 return true;
7908 }
7909
7910 //------------------------------inline_kyberAddPoly_3
7911 bool LibraryCallKit::inline_kyberAddPoly_3() {
7912 address stubAddr;
7913 const char *stubName;
7914 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7915 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7916
7917 stubAddr = StubRoutines::kyberAddPoly_3();
7918 stubName = "kyberAddPoly_3";
7919 if (!stubAddr) return false;
7920
7921 Node* result = argument(0);
7922 Node* a = argument(1);
7923 Node* b = argument(2);
7924 Node* c = argument(3);
7925
7926 result = must_be_not_null(result, true);
7927 a = must_be_not_null(a, true);
7928 b = must_be_not_null(b, true);
7929 c = must_be_not_null(c, true);
7930
7931 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7932 assert(result_start, "result is null");
7933 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7934 assert(a_start, "a is null");
7935 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7936 assert(b_start, "b is null");
7937 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7938 assert(c_start, "c is null");
7939 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7940 OptoRuntime::kyberAddPoly_3_Type(),
7941 stubAddr, stubName, TypePtr::BOTTOM,
7942 result_start, a_start, b_start, c_start);
7943 // return an int
7944 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7945 set_result(retvalue);
7946 return true;
7947 }
7948
7949 //------------------------------inline_kyber12To16
7950 bool LibraryCallKit::inline_kyber12To16() {
7951 address stubAddr;
7952 const char *stubName;
7953 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7954 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7955
7956 stubAddr = StubRoutines::kyber12To16();
7957 stubName = "kyber12To16";
7958 if (!stubAddr) return false;
7959
7960 Node* condensed = argument(0);
7961 Node* condensedOffs = argument(1);
7962 Node* parsed = argument(2);
7963 Node* parsedLength = argument(3);
7964
7965 condensed = must_be_not_null(condensed, true);
7966 parsed = must_be_not_null(parsed, true);
7967
7968 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7969 assert(condensed_start, "condensed is null");
7970 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7971 assert(parsed_start, "parsed is null");
7972 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7973 OptoRuntime::kyber12To16_Type(),
7974 stubAddr, stubName, TypePtr::BOTTOM,
7975 condensed_start, condensedOffs, parsed_start, parsedLength);
7976 // return an int
7977 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7978 set_result(retvalue);
7979 return true;
7980
7981 }
7982
7983 //------------------------------inline_kyberBarrettReduce
7984 bool LibraryCallKit::inline_kyberBarrettReduce() {
7985 address stubAddr;
7986 const char *stubName;
7987 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7988 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7989
7990 stubAddr = StubRoutines::kyberBarrettReduce();
7991 stubName = "kyberBarrettReduce";
7992 if (!stubAddr) return false;
7993
7994 Node* coeffs = argument(0);
7995
7996 coeffs = must_be_not_null(coeffs, true);
7997
7998 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7999 assert(coeffs_start, "coeffs is null");
8000 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8001 OptoRuntime::kyberBarrettReduce_Type(),
8002 stubAddr, stubName, TypePtr::BOTTOM,
8003 coeffs_start);
8004 // return an int
8005 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8006 set_result(retvalue);
8007 return true;
8008 }
8009
8010 //------------------------------inline_dilithiumAlmostNtt
8011 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8012 address stubAddr;
8013 const char *stubName;
8014 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8015 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8016
8017 stubAddr = StubRoutines::dilithiumAlmostNtt();
8018 stubName = "dilithiumAlmostNtt";
8019 if (!stubAddr) return false;
8020
8021 Node* coeffs = argument(0);
8022 Node* ntt_zetas = argument(1);
8023
8024 coeffs = must_be_not_null(coeffs, true);
8025 ntt_zetas = must_be_not_null(ntt_zetas, true);
8026
8027 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8028 assert(coeffs_start, "coeffs is null");
8029 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8030 assert(ntt_zetas_start, "ntt_zetas is null");
8031 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8032 OptoRuntime::dilithiumAlmostNtt_Type(),
8033 stubAddr, stubName, TypePtr::BOTTOM,
8034 coeffs_start, ntt_zetas_start);
8035 // return an int
8036 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8037 set_result(retvalue);
8038 return true;
8039 }
8040
8041 //------------------------------inline_dilithiumAlmostInverseNtt
8042 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8043 address stubAddr;
8044 const char *stubName;
8045 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8046 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8047
8048 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8049 stubName = "dilithiumAlmostInverseNtt";
8050 if (!stubAddr) return false;
8051
8052 Node* coeffs = argument(0);
8053 Node* zetas = argument(1);
8054
8055 coeffs = must_be_not_null(coeffs, true);
8056 zetas = must_be_not_null(zetas, true);
8057
8058 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8059 assert(coeffs_start, "coeffs is null");
8060 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8061 assert(zetas_start, "inverseNtt_zetas is null");
