1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciFlatArrayKlass.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "ci/ciSymbols.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "opto/vectornode.hpp"
55 #include "prims/jvmtiExport.hpp"
56 #include "prims/jvmtiThreadState.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/jniHandles.inline.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 if (jvms->has_method()) {
140 // Not a root compile.
141 const char* msg;
142 if (callee->intrinsic_candidate()) {
143 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
144 } else {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
146 : "failed to inline (intrinsic), method not annotated";
147 }
148 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
149 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
150 } else {
151 // Root compile
152 ResourceMark rm;
153 stringStream msg_stream;
154 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
155 vmIntrinsics::name_at(intrinsic_id()),
156 is_virtual() ? " (virtual)" : "", bci);
157 const char *msg = msg_stream.freeze();
158 log_debug(jit, inlining)("%s", msg);
159 if (C->print_intrinsics() || C->print_inlining()) {
160 tty->print("%s", msg);
161 }
162 }
163 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
164
165 return nullptr;
166 }
167
168 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
169 LibraryCallKit kit(jvms, this);
170 Compile* C = kit.C;
171 int nodes = C->unique();
172 _last_predicate = predicate;
173 #ifndef PRODUCT
174 assert(is_predicated() && predicate < predicates_count(), "sanity");
175 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
176 char buf[1000];
177 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
178 tty->print_cr("Predicate for intrinsic %s", str);
179 }
180 #endif
181 ciMethod* callee = kit.callee();
182 const int bci = kit.bci();
183
184 Node* slow_ctl = kit.try_to_predicate(predicate);
185 if (!kit.failing()) {
186 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
187 : "(intrinsic, predicate)";
188 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
189 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
190
191 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
192 if (C->log()) {
193 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
194 vmIntrinsics::name_at(intrinsic_id()),
195 (is_virtual() ? " virtual='1'" : ""),
196 C->unique() - nodes);
197 }
198 return slow_ctl; // Could be null if the check folds.
199 }
200
201 // The intrinsic bailed out
202 if (jvms->has_method()) {
203 // Not a root compile.
204 const char* msg = "failed to generate predicate for intrinsic";
205 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
206 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
207 } else {
208 // Root compile
209 ResourceMark rm;
210 stringStream msg_stream;
211 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
212 vmIntrinsics::name_at(intrinsic_id()),
213 is_virtual() ? " (virtual)" : "", bci);
214 const char *msg = msg_stream.freeze();
215 log_debug(jit, inlining)("%s", msg);
216 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
217 }
218 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
219 return nullptr;
220 }
221
222 bool LibraryCallKit::try_to_inline(int predicate) {
223 // Handle symbolic names for otherwise undistinguished boolean switches:
224 const bool is_store = true;
225 const bool is_compress = true;
226 const bool is_static = true;
227 const bool is_volatile = true;
228
229 if (!jvms()->has_method()) {
230 // Root JVMState has a null method.
231 assert(map()->memory()->Opcode() == Op_Parm, "");
232 // Insert the memory aliasing node
233 set_all_memory(reset_memory());
234 }
235 assert(merged_memory(), "");
236
237 switch (intrinsic_id()) {
238 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
239 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
240 case vmIntrinsics::_getClass: return inline_native_getClass();
241
242 case vmIntrinsics::_ceil:
243 case vmIntrinsics::_floor:
244 case vmIntrinsics::_rint:
245 case vmIntrinsics::_dsin:
246 case vmIntrinsics::_dcos:
247 case vmIntrinsics::_dtan:
248 case vmIntrinsics::_dtanh:
249 case vmIntrinsics::_dabs:
250 case vmIntrinsics::_fabs:
251 case vmIntrinsics::_iabs:
252 case vmIntrinsics::_labs:
253 case vmIntrinsics::_datan2:
254 case vmIntrinsics::_dsqrt:
255 case vmIntrinsics::_dsqrt_strict:
256 case vmIntrinsics::_dexp:
257 case vmIntrinsics::_dlog:
258 case vmIntrinsics::_dlog10:
259 case vmIntrinsics::_dpow:
260 case vmIntrinsics::_dcopySign:
261 case vmIntrinsics::_fcopySign:
262 case vmIntrinsics::_dsignum:
263 case vmIntrinsics::_roundF:
264 case vmIntrinsics::_roundD:
265 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
266
267 case vmIntrinsics::_notify:
268 case vmIntrinsics::_notifyAll:
269 return inline_notify(intrinsic_id());
270
271 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
272 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
273 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
274 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
275 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
276 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
277 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
278 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
279 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
280 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
281 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
282 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
283 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
284 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
285
286 case vmIntrinsics::_arraycopy: return inline_arraycopy();
287
288 case vmIntrinsics::_arraySort: return inline_array_sort();
289 case vmIntrinsics::_arrayPartition: return inline_array_partition();
290
291 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
292 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
293 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
294 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
295
296 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
297 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
298 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
299 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
303 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
304
305 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
306
307 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
308
309 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
310 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
311 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
312 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
313
314 case vmIntrinsics::_compressStringC:
315 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
316 case vmIntrinsics::_inflateStringC:
317 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
318
319 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
320 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
321 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
322 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
323 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
324 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
325 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
326 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
327 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
328 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
329 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
330 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
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 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
342
343 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
344 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
345 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
346 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
347 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
348 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
349 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
350 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
351 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
352
353 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
354 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
355 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
356 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
357 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
358 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
359 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
360 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
361 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
362
363 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
374 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
375 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
376 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
377 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
378 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
379 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
380 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
381 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