8062 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8063 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8064 stubAddr, stubName, TypePtr::BOTTOM,
8065 coeffs_start, zetas_start);
8066 // return an int
8067 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8068 set_result(retvalue);
8069 return true;
8070 }
8071
8072 //------------------------------inline_dilithiumNttMult
8073 bool LibraryCallKit::inline_dilithiumNttMult() {
8074 address stubAddr;
8075 const char *stubName;
8076 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8077 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8078
8079 stubAddr = StubRoutines::dilithiumNttMult();
8080 stubName = "dilithiumNttMult";
8081 if (!stubAddr) return false;
8082
8083 Node* result = argument(0);
8084 Node* ntta = argument(1);
8085 Node* nttb = argument(2);
8086 Node* zetas = argument(3);
8087
8088 result = must_be_not_null(result, true);
8089 ntta = must_be_not_null(ntta, true);
8090 nttb = must_be_not_null(nttb, true);
8091 zetas = must_be_not_null(zetas, true);
8092
8093 Node* result_start = array_element_address(result, intcon(0), T_INT);
8094 assert(result_start, "result is null");
8095 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8096 assert(ntta_start, "ntta is null");
8097 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8098 assert(nttb_start, "nttb is null");
8099 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8100 OptoRuntime::dilithiumNttMult_Type(),
8101 stubAddr, stubName, TypePtr::BOTTOM,
8102 result_start, ntta_start, nttb_start);
8103
8104 // return an int
8105 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8106 set_result(retvalue);
8107
8108 return true;
8109 }
8110
8111 //------------------------------inline_dilithiumMontMulByConstant
8112 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8113 address stubAddr;
8114 const char *stubName;
8115 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8116 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8117
8118 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8119 stubName = "dilithiumMontMulByConstant";
8120 if (!stubAddr) return false;
8121
8122 Node* coeffs = argument(0);
8123 Node* constant = argument(1);
8124
8125 coeffs = must_be_not_null(coeffs, true);
8126
8127 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8128 assert(coeffs_start, "coeffs is null");
8129 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8130 OptoRuntime::dilithiumMontMulByConstant_Type(),
8131 stubAddr, stubName, TypePtr::BOTTOM,
8132 coeffs_start, constant);
8133
8134 // return an int
8135 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8136 set_result(retvalue);
8137 return true;
8138 }
8139
8140
8141 //------------------------------inline_dilithiumDecomposePoly
8142 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8143 address stubAddr;
8144 const char *stubName;
8145 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8146 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8147
8148 stubAddr = StubRoutines::dilithiumDecomposePoly();
8149 stubName = "dilithiumDecomposePoly";
8150 if (!stubAddr) return false;
8151
8152 Node* input = argument(0);
8153 Node* lowPart = argument(1);
8154 Node* highPart = argument(2);
8155 Node* twoGamma2 = argument(3);
8156 Node* multiplier = argument(4);
8157
8158 input = must_be_not_null(input, true);
8159 lowPart = must_be_not_null(lowPart, true);
8160 highPart = must_be_not_null(highPart, true);
8161
8162 Node* input_start = array_element_address(input, intcon(0), T_INT);
8163 assert(input_start, "input is null");
8164 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8165 assert(lowPart_start, "lowPart is null");
8166 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8167 assert(highPart_start, "highPart is null");
8168
8169 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8170 OptoRuntime::dilithiumDecomposePoly_Type(),
8171 stubAddr, stubName, TypePtr::BOTTOM,
8172 input_start, lowPart_start, highPart_start,
8173 twoGamma2, multiplier);
8174
8175 // return an int
8176 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8177 set_result(retvalue);
8178 return true;
8179 }
8180
8181 bool LibraryCallKit::inline_base64_encodeBlock() {
8182 address stubAddr;
8183 const char *stubName;
8184 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8185 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8186 stubAddr = StubRoutines::base64_encodeBlock();
8187 stubName = "encodeBlock";
8188
8189 if (!stubAddr) return false;
8190 Node* base64obj = argument(0);
8191 Node* src = argument(1);
8192 Node* offset = argument(2);
8193 Node* len = argument(3);
8194 Node* dest = argument(4);
8195 Node* dp = argument(5);
8196 Node* isURL = argument(6);
8197
8198 src = must_be_not_null(src, true);
8199 dest = must_be_not_null(dest, true);
8200
8201 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8202 assert(src_start, "source array is null");
8203 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8204 assert(dest_start, "destination array is null");
8205
8206 Node* base64 = make_runtime_call(RC_LEAF,
8207 OptoRuntime::base64_encodeBlock_Type(),
8208 stubAddr, stubName, TypePtr::BOTTOM,
8209 src_start, offset, len, dest_start, dp, isURL);
8210 return true;
8211 }
8212
8213 bool LibraryCallKit::inline_base64_decodeBlock() {
8214 address stubAddr;
8215 const char *stubName;
8216 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8217 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8218 stubAddr = StubRoutines::base64_decodeBlock();
8219 stubName = "decodeBlock";
8220
8221 if (!stubAddr) return false;
8222 Node* base64obj = argument(0);
8223 Node* src = argument(1);
8224 Node* src_offset = argument(2);
8225 Node* len = argument(3);
8226 Node* dest = argument(4);
8227 Node* dest_offset = argument(5);
8228 Node* isURL = argument(6);
8229 Node* isMIME = argument(7);
8230
8231 src = must_be_not_null(src, true);
8232 dest = must_be_not_null(dest, true);
8233
8234 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8235 assert(src_start, "source array is null");
8236 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8237 assert(dest_start, "destination array is null");
8238
8239 Node* call = make_runtime_call(RC_LEAF,
8240 OptoRuntime::base64_decodeBlock_Type(),
8241 stubAddr, stubName, TypePtr::BOTTOM,
8242 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8243 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8244 set_result(result);
8245 return true;
8246 }
8247
8248 bool LibraryCallKit::inline_poly1305_processBlocks() {
8249 address stubAddr;
8250 const char *stubName;
8251 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8252 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8253 stubAddr = StubRoutines::poly1305_processBlocks();
8254 stubName = "poly1305_processBlocks";
8255
8256 if (!stubAddr) return false;
8257 null_check_receiver(); // null-check receiver
8258 if (stopped()) return true;
8259
8260 Node* input = argument(1);
8261 Node* input_offset = argument(2);
8262 Node* len = argument(3);
8263 Node* alimbs = argument(4);
8264 Node* rlimbs = argument(5);
8265
8266 input = must_be_not_null(input, true);
8267 alimbs = must_be_not_null(alimbs, true);
8268 rlimbs = must_be_not_null(rlimbs, true);
8269
8270 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8271 assert(input_start, "input array is null");
8272 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8273 assert(acc_start, "acc array is null");
8274 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8275 assert(r_start, "r array is null");
8276
8277 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8278 OptoRuntime::poly1305_processBlocks_Type(),
8279 stubAddr, stubName, TypePtr::BOTTOM,
8280 input_start, len, acc_start, r_start);
8281 return true;
8282 }
8283
8284 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8285 address stubAddr;
8286 const char *stubName;
8287 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8288 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8289 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8290 stubName = "intpoly_montgomeryMult_P256";
8291
8292 if (!stubAddr) return false;
8293 null_check_receiver(); // null-check receiver
8294 if (stopped()) return true;
8295
8296 Node* a = argument(1);
8297 Node* b = argument(2);
8298 Node* r = argument(3);
8299
8300 a = must_be_not_null(a, true);
8301 b = must_be_not_null(b, true);
8302 r = must_be_not_null(r, true);
8303
8304 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8305 assert(a_start, "a array is null");
8306 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8307 assert(b_start, "b array is null");
8308 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8309 assert(r_start, "r array is null");
8310
8311 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8312 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8313 stubAddr, stubName, TypePtr::BOTTOM,
8314 a_start, b_start, r_start);
8315 return true;
8316 }
8317
8318 bool LibraryCallKit::inline_intpoly_assign() {
8319 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8320 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8321 const char *stubName = "intpoly_assign";
8322 address stubAddr = StubRoutines::intpoly_assign();
8323 if (!stubAddr) return false;
8324
8325 Node* set = argument(0);
8326 Node* a = argument(1);
8327 Node* b = argument(2);
8328 Node* arr_length = load_array_length(a);
8329
8330 a = must_be_not_null(a, true);
8331 b = must_be_not_null(b, true);
8332
8333 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8334 assert(a_start, "a array is null");
8335 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8336 assert(b_start, "b array is null");
8337
8338 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8339 OptoRuntime::intpoly_assign_Type(),
8340 stubAddr, stubName, TypePtr::BOTTOM,
8341 set, a_start, b_start, arr_length);
8342 return true;
8343 }
8344
8345 //------------------------------inline_digestBase_implCompress-----------------------
8346 //
8347 // Calculate MD5 for single-block byte[] array.
8348 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8349 //
8350 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8351 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8352 //
8353 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8354 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8355 //
8356 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8357 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8358 //
8359 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8360 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8361 //
8362 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8363 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8364
8365 Node* digestBase_obj = argument(0);
8366 Node* src = argument(1); // type oop
8367 Node* ofs = argument(2); // type int
8368
8369 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8370 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8371 // failed array check
8372 return false;
8373 }
8374 // Figure out the size and type of the elements we will be copying.