382
383 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
384 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
385 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
386 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
387 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
388 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
389 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
390 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
391 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
392
393 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
394 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
395 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
396 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
397 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
398 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
399 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
400 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
401 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
402
403 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
404 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
405 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
406 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
407 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
408 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
409 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
410 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
411 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
412
413 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
417 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
418
419 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
420 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
421 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
422 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
423 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
439
440 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
452 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
453 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
454 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
455
456 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
459 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
460
461 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
465 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
466
467 case vmIntrinsics::_loadFence:
468 case vmIntrinsics::_storeFence:
469 case vmIntrinsics::_storeStoreFence:
470 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
471
472 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
473
474 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
475 case vmIntrinsics::_currentThread: return inline_native_currentThread();
476 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
477
478 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
479 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
480
481 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
482 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
483
484 #if INCLUDE_JVMTI
485 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
486 "notifyJvmtiStart", true, false);
487 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
488 "notifyJvmtiEnd", false, true);
489 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
490 "notifyJvmtiMount", false, false);
491 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
492 "notifyJvmtiUnmount", false, false);
493 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
494 #endif
495
496 #ifdef JFR_HAVE_INTRINSICS
497 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
498 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
499 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
500 #endif
501 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
502 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
503 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
504 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
505 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
506 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
507 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
508 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
509 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
510 case vmIntrinsics::_getLength: return inline_native_getLength();
511 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
512 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
513 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
514 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
515 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
516 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
517 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
518
519 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
520 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
521 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
522 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
523 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
524
525 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
526
527 case vmIntrinsics::_isInstance:
528 case vmIntrinsics::_isHidden:
529 case vmIntrinsics::_getSuperclass:
530 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
531
532 case vmIntrinsics::_floatToRawIntBits:
533 case vmIntrinsics::_floatToIntBits:
534 case vmIntrinsics::_intBitsToFloat:
535 case vmIntrinsics::_doubleToRawLongBits:
536 case vmIntrinsics::_doubleToLongBits:
537 case vmIntrinsics::_longBitsToDouble:
538 case vmIntrinsics::_floatToFloat16:
539 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
540 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
541 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
542 case vmIntrinsics::_floatIsFinite:
543 case vmIntrinsics::_floatIsInfinite:
544 case vmIntrinsics::_doubleIsFinite:
545 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
546
547 case vmIntrinsics::_numberOfLeadingZeros_i:
548 case vmIntrinsics::_numberOfLeadingZeros_l:
549 case vmIntrinsics::_numberOfTrailingZeros_i:
550 case vmIntrinsics::_numberOfTrailingZeros_l:
551 case vmIntrinsics::_bitCount_i:
552 case vmIntrinsics::_bitCount_l:
553 case vmIntrinsics::_reverse_i:
554 case vmIntrinsics::_reverse_l:
555 case vmIntrinsics::_reverseBytes_i:
556 case vmIntrinsics::_reverseBytes_l:
557 case vmIntrinsics::_reverseBytes_s:
558 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
559
560 case vmIntrinsics::_compress_i:
561 case vmIntrinsics::_compress_l:
562 case vmIntrinsics::_expand_i:
563 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
564
565 case vmIntrinsics::_compareUnsigned_i:
566 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
567
568 case vmIntrinsics::_divideUnsigned_i:
569 case vmIntrinsics::_divideUnsigned_l:
570 case vmIntrinsics::_remainderUnsigned_i:
571 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
572
573 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
574
575 case vmIntrinsics::_Reference_get: return inline_reference_get();
576 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
577 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
578 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
579 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
580
581 case vmIntrinsics::_Class_cast: return inline_Class_cast();
582
583 case vmIntrinsics::_aescrypt_encryptBlock:
584 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
585
586 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
587 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
588 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
589
590 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
591 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
592 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
593
594 case vmIntrinsics::_counterMode_AESCrypt:
595 return inline_counterMode_AESCrypt(intrinsic_id());
596
597 case vmIntrinsics::_galoisCounterMode_AESCrypt:
598 return inline_galoisCounterMode_AESCrypt();
599
600 case vmIntrinsics::_md5_implCompress:
601 case vmIntrinsics::_sha_implCompress:
602 case vmIntrinsics::_sha2_implCompress:
603 case vmIntrinsics::_sha5_implCompress:
604 case vmIntrinsics::_sha3_implCompress:
605 return inline_digestBase_implCompress(intrinsic_id());
606 case vmIntrinsics::_double_keccak:
607 return inline_double_keccak();
608
609 case vmIntrinsics::_digestBase_implCompressMB:
610 return inline_digestBase_implCompressMB(predicate);
611
612 case vmIntrinsics::_multiplyToLen:
613 return inline_multiplyToLen();
614
615 case vmIntrinsics::_squareToLen:
616 return inline_squareToLen();
617
618 case vmIntrinsics::_mulAdd:
619 return inline_mulAdd();
620
621 case vmIntrinsics::_montgomeryMultiply:
622 return inline_montgomeryMultiply();
623 case vmIntrinsics::_montgomerySquare:
624 return inline_montgomerySquare();
625
626 case vmIntrinsics::_bigIntegerRightShiftWorker:
627 return inline_bigIntegerShift(true);
628 case vmIntrinsics::_bigIntegerLeftShiftWorker:
629 return inline_bigIntegerShift(false);
630
631 case vmIntrinsics::_vectorizedMismatch:
632 return inline_vectorizedMismatch();
633
634 case vmIntrinsics::_ghash_processBlocks:
635 return inline_ghash_processBlocks();
636 case vmIntrinsics::_chacha20Block:
637 return inline_chacha20Block();
638 case vmIntrinsics::_kyberNtt:
639 return inline_kyberNtt();
640 case vmIntrinsics::_kyberInverseNtt:
641 return inline_kyberInverseNtt();
642 case vmIntrinsics::_kyberNttMult:
643 return inline_kyberNttMult();
644 case vmIntrinsics::_kyberAddPoly_2:
645 return inline_kyberAddPoly_2();
646 case vmIntrinsics::_kyberAddPoly_3:
647 return inline_kyberAddPoly_3();
648 case vmIntrinsics::_kyber12To16:
649 return inline_kyber12To16();
650 case vmIntrinsics::_kyberBarrettReduce:
651 return inline_kyberBarrettReduce();
652 case vmIntrinsics::_dilithiumAlmostNtt:
653 return inline_dilithiumAlmostNtt();
654 case vmIntrinsics::_dilithiumAlmostInverseNtt:
655 return inline_dilithiumAlmostInverseNtt();
656 case vmIntrinsics::_dilithiumNttMult:
657 return inline_dilithiumNttMult();
658 case vmIntrinsics::_dilithiumMontMulByConstant:
659 return inline_dilithiumMontMulByConstant();
660 case vmIntrinsics::_dilithiumDecomposePoly:
661 return inline_dilithiumDecomposePoly();
662 case vmIntrinsics::_base64_encodeBlock:
663 return inline_base64_encodeBlock();
664 case vmIntrinsics::_base64_decodeBlock:
665 return inline_base64_decodeBlock();
666 case vmIntrinsics::_poly1305_processBlocks:
667 return inline_poly1305_processBlocks();
668 case vmIntrinsics::_intpoly_montgomeryMult_P256:
669 return inline_intpoly_montgomeryMult_P256();
670 case vmIntrinsics::_intpoly_assign:
671 return inline_intpoly_assign();
672 case vmIntrinsics::_encodeISOArray:
673 case vmIntrinsics::_encodeByteISOArray:
674 return inline_encodeISOArray(false);
675 case vmIntrinsics::_encodeAsciiArray:
676 return inline_encodeISOArray(true);
677
678 case vmIntrinsics::_updateCRC32:
679 return inline_updateCRC32();
680 case vmIntrinsics::_updateBytesCRC32:
681 return inline_updateBytesCRC32();
682 case vmIntrinsics::_updateByteBufferCRC32:
683 return inline_updateByteBufferCRC32();
684
685 case vmIntrinsics::_updateBytesCRC32C:
686 return inline_updateBytesCRC32C();
687 case vmIntrinsics::_updateDirectByteBufferCRC32C:
688 return inline_updateDirectByteBufferCRC32C();
689
690 case vmIntrinsics::_updateBytesAdler32:
691 return inline_updateBytesAdler32();
692 case vmIntrinsics::_updateByteBufferAdler32:
693 return inline_updateByteBufferAdler32();
694
695 case vmIntrinsics::_profileBoolean:
696 return inline_profileBoolean();
697 case vmIntrinsics::_isCompileConstant:
698 return inline_isCompileConstant();
699
700 case vmIntrinsics::_countPositives:
701 return inline_countPositives();
702
703 case vmIntrinsics::_fmaD:
704 case vmIntrinsics::_fmaF:
705 return inline_fma(intrinsic_id());
706
707 case vmIntrinsics::_isDigit:
708 case vmIntrinsics::_isLowerCase:
709 case vmIntrinsics::_isUpperCase:
710 case vmIntrinsics::_isWhitespace:
711 return inline_character_compare(intrinsic_id());
712
713 case vmIntrinsics::_min:
714 case vmIntrinsics::_max:
715 case vmIntrinsics::_min_strict:
716 case vmIntrinsics::_max_strict:
717 case vmIntrinsics::_minL:
718 case vmIntrinsics::_maxL:
719 case vmIntrinsics::_minF:
720 case vmIntrinsics::_maxF:
721 case vmIntrinsics::_minD:
722 case vmIntrinsics::_maxD:
723 case vmIntrinsics::_minF_strict:
724 case vmIntrinsics::_maxF_strict:
725 case vmIntrinsics::_minD_strict:
726 case vmIntrinsics::_maxD_strict:
727 return inline_min_max(intrinsic_id());
728
729 case vmIntrinsics::_VectorUnaryOp:
730 return inline_vector_nary_operation(1);
731 case vmIntrinsics::_VectorBinaryOp:
732 return inline_vector_nary_operation(2);
733 case vmIntrinsics::_VectorTernaryOp:
734 return inline_vector_nary_operation(3);
735 case vmIntrinsics::_VectorFromBitsCoerced:
736 return inline_vector_frombits_coerced();
737 case vmIntrinsics::_VectorMaskOp:
738 return inline_vector_mask_operation();
739 case vmIntrinsics::_VectorLoadOp:
740 return inline_vector_mem_operation(/*is_store=*/false);
741 case vmIntrinsics::_VectorLoadMaskedOp:
742 return inline_vector_mem_masked_operation(/*is_store*/false);
743 case vmIntrinsics::_VectorStoreOp:
744 return inline_vector_mem_operation(/*is_store=*/true);
745 case vmIntrinsics::_VectorStoreMaskedOp:
746 return inline_vector_mem_masked_operation(/*is_store=*/true);
747 case vmIntrinsics::_VectorGatherOp:
748 return inline_vector_gather_scatter(/*is_scatter*/ false);
749 case vmIntrinsics::_VectorScatterOp:
750 return inline_vector_gather_scatter(/*is_scatter*/ true);
751 case vmIntrinsics::_VectorReductionCoerced:
752 return inline_vector_reduction();
753 case vmIntrinsics::_VectorTest:
754 return inline_vector_test();
755 case vmIntrinsics::_VectorBlend:
756 return inline_vector_blend();
757 case vmIntrinsics::_VectorRearrange:
758 return inline_vector_rearrange();
759 case vmIntrinsics::_VectorSelectFrom:
760 return inline_vector_select_from();
761 case vmIntrinsics::_VectorCompare:
762 return inline_vector_compare();
763 case vmIntrinsics::_VectorBroadcastInt:
764 return inline_vector_broadcast_int();
765 case vmIntrinsics::_VectorConvert:
766 return inline_vector_convert();
767 case vmIntrinsics::_VectorInsert:
768 return inline_vector_insert();
769 case vmIntrinsics::_VectorExtract:
770 return inline_vector_extract();
771 case vmIntrinsics::_VectorCompressExpand:
772 return inline_vector_compress_expand();
773 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
774 return inline_vector_select_from_two_vectors();
775 case vmIntrinsics::_IndexVector:
776 return inline_index_vector();
777 case vmIntrinsics::_IndexPartiallyInUpperRange:
778 return inline_index_partially_in_upper_range();
779
780 case vmIntrinsics::_getObjectSize:
781 return inline_getObjectSize();
782
783 case vmIntrinsics::_blackhole:
784 return inline_blackhole();
785
786 default:
787 // If you get here, it may be that someone has added a new intrinsic
788 // to the list in vmIntrinsics.hpp without implementing it here.
789 #ifndef PRODUCT
790 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
791 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
792 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
793 }
794 #endif
795 return false;
796 }
797 }
798
799 Node* LibraryCallKit::try_to_predicate(int predicate) {
800 if (!jvms()->has_method()) {
801 // Root JVMState has a null method.
802 assert(map()->memory()->Opcode() == Op_Parm, "");
803 // Insert the memory aliasing node
804 set_all_memory(reset_memory());
805 }
806 assert(merged_memory(), "");
807
808 switch (intrinsic_id()) {
809 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt_predicate(false);
811 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
812 return inline_cipherBlockChaining_AESCrypt_predicate(true);
813 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
814 return inline_electronicCodeBook_AESCrypt_predicate(false);
815 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
816 return inline_electronicCodeBook_AESCrypt_predicate(true);
817 case vmIntrinsics::_counterMode_AESCrypt:
818 return inline_counterMode_AESCrypt_predicate();
819 case vmIntrinsics::_digestBase_implCompressMB:
820 return inline_digestBase_implCompressMB_predicate(predicate);
821 case vmIntrinsics::_galoisCounterMode_AESCrypt:
822 return inline_galoisCounterMode_AESCrypt_predicate();
823
824 default:
825 // If you get here, it may be that someone has added a new intrinsic
826 // to the list in vmIntrinsics.hpp without implementing it here.
827 #ifndef PRODUCT
828 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
829 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
830 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
831 }
832 #endif
833 Node* slow_ctl = control();
834 set_control(top()); // No fast path intrinsic
835 return slow_ctl;
836 }
837 }
838
839 //------------------------------set_result-------------------------------
840 // Helper function for finishing intrinsics.
841 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
842 record_for_igvn(region);
843 set_control(_gvn.transform(region));
844 set_result( _gvn.transform(value));
845 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
846 }
847
848 //------------------------------generate_guard---------------------------
849 // Helper function for generating guarded fast-slow graph structures.
850 // The given 'test', if true, guards a slow path. If the test fails
851 // then a fast path can be taken. (We generally hope it fails.)
852 // In all cases, GraphKit::control() is updated to the fast path.
853 // The returned value represents the control for the slow path.
854 // The return value is never 'top'; it is either a valid control
855 // or null if it is obvious that the slow path can never be taken.
856 // Also, if region and the slow control are not null, the slow edge
857 // is appended to the region.
858 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
859 if (stopped()) {
860 // Already short circuited.
861 return nullptr;
862 }
863
864 // Build an if node and its projections.
865 // If test is true we take the slow path, which we assume is uncommon.
866 if (_gvn.type(test) == TypeInt::ZERO) {
867 // The slow branch is never taken. No need to build this guard.
868 return nullptr;
869 }
870
871 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
872
873 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
874 if (if_slow == top()) {
875 // The slow branch is never taken. No need to build this guard.