8375 BasicType src_elem = src_type->elem()->array_element_basic_type();
8376 if (src_elem != T_BYTE) {
8377 return false;
8378 }
8379 // 'src_start' points to src array + offset
8380 src = must_be_not_null(src, true);
8381 Node* src_start = array_element_address(src, ofs, src_elem);
8382 Node* state = nullptr;
8383 Node* block_size = nullptr;
8384 address stubAddr;
8385 const char *stubName;
8386
8387 switch(id) {
8388 case vmIntrinsics::_md5_implCompress:
8389 assert(UseMD5Intrinsics, "need MD5 instruction support");
8390 state = get_state_from_digest_object(digestBase_obj, T_INT);
8391 stubAddr = StubRoutines::md5_implCompress();
8392 stubName = "md5_implCompress";
8393 break;
8394 case vmIntrinsics::_sha_implCompress:
8395 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8396 state = get_state_from_digest_object(digestBase_obj, T_INT);
8397 stubAddr = StubRoutines::sha1_implCompress();
8398 stubName = "sha1_implCompress";
8399 break;
8400 case vmIntrinsics::_sha2_implCompress:
8401 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8402 state = get_state_from_digest_object(digestBase_obj, T_INT);
8403 stubAddr = StubRoutines::sha256_implCompress();
8404 stubName = "sha256_implCompress";
8405 break;
8406 case vmIntrinsics::_sha5_implCompress:
8407 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8408 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8409 stubAddr = StubRoutines::sha512_implCompress();
8410 stubName = "sha512_implCompress";
8411 break;
8412 case vmIntrinsics::_sha3_implCompress:
8413 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8414 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8415 stubAddr = StubRoutines::sha3_implCompress();
8416 stubName = "sha3_implCompress";
8417 block_size = get_block_size_from_digest_object(digestBase_obj);
8418 if (block_size == nullptr) return false;
8419 break;
8420 default:
8421 fatal_unexpected_iid(id);
8422 return false;
8423 }
8424 if (state == nullptr) return false;
8425
8426 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8427 if (stubAddr == nullptr) return false;
8428
8429 // Call the stub.
8430 Node* call;
8431 if (block_size == nullptr) {
8432 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8433 stubAddr, stubName, TypePtr::BOTTOM,
8434 src_start, state);
8435 } else {
8436 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8437 stubAddr, stubName, TypePtr::BOTTOM,
8438 src_start, state, block_size);
8439 }
8440
8441 return true;
8442 }
8443
8444 //------------------------------inline_double_keccak
8445 bool LibraryCallKit::inline_double_keccak() {
8446 address stubAddr;
8447 const char *stubName;
8448 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8449 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8450
8451 stubAddr = StubRoutines::double_keccak();
8452 stubName = "double_keccak";
8453 if (!stubAddr) return false;
8454
8455 Node* status0 = argument(0);
8456 Node* status1 = argument(1);
8457
8458 status0 = must_be_not_null(status0, true);
8459 status1 = must_be_not_null(status1, true);
8460
8461 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8462 assert(status0_start, "status0 is null");
8463 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8464 assert(status1_start, "status1 is null");
8465 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8466 OptoRuntime::double_keccak_Type(),
8467 stubAddr, stubName, TypePtr::BOTTOM,
8468 status0_start, status1_start);
8469 // return an int
8470 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8471 set_result(retvalue);
8472 return true;
8473 }
8474
8475
8476 //------------------------------inline_digestBase_implCompressMB-----------------------
8477 //
8478 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8479 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8480 //
8481 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8482 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8483 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8484 assert((uint)predicate < 5, "sanity");
8485 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8486
8487 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8488 Node* src = argument(1); // byte[] array
8489 Node* ofs = argument(2); // type int
8490 Node* limit = argument(3); // type int
8491
8492 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8493 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8494 // failed array check
8495 return false;
8496 }
8497 // Figure out the size and type of the elements we will be copying.
8498 BasicType src_elem = src_type->elem()->array_element_basic_type();
8499 if (src_elem != T_BYTE) {
8500 return false;
8501 }
8502 // 'src_start' points to src array + offset
8503 src = must_be_not_null(src, false);
8504 Node* src_start = array_element_address(src, ofs, src_elem);
8505
8506 const char* klass_digestBase_name = nullptr;
8507 const char* stub_name = nullptr;
8508 address stub_addr = nullptr;
8509 BasicType elem_type = T_INT;
8510
8511 switch (predicate) {
8512 case 0:
8513 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8514 klass_digestBase_name = "sun/security/provider/MD5";
8515 stub_name = "md5_implCompressMB";
8516 stub_addr = StubRoutines::md5_implCompressMB();
8517 }
8518 break;
8519 case 1:
8520 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8521 klass_digestBase_name = "sun/security/provider/SHA";
8522 stub_name = "sha1_implCompressMB";
8523 stub_addr = StubRoutines::sha1_implCompressMB();
8524 }
8525 break;
8526 case 2:
8527 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8528 klass_digestBase_name = "sun/security/provider/SHA2";
8529 stub_name = "sha256_implCompressMB";
8530 stub_addr = StubRoutines::sha256_implCompressMB();
8531 }
8532 break;
8533 case 3:
8534 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8535 klass_digestBase_name = "sun/security/provider/SHA5";
8536 stub_name = "sha512_implCompressMB";
8537 stub_addr = StubRoutines::sha512_implCompressMB();
8538 elem_type = T_LONG;
8539 }
8540 break;
8541 case 4:
8542 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8543 klass_digestBase_name = "sun/security/provider/SHA3";
8544 stub_name = "sha3_implCompressMB";
8545 stub_addr = StubRoutines::sha3_implCompressMB();
8546 elem_type = T_LONG;
8547 }
8548 break;
8549 default:
8550 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8551 }
8552 if (klass_digestBase_name != nullptr) {
8553 assert(stub_addr != nullptr, "Stub is generated");
8554 if (stub_addr == nullptr) return false;
8555
8556 // get DigestBase klass to lookup for SHA klass
8557 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8558 assert(tinst != nullptr, "digestBase_obj is not instance???");
8559 assert(tinst->is_loaded(), "DigestBase is not loaded");
8560
8561 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8562 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8563 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8564 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8565 }
8566 return false;
8567 }
8568
8569 //------------------------------inline_digestBase_implCompressMB-----------------------
8570 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8571 BasicType elem_type, address stubAddr, const char *stubName,
8572 Node* src_start, Node* ofs, Node* limit) {
8573 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8574 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8575 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8576 digest_obj = _gvn.transform(digest_obj);
8577
8578 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8579 if (state == nullptr) return false;
8580
8581 Node* block_size = nullptr;
8582 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8583 block_size = get_block_size_from_digest_object(digest_obj);
8584 if (block_size == nullptr) return false;
8585 }
8586
8587 // Call the stub.
8588 Node* call;
8589 if (block_size == nullptr) {
8590 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8591 OptoRuntime::digestBase_implCompressMB_Type(false),
8592 stubAddr, stubName, TypePtr::BOTTOM,
8593 src_start, state, ofs, limit);
8594 } else {
8595 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8596 OptoRuntime::digestBase_implCompressMB_Type(true),
8597 stubAddr, stubName, TypePtr::BOTTOM,
8598 src_start, state, block_size, ofs, limit);
8599 }
8600
8601 // return ofs (int)
8602 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8603 set_result(result);
8604
8605 return true;
8606 }
8607
8608 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8609 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8610 assert(UseAES, "need AES instruction support");
8611 address stubAddr = nullptr;
8612 const char *stubName = nullptr;
8613 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8614 stubName = "galoisCounterMode_AESCrypt";
8615
8616 if (stubAddr == nullptr) return false;
8617
8618 Node* in = argument(0);
8619 Node* inOfs = argument(1);
8620 Node* len = argument(2);
8621 Node* ct = argument(3);
8622 Node* ctOfs = argument(4);
8623 Node* out = argument(5);
8624 Node* outOfs = argument(6);
8625 Node* gctr_object = argument(7);
8626 Node* ghash_object = argument(8);
8627
8628 // (1) in, ct and out are arrays.
8629 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8630 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8631 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8632 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8633 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8634 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8635
8636 // checks are the responsibility of the caller
8637 Node* in_start = in;
8638 Node* ct_start = ct;
8639 Node* out_start = out;
8640 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8641 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8642 in_start = array_element_address(in, inOfs, T_BYTE);
8643 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8644 out_start = array_element_address(out, outOfs, T_BYTE);
8645 }
8646
8647 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8648 // (because of the predicated logic executed earlier).
8649 // so we cast it here safely.
8650 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8651 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8652 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8653 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8654 Node* state = load_field_from_object(ghash_object, "state", "[J");
8655
8656 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8657 return false;
8658 }
8659 // cast it to what we know it will be at runtime
8660 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8661 assert(tinst != nullptr, "GCTR obj is null");
8662 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8663 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8664 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8665 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8666 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8667 const TypeOopPtr* xtype = aklass->as_instance_type();
8668 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8669 aescrypt_object = _gvn.transform(aescrypt_object);
8670 // we need to get the start of the aescrypt_object's expanded key array
8671 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8672 if (k_start == nullptr) return false;
8673 // similarly, get the start address of the r vector
8674 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8675 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8676 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8677
8678
8679 // Call the stub, passing params
8680 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8681 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8682 stubAddr, stubName, TypePtr::BOTTOM,
8683 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8684
8685 // return cipher length (int)
8686 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8687 set_result(retvalue);
8688
8689 return true;
8690 }
8691
8692 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8693 // Return node representing slow path of predicate check.
8694 // the pseudo code we want to emulate with this predicate is:
8695 // for encryption:
8696 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8697 // for decryption:
8698 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8699 // note cipher==plain is more conservative than the original java code but that's OK
8700 //
8701
8702 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8703 // The receiver was checked for null already.
8704 Node* objGCTR = argument(7);
8705 // Load embeddedCipher field of GCTR object.