876 return nullptr;
877 }
878
879 if (region != nullptr)
880 region->add_req(if_slow);
881
882 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
883 set_control(if_fast);
884
885 return if_slow;
886 }
887
888 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
889 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
890 }
891 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
892 return generate_guard(test, region, PROB_FAIR);
893 }
894
895 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
896 Node* *pos_index) {
897 if (stopped())
898 return nullptr; // already stopped
899 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
900 return nullptr; // index is already adequately typed
901 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
902 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
903 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
904 if (is_neg != nullptr && pos_index != nullptr) {
905 // Emulate effect of Parse::adjust_map_after_if.
906 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
907 (*pos_index) = _gvn.transform(ccast);
908 }
909 return is_neg;
910 }
911
912 // Make sure that 'position' is a valid limit index, in [0..length].
913 // There are two equivalent plans for checking this:
914 // A. (offset + copyLength) unsigned<= arrayLength
915 // B. offset <= (arrayLength - copyLength)
916 // We require that all of the values above, except for the sum and
917 // difference, are already known to be non-negative.
918 // Plan A is robust in the face of overflow, if offset and copyLength
919 // are both hugely positive.
920 //
921 // Plan B is less direct and intuitive, but it does not overflow at
922 // all, since the difference of two non-negatives is always
923 // representable. Whenever Java methods must perform the equivalent
924 // check they generally use Plan B instead of Plan A.
925 // For the moment we use Plan A.
926 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
927 Node* subseq_length,
928 Node* array_length,
929 RegionNode* region) {
930 if (stopped())
931 return nullptr; // already stopped
932 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
933 if (zero_offset && subseq_length->eqv_uncast(array_length))
934 return nullptr; // common case of whole-array copy
935 Node* last = subseq_length;
936 if (!zero_offset) // last += offset
937 last = _gvn.transform(new AddINode(last, offset));
938 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
939 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
940 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
941 return is_over;
942 }
943
944 // Emit range checks for the given String.value byte array
945 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
946 if (stopped()) {
947 return; // already stopped
948 }
949 RegionNode* bailout = new RegionNode(1);
950 record_for_igvn(bailout);
951 if (char_count) {
952 // Convert char count to byte count
953 count = _gvn.transform(new LShiftINode(count, intcon(1)));
954 }
955
956 // Offset and count must not be negative
957 generate_negative_guard(offset, bailout);
958 generate_negative_guard(count, bailout);
959 // Offset + count must not exceed length of array
960 generate_limit_guard(offset, count, load_array_length(array), bailout);
961
962 if (bailout->req() > 1) {
963 PreserveJVMState pjvms(this);
964 set_control(_gvn.transform(bailout));
965 uncommon_trap(Deoptimization::Reason_intrinsic,
966 Deoptimization::Action_maybe_recompile);
967 }
968 }
969
970 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
971 bool is_immutable) {
972 ciKlass* thread_klass = env()->Thread_klass();
973 const Type* thread_type
974 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
975
976 Node* thread = _gvn.transform(new ThreadLocalNode());
977 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
978 tls_output = thread;
979
980 Node* thread_obj_handle
981 = (is_immutable
982 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
983 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
984 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
985 thread_obj_handle = _gvn.transform(thread_obj_handle);
986
987 DecoratorSet decorators = IN_NATIVE;
988 if (is_immutable) {
989 decorators |= C2_IMMUTABLE_MEMORY;
990 }
991 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
992 }
993
994 //--------------------------generate_current_thread--------------------
995 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
996 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
997 /*is_immutable*/false);
998 }
999
1000 //--------------------------generate_virtual_thread--------------------
1001 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1002 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1003 !C->method()->changes_current_thread());
1004 }
1005
1006 //------------------------------make_string_method_node------------------------
1007 // Helper method for String intrinsic functions. This version is called with
1008 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1009 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1010 // containing the lengths of str1 and str2.
1011 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1012 Node* result = nullptr;
1013 switch (opcode) {
1014 case Op_StrIndexOf:
1015 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1016 str1_start, cnt1, str2_start, cnt2, ae);
1017 break;
1018 case Op_StrComp:
1019 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1020 str1_start, cnt1, str2_start, cnt2, ae);
1021 break;
1022 case Op_StrEquals:
1023 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1024 // Use the constant length if there is one because optimized match rule may exist.
1025 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1026 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1027 break;
1028 default:
1029 ShouldNotReachHere();
1030 return nullptr;
1031 }
1032
1033 // All these intrinsics have checks.
1034 C->set_has_split_ifs(true); // Has chance for split-if optimization
1035 clear_upper_avx();
1036
1037 return _gvn.transform(result);
1038 }
1039
1040 //------------------------------inline_string_compareTo------------------------
1041 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1042 Node* arg1 = argument(0);
1043 Node* arg2 = argument(1);
1044
1045 arg1 = must_be_not_null(arg1, true);
1046 arg2 = must_be_not_null(arg2, true);
1047
1048 // Get start addr and length of first argument
1049 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1050 Node* arg1_cnt = load_array_length(arg1);
1051
1052 // Get start addr and length of second argument
1053 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1054 Node* arg2_cnt = load_array_length(arg2);
1055
1056 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1057 set_result(result);
1058 return true;
1059 }
1060
1061 //------------------------------inline_string_equals------------------------
1062 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1063 Node* arg1 = argument(0);
1064 Node* arg2 = argument(1);
1065
1066 // paths (plus control) merge
1067 RegionNode* region = new RegionNode(3);
1068 Node* phi = new PhiNode(region, TypeInt::BOOL);
1069
1070 if (!stopped()) {
1071
1072 arg1 = must_be_not_null(arg1, true);
1073 arg2 = must_be_not_null(arg2, true);
1074
1075 // Get start addr and length of first argument
1076 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1077 Node* arg1_cnt = load_array_length(arg1);
1078
1079 // Get start addr and length of second argument
1080 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1081 Node* arg2_cnt = load_array_length(arg2);
1082
1083 // Check for arg1_cnt != arg2_cnt
1084 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1085 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1086 Node* if_ne = generate_slow_guard(bol, nullptr);
1087 if (if_ne != nullptr) {
1088 phi->init_req(2, intcon(0));
1089 region->init_req(2, if_ne);
1090 }
1091
1092 // Check for count == 0 is done by assembler code for StrEquals.