8706 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8707 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8708
8709 // get AESCrypt klass for instanceOf check
8710 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8711 // will have same classloader as CipherBlockChaining object
8712 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8713 assert(tinst != nullptr, "GCTR obj is null");
8714 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8715
8716 // we want to do an instanceof comparison against the AESCrypt class
8717 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8718 if (!klass_AESCrypt->is_loaded()) {
8719 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8720 Node* ctrl = control();
8721 set_control(top()); // no regular fast path
8722 return ctrl;
8723 }
8724
8725 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8726 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8727 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8728 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8729 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8730
8731 return instof_false; // even if it is null
8732 }
8733
8734 //------------------------------get_state_from_digest_object-----------------------
8735 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8736 const char* state_type;
8737 switch (elem_type) {
8738 case T_BYTE: state_type = "[B"; break;
8739 case T_INT: state_type = "[I"; break;
8740 case T_LONG: state_type = "[J"; break;
8741 default: ShouldNotReachHere();
8742 }
8743 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8744 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8745 if (digest_state == nullptr) return (Node *) nullptr;
8746
8747 // now have the array, need to get the start address of the state array
8748 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8749 return state;
8750 }
8751
8752 //------------------------------get_block_size_from_sha3_object----------------------------------
8753 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8754 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8755 assert (block_size != nullptr, "sanity");
8756 return block_size;
8757 }
8758
8759 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8760 // Return node representing slow path of predicate check.
8761 // the pseudo code we want to emulate with this predicate is:
8762 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8763 //
8764 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8765 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8766 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8767 assert((uint)predicate < 5, "sanity");
8768
8769 // The receiver was checked for null already.
8770 Node* digestBaseObj = argument(0);
8771
8772 // get DigestBase klass for instanceOf check
8773 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8774 assert(tinst != nullptr, "digestBaseObj is null");
8775 assert(tinst->is_loaded(), "DigestBase is not loaded");
8776
8777 const char* klass_name = nullptr;
8778 switch (predicate) {
8779 case 0:
8780 if (UseMD5Intrinsics) {
8781 // we want to do an instanceof comparison against the MD5 class
8782 klass_name = "sun/security/provider/MD5";
8783 }
8784 break;
8785 case 1:
8786 if (UseSHA1Intrinsics) {
8787 // we want to do an instanceof comparison against the SHA class
8788 klass_name = "sun/security/provider/SHA";
8789 }
8790 break;
8791 case 2:
8792 if (UseSHA256Intrinsics) {
8793 // we want to do an instanceof comparison against the SHA2 class
8794 klass_name = "sun/security/provider/SHA2";
8795 }
8796 break;
8797 case 3:
8798 if (UseSHA512Intrinsics) {
8799 // we want to do an instanceof comparison against the SHA5 class
8800 klass_name = "sun/security/provider/SHA5";
8801 }
8802 break;
8803 case 4:
8804 if (UseSHA3Intrinsics) {
8805 // we want to do an instanceof comparison against the SHA3 class
8806 klass_name = "sun/security/provider/SHA3";
8807 }
8808 break;
8809 default:
8810 fatal("unknown SHA intrinsic predicate: %d", predicate);
8811 }
8812
8813 ciKlass* klass = nullptr;
8814 if (klass_name != nullptr) {
8815 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8816 }
8817 if ((klass == nullptr) || !klass->is_loaded()) {
8818 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8819 Node* ctrl = control();
8820 set_control(top()); // no intrinsic path
8821 return ctrl;
8822 }
8823 ciInstanceKlass* instklass = klass->as_instance_klass();
8824
8825 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8826 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8827 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8828 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8829
8830 return instof_false; // even if it is null
8831 }
8832
8833 //-------------inline_fma-----------------------------------
8834 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8835 Node *a = nullptr;
8836 Node *b = nullptr;
8837 Node *c = nullptr;
8838 Node* result = nullptr;
8839 switch (id) {
8840 case vmIntrinsics::_fmaD:
8841 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8842 // no receiver since it is static method
8843 a = argument(0);
8844 b = argument(2);
8845 c = argument(4);
8846 result = _gvn.transform(new FmaDNode(a, b, c));
8847 break;
8848 case vmIntrinsics::_fmaF:
8849 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8850 a = argument(0);
8851 b = argument(1);
8852 c = argument(2);
8853 result = _gvn.transform(new FmaFNode(a, b, c));
8854 break;
8855 default:
8856 fatal_unexpected_iid(id); break;
8857 }
8858 set_result(result);
8859 return true;
8860 }
8861
8862 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8863 // argument(0) is receiver
8864 Node* codePoint = argument(1);
8865 Node* n = nullptr;
8866
8867 switch (id) {
8868 case vmIntrinsics::_isDigit :
8869 n = new DigitNode(control(), codePoint);
8870 break;
8871 case vmIntrinsics::_isLowerCase :
8872 n = new LowerCaseNode(control(), codePoint);
8873 break;
8874 case vmIntrinsics::_isUpperCase :
8875 n = new UpperCaseNode(control(), codePoint);
8876 break;
8877 case vmIntrinsics::_isWhitespace :
8878 n = new WhitespaceNode(control(), codePoint);
8879 break;
8880 default:
8881 fatal_unexpected_iid(id);
8882 }
8883
8884 set_result(_gvn.transform(n));
8885 return true;
8886 }
8887
8888 bool LibraryCallKit::inline_profileBoolean() {
8889 Node* counts = argument(1);
8890 const TypeAryPtr* ary = nullptr;
8891 ciArray* aobj = nullptr;
8892 if (counts->is_Con()
8893 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8894 && (aobj = ary->const_oop()->as_array()) != nullptr
8895 && (aobj->length() == 2)) {
8896 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8897 jint false_cnt = aobj->element_value(0).as_int();
8898 jint true_cnt = aobj->element_value(1).as_int();
8899
8900 if (C->log() != nullptr) {
8901 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8902 false_cnt, true_cnt);
8903 }
8904
8905 if (false_cnt + true_cnt == 0) {
8906 // According to profile, never executed.