1093
1094 if (!stopped()) {
1095 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1096 phi->init_req(1, equals);
1097 region->init_req(1, control());
1098 }
1099 }
1100
1101 // post merge
1102 set_control(_gvn.transform(region));
1103 record_for_igvn(region);
1104
1105 set_result(_gvn.transform(phi));
1106 return true;
1107 }
1108
1109 //------------------------------inline_array_equals----------------------------
1110 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1111 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1112 Node* arg1 = argument(0);
1113 Node* arg2 = argument(1);
1114
1115 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1116 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1117 clear_upper_avx();
1118
1119 return true;
1120 }
1121
1122
1123 //------------------------------inline_countPositives------------------------------
1124 bool LibraryCallKit::inline_countPositives() {
1125 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1126 return false;
1127 }
1128
1129 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1130 // no receiver since it is static method
1131 Node* ba = argument(0);
1132 Node* offset = argument(1);
1133 Node* len = argument(2);
1134
1135 ba = must_be_not_null(ba, true);
1136
1137 // Range checks
1138 generate_string_range_check(ba, offset, len, false);
1139 if (stopped()) {
1140 return true;
1141 }
1142 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1143 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1144 set_result(_gvn.transform(result));
1145 clear_upper_avx();
1146 return true;
1147 }
1148
1149 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1150 Node* index = argument(0);
1151 Node* length = bt == T_INT ? argument(1) : argument(2);
1152 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1153 return false;
1154 }
1155
1156 // check that length is positive
1157 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1158 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1159
1160 {
1161 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1162 uncommon_trap(Deoptimization::Reason_intrinsic,
1163 Deoptimization::Action_make_not_entrant);
1164 }
1165
1166 if (stopped()) {
1167 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1168 return true;
1169 }
1170
1171 // length is now known positive, add a cast node to make this explicit
1172 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1173 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1174 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1175 ConstraintCastNode::RegularDependency, bt);
1176 casted_length = _gvn.transform(casted_length);
1177 replace_in_map(length, casted_length);
1178 length = casted_length;
1179
1180 // Use an unsigned comparison for the range check itself
1181 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1182 BoolTest::mask btest = BoolTest::lt;
1183 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1184 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1185 _gvn.set_type(rc, rc->Value(&_gvn));
1186 if (!rc_bool->is_Con()) {
1187 record_for_igvn(rc);
1188 }
1189 set_control(_gvn.transform(new IfTrueNode(rc)));
1190 {
1191 PreserveJVMState pjvms(this);
1192 set_control(_gvn.transform(new IfFalseNode(rc)));
1193 uncommon_trap(Deoptimization::Reason_range_check,
1194 Deoptimization::Action_make_not_entrant);
1195 }
1196
1197 if (stopped()) {
1198 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1199 return true;
1200 }
1201
1202 // index is now known to be >= 0 and < length, cast it
1203 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1204 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1205 ConstraintCastNode::RegularDependency, bt);
1206 result = _gvn.transform(result);
1207 set_result(result);
1208 replace_in_map(index, result);
1209 return true;
1210 }
1211
1212 //------------------------------inline_string_indexOf------------------------
1213 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1214 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1215 return false;
1216 }
1217 Node* src = argument(0);
1218 Node* tgt = argument(1);
1219
1220 // Make the merge point
1221 RegionNode* result_rgn = new RegionNode(4);
1222 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1223
1224 src = must_be_not_null(src, true);
1225 tgt = must_be_not_null(tgt, true);
1226
1227 // Get start addr and length of source string
1228 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1229 Node* src_count = load_array_length(src);
1230
1231 // Get start addr and length of substring
1232 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1233 Node* tgt_count = load_array_length(tgt);
1234
1235 Node* result = nullptr;
1236 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1237
1238 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1239 // Divide src size by 2 if String is UTF16 encoded
1240 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1241 }
1242 if (ae == StrIntrinsicNode::UU) {
1243 // Divide substring size by 2 if String is UTF16 encoded
1244 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1245 }
1246
1247 if (call_opt_stub) {
1248 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1249 StubRoutines::_string_indexof_array[ae],
1250 "stringIndexOf", TypePtr::BOTTOM, src_start,
1251 src_count, tgt_start, tgt_count);
1252 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1253 } else {
1254 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1255 result_rgn, result_phi, ae);
1256 }
1257 if (result != nullptr) {
1258 result_phi->init_req(3, result);
1259 result_rgn->init_req(3, control());
1260 }
1261 set_control(_gvn.transform(result_rgn));
1262 record_for_igvn(result_rgn);
1263 set_result(_gvn.transform(result_phi));
1264
1265 return true;
1266 }
1267
1268 //-----------------------------inline_string_indexOfI-----------------------
1269 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1270 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1271 return false;
1272 }
1273 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1274 return false;
1275 }
1276
1277 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1278 Node* src = argument(0); // byte[]
1279 Node* src_count = argument(1); // char count
1280 Node* tgt = argument(2); // byte[]
1281 Node* tgt_count = argument(3); // char count
1282 Node* from_index = argument(4); // char index
1283
1284 src = must_be_not_null(src, true);
1285 tgt = must_be_not_null(tgt, true);
1286
1287 // Multiply byte array index by 2 if String is UTF16 encoded
1288 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1289 src_count = _gvn.transform(new SubINode(src_count, from_index));
1290 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1291 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1292
1293 // Range checks
1294 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1295 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1296 if (stopped()) {
1297 return true;
1298 }
1299
1300 RegionNode* region = new RegionNode(5);
1301 Node* phi = new PhiNode(region, TypeInt::INT);
1302 Node* result = nullptr;
1303
1304 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1305
1306 if (call_opt_stub) {
1307 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1308 StubRoutines::_string_indexof_array[ae],
1309 "stringIndexOf", TypePtr::BOTTOM, src_start,
1310 src_count, tgt_start, tgt_count);
1311 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1312 } else {
1313 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1314 region, phi, ae);
1315 }
1316 if (result != nullptr) {
1317 // The result is index relative to from_index if substring was found, -1 otherwise.
1318 // Generate code which will fold into cmove.
1319 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1320 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1321
1322 Node* if_lt = generate_slow_guard(bol, nullptr);
1323 if (if_lt != nullptr) {
1324 // result == -1
1325 phi->init_req(3, result);
1326 region->init_req(3, if_lt);
1327 }
1328 if (!stopped()) {
1329 result = _gvn.transform(new AddINode(result, from_index));
1330 phi->init_req(4, result);
1331 region->init_req(4, control());
1332 }
1333 }
1334
1335 set_control(_gvn.transform(region));
1336 record_for_igvn(region);
1337 set_result(_gvn.transform(phi));
1338 clear_upper_avx();
1339
1340 return true;
1341 }
1342
1343 // Create StrIndexOfNode with fast path checks
1344 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1345 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1346 // Check for substr count > string count
1347 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1348 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1349 Node* if_gt = generate_slow_guard(bol, nullptr);
1350 if (if_gt != nullptr) {
1351 phi->init_req(1, intcon(-1));
1352 region->init_req(1, if_gt);
1353 }
1354 if (!stopped()) {
1355 // Check for substr count == 0
1356 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1357 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1358 Node* if_zero = generate_slow_guard(bol, nullptr);
1359 if (if_zero != nullptr) {
1360 phi->init_req(2, intcon(0));
1361 region->init_req(2, if_zero);
1362 }
1363 }
1364 if (!stopped()) {
1365 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1366 }
1367 return nullptr;
1368 }
1369
1370 //-----------------------------inline_string_indexOfChar-----------------------
1371 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1372 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1373 return false;
1374 }
1375 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1376 return false;
1377 }
1378 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1379 Node* src = argument(0); // byte[]
1380 Node* int_ch = argument(1);
1381 Node* from_index = argument(2);
1382 Node* max = argument(3);
1383
1384 src = must_be_not_null(src, true);
1385
1386 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1387 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1388 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1389
1390 // Range checks
1391 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1392
1393 // Check for int_ch >= 0
1394 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1395 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1396 {
1397 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1398 uncommon_trap(Deoptimization::Reason_intrinsic,
1399 Deoptimization::Action_maybe_recompile);
1400 }
1401 if (stopped()) {
1402 return true;
1403 }
1404
1405 RegionNode* region = new RegionNode(3);
1406 Node* phi = new PhiNode(region, TypeInt::INT);
1407
1408 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1409 C->set_has_split_ifs(true); // Has chance for split-if optimization
1410 _gvn.transform(result);
1411
1412 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1413 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1414
1415 Node* if_lt = generate_slow_guard(bol, nullptr);
1416 if (if_lt != nullptr) {
1417 // result == -1
1418 phi->init_req(2, result);
1419 region->init_req(2, if_lt);
1420 }
1421 if (!stopped()) {
1422 result = _gvn.transform(new AddINode(result, from_index));
1423 phi->init_req(1, result);
1424 region->init_req(1, control());
1425 }
1426 set_control(_gvn.transform(region));
1427 record_for_igvn(region);
1428 set_result(_gvn.transform(phi));
1429 clear_upper_avx();
1430
1431 return true;
1432 }
1433 //---------------------------inline_string_copy---------------------
1434 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1435 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1436 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1437 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1438 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1439 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1440 bool LibraryCallKit::inline_string_copy(bool compress) {
1441 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1442 return false;
1443 }
1444 int nargs = 5; // 2 oops, 3 ints
1445 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1446
1447 Node* src = argument(0);
1448 Node* src_offset = argument(1);
1449 Node* dst = argument(2);
1450 Node* dst_offset = argument(3);
1451 Node* length = argument(4);
1452
1453 // Check for allocation before we add nodes that would confuse
1454 // tightly_coupled_allocation()
1455 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1456
1457 // Figure out the size and type of the elements we will be copying.
1458 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1459 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1460 if (src_type == nullptr || dst_type == nullptr) {
1461 return false;
1462 }
1463 BasicType src_elem = src_type->elem()->array_element_basic_type();
1464 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1465 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1466 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1467 "Unsupported array types for inline_string_copy");
1468
1469 src = must_be_not_null(src, true);
1470 dst = must_be_not_null(dst, true);
1471
1472 // Convert char[] offsets to byte[] offsets
1473 bool convert_src = (compress && src_elem == T_BYTE);
1474 bool convert_dst = (!compress && dst_elem == T_BYTE);
1475 if (convert_src) {
1476 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1477 } else if (convert_dst) {
1478 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1479 }
1480
1481 // Range checks
1482 generate_string_range_check(src, src_offset, length, convert_src);
1483 generate_string_range_check(dst, dst_offset, length, convert_dst);
1484 if (stopped()) {
1485 return true;
1486 }
1487
1488 Node* src_start = array_element_address(src, src_offset, src_elem);
1489 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1490 // 'src_start' points to src array + scaled offset
1491 // 'dst_start' points to dst array + scaled offset
1492 Node* count = nullptr;
1493 if (compress) {
1494 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1495 } else {
1496 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1497 }
1498
1499 if (alloc != nullptr) {
1500 if (alloc->maybe_set_complete(&_gvn)) {
1501 // "You break it, you buy it."
1502 InitializeNode* init = alloc->initialization();
1503 assert(init->is_complete(), "we just did this");
1504 init->set_complete_with_arraycopy();
1505 assert(dst->is_CheckCastPP(), "sanity");
1506 assert(dst->in(0)->in(0) == init, "dest pinned");
1507 }
1508 // Do not let stores that initialize this object be reordered with
1509 // a subsequent store that would make this object accessible by
1510 // other threads.
1511 // Record what AllocateNode this StoreStore protects so that
1512 // escape analysis can go from the MemBarStoreStoreNode to the
1513 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1514 // based on the escape status of the AllocateNode.
1515 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1516 }
1517 if (compress) {
1518 set_result(_gvn.transform(count));
1519 }
1520 clear_upper_avx();
1521
1522 return true;
1523 }
1524
1525 #ifdef _LP64
1526 #define XTOP ,top() /*additional argument*/
1527 #else //_LP64
1528 #define XTOP /*no additional argument*/
1529 #endif //_LP64
1530
1531 //------------------------inline_string_toBytesU--------------------------
1532 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1533 bool LibraryCallKit::inline_string_toBytesU() {
1534 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1535 return false;
1536 }
1537 // Get the arguments.
1538 Node* value = argument(0);
1539 Node* offset = argument(1);
1540 Node* length = argument(2);
1541
1542 Node* newcopy = nullptr;
1543
1544 // Set the original stack and the reexecute bit for the interpreter to reexecute
1545 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1546 { PreserveReexecuteState preexecs(this);
1547 jvms()->set_should_reexecute(true);
1548
1549 // Check if a null path was taken unconditionally.
1550 value = null_check(value);
1551
1552 RegionNode* bailout = new RegionNode(1);
1553 record_for_igvn(bailout);
1554
1555 // Range checks
1556 generate_negative_guard(offset, bailout);
1557 generate_negative_guard(length, bailout);
1558 generate_limit_guard(offset, length, load_array_length(value), bailout);
1559 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1560 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1561
1562 if (bailout->req() > 1) {
1563 PreserveJVMState pjvms(this);
1564 set_control(_gvn.transform(bailout));
1565 uncommon_trap(Deoptimization::Reason_intrinsic,
1566 Deoptimization::Action_maybe_recompile);
1567 }
1568 if (stopped()) {
1569 return true;
1570 }
1571
1572 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1573 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1574 newcopy = new_array(klass_node, size, 0); // no arguments to push
1575 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1576 guarantee(alloc != nullptr, "created above");
1577
1578 // Calculate starting addresses.
1579 Node* src_start = array_element_address(value, offset, T_CHAR);
1580 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1581
1582 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1583 const TypeInt* toffset = gvn().type(offset)->is_int();
1584 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1585
1586 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1587 const char* copyfunc_name = "arraycopy";
1588 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1589 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1590 OptoRuntime::fast_arraycopy_Type(),
1591 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1592 src_start, dst_start, ConvI2X(length) XTOP);
1593 // Do not let reads from the cloned object float above the arraycopy.
1594 if (alloc->maybe_set_complete(&_gvn)) {
1595 // "You break it, you buy it."
1596 InitializeNode* init = alloc->initialization();
1597 assert(init->is_complete(), "we just did this");
1598 init->set_complete_with_arraycopy();
1599 assert(newcopy->is_CheckCastPP(), "sanity");
1600 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1601 }
1602 // Do not let stores that initialize this object be reordered with
1603 // a subsequent store that would make this object accessible by
1604 // other threads.