8907 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8908 Deoptimization::Action_reinterpret);
8909 return true;
8910 }
8911
8912 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8913 // is a number of each value occurrences.
8914 Node* result = argument(0);
8915 if (false_cnt == 0 || true_cnt == 0) {
8916 // According to profile, one value has been never seen.
8917 int expected_val = (false_cnt == 0) ? 1 : 0;
8918
8919 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8920 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8921
8922 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8923 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8924 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8925
8926 { // Slow path: uncommon trap for never seen value and then reexecute
8927 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8928 // the value has been seen at least once.
8929 PreserveJVMState pjvms(this);
8930 PreserveReexecuteState preexecs(this);
8931 jvms()->set_should_reexecute(true);
8932
8933 set_control(slow_path);
8934 set_i_o(i_o());
8935
8936 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8937 Deoptimization::Action_reinterpret);
8938 }
8939 // The guard for never seen value enables sharpening of the result and
8940 // returning a constant. It allows to eliminate branches on the same value
8941 // later on.
8942 set_control(fast_path);
8943 result = intcon(expected_val);
8944 }
8945 // Stop profiling.
8946 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8947 // By replacing method body with profile data (represented as ProfileBooleanNode
8948 // on IR level) we effectively disable profiling.
8949 // It enables full speed execution once optimized code is generated.
8950 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8951 C->record_for_igvn(profile);
8952 set_result(profile);
8953 return true;
8954 } else {
8955 // Continue profiling.
8956 // Profile data isn't available at the moment. So, execute method's bytecode version.
8957 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8958 // is compiled and counters aren't available since corresponding MethodHandle
8959 // isn't a compile-time constant.
8960 return false;
8961 }
8962 }
8963
8964 bool LibraryCallKit::inline_isCompileConstant() {
8965 Node* n = argument(0);
8966 set_result(n->is_Con() ? intcon(1) : intcon(0));
8967 return true;
8968 }
8969
8970 //------------------------------- inline_getObjectSize --------------------------------------
8971 //
8972 // Calculate the runtime size of the object/array.
8973 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8974 //
8975 bool LibraryCallKit::inline_getObjectSize() {
8976 Node* obj = argument(3);
8977 Node* klass_node = load_object_klass(obj);
8978
8979 jint layout_con = Klass::_lh_neutral_value;
8980 Node* layout_val = get_layout_helper(klass_node, layout_con);
8981 int layout_is_con = (layout_val == nullptr);
8982
8983 if (layout_is_con) {
8984 // Layout helper is constant, can figure out things at compile time.
8985
8986 if (Klass::layout_helper_is_instance(layout_con)) {
8987 // Instance case: layout_con contains the size itself.
8988 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8989 set_result(size);
8990 } else {
8991 // Array case: size is round(header + element_size*arraylength).
8992 // Since arraylength is different for every array instance, we have to
8993 // compute the whole thing at runtime.
8994
8995 Node* arr_length = load_array_length(obj);
8996
8997 int round_mask = MinObjAlignmentInBytes - 1;
8998 int hsize = Klass::layout_helper_header_size(layout_con);
8999 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9000
9001 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9002 round_mask = 0; // strength-reduce it if it goes away completely
9003 }
9004 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9005 Node* header_size = intcon(hsize + round_mask);
9006
9007 Node* lengthx = ConvI2X(arr_length);
9008 Node* headerx = ConvI2X(header_size);
9009
9010 Node* abody = lengthx;
9011 if (eshift != 0) {
9012 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9013 }
9014 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9015 if (round_mask != 0) {
9016 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9017 }
9018 size = ConvX2L(size);
9019 set_result(size);
9020 }
9021 } else {
9022 // Layout helper is not constant, need to test for array-ness at runtime.
9023
9024 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9025 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9026 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9027 record_for_igvn(result_reg);
9028
9029 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9030 if (array_ctl != nullptr) {
9031 // Array case: size is round(header + element_size*arraylength).
9032 // Since arraylength is different for every array instance, we have to
9033 // compute the whole thing at runtime.