1605 // Record what AllocateNode this StoreStore protects so that
1606 // escape analysis can go from the MemBarStoreStoreNode to the
1607 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1608 // based on the escape status of the AllocateNode.
1609 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1610 } // original reexecute is set back here
1611
1612 C->set_has_split_ifs(true); // Has chance for split-if optimization
1613 if (!stopped()) {
1614 set_result(newcopy);
1615 }
1616 clear_upper_avx();
1617
1618 return true;
1619 }
1620
1621 //------------------------inline_string_getCharsU--------------------------
1622 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1623 bool LibraryCallKit::inline_string_getCharsU() {
1624 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1625 return false;
1626 }
1627
1628 // Get the arguments.
1629 Node* src = argument(0);
1630 Node* src_begin = argument(1);
1631 Node* src_end = argument(2); // exclusive offset (i < src_end)
1632 Node* dst = argument(3);
1633 Node* dst_begin = argument(4);
1634
1635 // Check for allocation before we add nodes that would confuse
1636 // tightly_coupled_allocation()
1637 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1638
1639 // Check if a null path was taken unconditionally.
1640 src = null_check(src);
1641 dst = null_check(dst);
1642 if (stopped()) {
1643 return true;
1644 }
1645
1646 // Get length and convert char[] offset to byte[] offset
1647 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1648 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1649
1650 // Range checks
1651 generate_string_range_check(src, src_begin, length, true);
1652 generate_string_range_check(dst, dst_begin, length, false);
1653 if (stopped()) {
1654 return true;
1655 }
1656
1657 if (!stopped()) {
1658 // Calculate starting addresses.
1659 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1660 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1661
1662 // Check if array addresses are aligned to HeapWordSize
1663 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1664 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1665 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1666 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1667
1668 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1669 const char* copyfunc_name = "arraycopy";
1670 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1671 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1672 OptoRuntime::fast_arraycopy_Type(),
1673 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1674 src_start, dst_start, ConvI2X(length) XTOP);
1675 // Do not let reads from the cloned object float above the arraycopy.
1676 if (alloc != nullptr) {
1677 if (alloc->maybe_set_complete(&_gvn)) {
1678 // "You break it, you buy it."
1679 InitializeNode* init = alloc->initialization();
1680 assert(init->is_complete(), "we just did this");
1681 init->set_complete_with_arraycopy();
1682 assert(dst->is_CheckCastPP(), "sanity");
1683 assert(dst->in(0)->in(0) == init, "dest pinned");
1684 }
1685 // Do not let stores that initialize this object be reordered with
1686 // a subsequent store that would make this object accessible by
1687 // other threads.
1688 // Record what AllocateNode this StoreStore protects so that
1689 // escape analysis can go from the MemBarStoreStoreNode to the
1690 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1691 // based on the escape status of the AllocateNode.
1692 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1693 } else {
1694 insert_mem_bar(Op_MemBarCPUOrder);
1695 }
1696 }
1697
1698 C->set_has_split_ifs(true); // Has chance for split-if optimization
1699 return true;
1700 }
1701
1702 //----------------------inline_string_char_access----------------------------
1703 // Store/Load char to/from byte[] array.
1704 // static void StringUTF16.putChar(byte[] val, int index, int c)
1705 // static char StringUTF16.getChar(byte[] val, int index)
1706 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1707 Node* value = argument(0);
1708 Node* index = argument(1);
1709 Node* ch = is_store ? argument(2) : nullptr;
1710
1711 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1712 // correctly requires matched array shapes.
1713 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1714 "sanity: byte[] and char[] bases agree");
1715 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1716 "sanity: byte[] and char[] scales agree");
1717
1718 // Bail when getChar over constants is requested: constant folding would
1719 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1720 // Java method would constant fold nicely instead.
1721 if (!is_store && value->is_Con() && index->is_Con()) {
1722 return false;
1723 }
1724
1725 // Save state and restore on bailout
1726 uint old_sp = sp();
1727 SafePointNode* old_map = clone_map();
1728
1729 value = must_be_not_null(value, true);
1730
1731 Node* adr = array_element_address(value, index, T_CHAR);
1732 if (adr->is_top()) {
1733 set_map(old_map);
1734 set_sp(old_sp);
1735 return false;
1736 }
1737 destruct_map_clone(old_map);
1738 if (is_store) {
1739 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1740 } else {
1741 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
1742 set_result(ch);
1743 }
1744 return true;
1745 }
1746
1747
1748 //------------------------------inline_math-----------------------------------
1749 // public static double Math.abs(double)
1750 // public static double Math.sqrt(double)
1751 // public static double Math.log(double)
1752 // public static double Math.log10(double)
1753 // public static double Math.round(double)
1754 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1755 Node* arg = argument(0);
1756 Node* n = nullptr;
1757 switch (id) {
1758 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1759 case vmIntrinsics::_dsqrt:
1760 case vmIntrinsics::_dsqrt_strict:
1761 n = new SqrtDNode(C, control(), arg); break;
1762 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1763 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1764 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1765 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1766 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1767 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1768 default: fatal_unexpected_iid(id); break;
1769 }
1770 set_result(_gvn.transform(n));
1771 return true;
1772 }
1773
1774 //------------------------------inline_math-----------------------------------
1775 // public static float Math.abs(float)
1776 // public static int Math.abs(int)
1777 // public static long Math.abs(long)
1778 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1779 Node* arg = argument(0);
1780 Node* n = nullptr;
1781 switch (id) {
1782 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1783 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1784 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1785 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1786 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1787 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1788 default: fatal_unexpected_iid(id); break;
1789 }
1790 set_result(_gvn.transform(n));
1791 return true;
1792 }
1793
1794 //------------------------------runtime_math-----------------------------
1795 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1796 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1797 "must be (DD)D or (D)D type");
1798
1799 // Inputs
1800 Node* a = argument(0);
1801 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1802
1803 const TypePtr* no_memory_effects = nullptr;
1804 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1805 no_memory_effects,
1806 a, top(), b, b ? top() : nullptr);
1807 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1808 #ifdef ASSERT
1809 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1810 assert(value_top == top(), "second value must be top");
1811 #endif
1812
1813 set_result(value);
1814 return true;
1815 }
1816
1817 //------------------------------inline_math_pow-----------------------------
1818 bool LibraryCallKit::inline_math_pow() {
1819 Node* exp = argument(2);
1820 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1821 if (d != nullptr) {
1822 if (d->getd() == 2.0) {
1823 // Special case: pow(x, 2.0) => x * x
1824 Node* base = argument(0);
1825 set_result(_gvn.transform(new MulDNode(base, base)));
1826 return true;
1827 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1828 // Special case: pow(x, 0.5) => sqrt(x)
1829 Node* base = argument(0);
1830 Node* zero = _gvn.zerocon(T_DOUBLE);
1831
1832 RegionNode* region = new RegionNode(3);
1833 Node* phi = new PhiNode(region, Type::DOUBLE);
1834
1835 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1836 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1837 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1838 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1839 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1840
1841 Node* if_pow = generate_slow_guard(test, nullptr);
1842 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1843 phi->init_req(1, value_sqrt);
1844 region->init_req(1, control());
1845
1846 if (if_pow != nullptr) {
1847 set_control(if_pow);
1848 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1849 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1850 const TypePtr* no_memory_effects = nullptr;
1851 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1852 no_memory_effects, base, top(), exp, top());
1853 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1854 #ifdef ASSERT
1855 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1856 assert(value_top == top(), "second value must be top");
1857 #endif
1858 phi->init_req(2, value_pow);
1859 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1860 }
1861
1862 C->set_has_split_ifs(true); // Has chance for split-if optimization
1863 set_control(_gvn.transform(region));
1864 record_for_igvn(region);
1865 set_result(_gvn.transform(phi));
1866
1867 return true;
1868 }
1869 }
1870
1871 return StubRoutines::dpow() != nullptr ?