9034
9035 PreserveJVMState pjvms(this);
9036 set_control(array_ctl);
9037 Node* arr_length = load_array_length(obj);
9038
9039 int round_mask = MinObjAlignmentInBytes - 1;
9040 Node* mask = intcon(round_mask);
9041
9042 Node* hss = intcon(Klass::_lh_header_size_shift);
9043 Node* hsm = intcon(Klass::_lh_header_size_mask);
9044 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9045 header_size = _gvn.transform(new AndINode(header_size, hsm));
9046 header_size = _gvn.transform(new AddINode(header_size, mask));
9047
9048 // There is no need to mask or shift this value.
9049 // The semantics of LShiftINode include an implicit mask to 0x1F.
9050 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9051 Node* elem_shift = layout_val;
9052
9053 Node* lengthx = ConvI2X(arr_length);
9054 Node* headerx = ConvI2X(header_size);
9055
9056 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9057 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9058 if (round_mask != 0) {
9059 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9060 }
9061 size = ConvX2L(size);
9062
9063 result_reg->init_req(_array_path, control());
9064 result_val->init_req(_array_path, size);
9065 }
9066
9067 if (!stopped()) {
9068 // Instance case: the layout helper gives us instance size almost directly,
9069 // but we need to mask out the _lh_instance_slow_path_bit.
9070 Node* size = ConvI2X(layout_val);
9071 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9072 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9073 size = _gvn.transform(new AndXNode(size, mask));
9074 size = ConvX2L(size);
9075
9076 result_reg->init_req(_instance_path, control());
9077 result_val->init_req(_instance_path, size);
9078 }
9079
9080 set_result(result_reg, result_val);
9081 }
9082
9083 return true;
9084 }
9085
9086 //------------------------------- inline_blackhole --------------------------------------
9087 //
9088 // Make sure all arguments to this node are alive.
9089 // This matches methods that were requested to be blackholed through compile commands.
9090 //
9091 bool LibraryCallKit::inline_blackhole() {
9092 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9093 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9094 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9095
9096 // Blackhole node pinches only the control, not memory. This allows
9097 // the blackhole to be pinned in the loop that computes blackholed
9098 // values, but have no other side effects, like breaking the optimizations
9099 // across the blackhole.
9100
9101 Node* bh = _gvn.transform(new BlackholeNode(control()));
9102 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9103
9104 // Bind call arguments as blackhole arguments to keep them alive
9105 uint nargs = callee()->arg_size();
9106 for (uint i = 0; i < nargs; i++) {
9107 bh->add_req(argument(i));
9108 }
9109
9110 return true;
9111 }
9112
9113 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9114 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9115 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9116 return nullptr; // box klass is not Float16
9117 }
9118
9119 // Null check; get notnull casted pointer
9120 Node* null_ctl = top();
9121 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9122 // If not_null_box is dead, only null-path is taken
9123 if (stopped()) {
9124 set_control(null_ctl);
9125 return nullptr;
9126 }
9127 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9128 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9129 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9130 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9131 }
9132
9133 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9134 PreserveReexecuteState preexecs(this);
9135 jvms()->set_should_reexecute(true);
9136
9137 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9138 Node* klass_node = makecon(klass_type);
9139 Node* box = new_instance(klass_node);
9140
9141 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9142 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9143
9144 Node* field_store = _gvn.transform(access_store_at(box,
9145 value_field,
9146 value_adr_type,
9147 value,
9148 TypeInt::SHORT,
9149 T_SHORT,
9150 IN_HEAP));
9151 set_memory(field_store, value_adr_type);
9152 return box;
9153 }
9154
9155 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9156 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9157 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9158 return false;
9159 }
9160
9161 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9162 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9163 return false;
9164 }
9165
9166 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9167 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9168 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9169 ciSymbols::short_signature(),
9170 false);
9171 assert(field != nullptr, "");
9172
9173 // Transformed nodes
9174 Node* fld1 = nullptr;
9175 Node* fld2 = nullptr;
9176 Node* fld3 = nullptr;
9177 switch(num_args) {
9178 case 3:
9179 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9180 if (fld3 == nullptr) {
9181 return false;
9182 }
9183 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9184 // fall-through
9185 case 2:
9186 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9187 if (fld2 == nullptr) {
9188 return false;
9189 }
9190 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9191 // fall-through
9192 case 1:
9193 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9194 if (fld1 == nullptr) {
9195 return false;
9196 }
9197 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9198 break;
9199 default: fatal("Unsupported number of arguments %d", num_args);
9200 }
9201
9202 Node* result = nullptr;
9203 switch (id) {
9204 // Unary operations
9205 case vmIntrinsics::_sqrt_float16:
9206 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9207 break;
9208 // Ternary operations
9209 case vmIntrinsics::_fma_float16:
9210 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9211 break;
9212 default:
9213 fatal_unexpected_iid(id);
9214 break;
9215 }
9216 result = _gvn.transform(new ReinterpretHF2SNode(result));
9217 set_result(box_fp16_value(float16_box_type, field, result));
9218 return true;
9219 }
9220