1872 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1873 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1874 }
1875
1876 //------------------------------inline_math_native-----------------------------
1877 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1878 switch (id) {
1879 case vmIntrinsics::_dsin:
1880 return StubRoutines::dsin() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1883 case vmIntrinsics::_dcos:
1884 return StubRoutines::dcos() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1887 case vmIntrinsics::_dtan:
1888 return StubRoutines::dtan() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1891 case vmIntrinsics::_dtanh:
1892 return StubRoutines::dtanh() != nullptr ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1894 case vmIntrinsics::_dexp:
1895 return StubRoutines::dexp() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1898 case vmIntrinsics::_dlog:
1899 return StubRoutines::dlog() != nullptr ?
1900 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1901 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1902 case vmIntrinsics::_dlog10:
1903 return StubRoutines::dlog10() != nullptr ?
1904 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1905 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1906
1907 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_ceil:
1909 case vmIntrinsics::_floor:
1910 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1911
1912 case vmIntrinsics::_dsqrt:
1913 case vmIntrinsics::_dsqrt_strict:
1914 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1916 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1917 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1918 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1919
1920 case vmIntrinsics::_dpow: return inline_math_pow();
1921 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1922 case vmIntrinsics::_fcopySign: return inline_math(id);
1923 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1924 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1925 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1926
1927 // These intrinsics are not yet correctly implemented
1928 case vmIntrinsics::_datan2:
1929 return false;
1930
1931 default:
1932 fatal_unexpected_iid(id);
1933 return false;
1934 }
1935 }
1936
1937 //----------------------------inline_notify-----------------------------------*
1938 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1939 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1940 address func;
1941 if (id == vmIntrinsics::_notify) {
1942 func = OptoRuntime::monitor_notify_Java();
1943 } else {
1944 func = OptoRuntime::monitor_notifyAll_Java();
1945 }
1946 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1947 make_slow_call_ex(call, env()->Throwable_klass(), false);
1948 return true;
1949 }
1950
1951
1952 //----------------------------inline_min_max-----------------------------------
1953 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1954 Node* a = nullptr;
1955 Node* b = nullptr;
1956 Node* n = nullptr;
1957 switch (id) {
1958 case vmIntrinsics::_min:
1959 case vmIntrinsics::_max:
1960 case vmIntrinsics::_minF:
1961 case vmIntrinsics::_maxF:
1962 case vmIntrinsics::_minF_strict:
1963 case vmIntrinsics::_maxF_strict:
1964 case vmIntrinsics::_min_strict:
1965 case vmIntrinsics::_max_strict:
1966 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1967 a = argument(0);
1968 b = argument(1);
1969 break;
1970 case vmIntrinsics::_minD:
1971 case vmIntrinsics::_maxD:
1972 case vmIntrinsics::_minD_strict:
1973 case vmIntrinsics::_maxD_strict:
1974 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1975 a = argument(0);
1976 b = argument(2);
1977 break;
1978 case vmIntrinsics::_minL:
1979 case vmIntrinsics::_maxL:
1980 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1981 a = argument(0);
1982 b = argument(2);
1983 break;
1984 default:
1985 fatal_unexpected_iid(id);
1986 break;
1987 }
1988
1989 switch (id) {
1990 case vmIntrinsics::_min:
1991 case vmIntrinsics::_min_strict:
1992 n = new MinINode(a, b);
1993 break;
1994 case vmIntrinsics::_max:
1995 case vmIntrinsics::_max_strict:
1996 n = new MaxINode(a, b);
1997 break;
1998 case vmIntrinsics::_minF:
1999 case vmIntrinsics::_minF_strict:
2000 n = new MinFNode(a, b);
2001 break;
2002 case vmIntrinsics::_maxF:
2003 case vmIntrinsics::_maxF_strict:
2004 n = new MaxFNode(a, b);
2005 break;
2006 case vmIntrinsics::_minD:
2007 case vmIntrinsics::_minD_strict:
2008 n = new MinDNode(a, b);
2009 break;
2010 case vmIntrinsics::_maxD:
2011 case vmIntrinsics::_maxD_strict:
2012 n = new MaxDNode(a, b);
2013 break;
2014 case vmIntrinsics::_minL:
2015 n = new MinLNode(_gvn.C, a, b);
2016 break;
2017 case vmIntrinsics::_maxL:
2018 n = new MaxLNode(_gvn.C, a, b);
2019 break;
2020 default:
2021 fatal_unexpected_iid(id);
2022 break;
2023 }
2024
2025 set_result(_gvn.transform(n));
2026 return true;
2027 }
2028
2029 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2030 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2031 env()->ArithmeticException_instance())) {
2032 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2033 // so let's bail out intrinsic rather than risking deopting again.
2034 return false;
2035 }
2036
2037 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2038 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2039 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2040 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2041
2042 {
2043 PreserveJVMState pjvms(this);
2044 PreserveReexecuteState preexecs(this);
2045 jvms()->set_should_reexecute(true);
2046
2047 set_control(slow_path);
2048 set_i_o(i_o());
2049
2050 builtin_throw(Deoptimization::Reason_intrinsic,
2051 env()->ArithmeticException_instance(),
2052 /*allow_too_many_traps*/ false);
2053 }
2054
2055 set_control(fast_path);
2056 set_result(math);
2057 return true;
2058 }
2059
2060 template <typename OverflowOp>
2061 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2062 typedef typename OverflowOp::MathOp MathOp;
2063
2064 MathOp* mathOp = new MathOp(arg1, arg2);
2065 Node* operation = _gvn.transform( mathOp );
2066 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2067 return inline_math_mathExact(operation, ofcheck);
2068 }
2069
2070 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2071 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2072 }
2073
2074 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2075 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2076 }
2077
2078 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2079 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2080 }
2081
2082 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2083 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2084 }
2085
2086 bool LibraryCallKit::inline_math_negateExactI() {
2087 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2088 }
2089
2090 bool LibraryCallKit::inline_math_negateExactL() {
2091 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2092 }
2093
2094 bool LibraryCallKit::inline_math_multiplyExactI() {
2095 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2096 }
2097
2098 bool LibraryCallKit::inline_math_multiplyExactL() {
2099 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2100 }
2101
2102 bool LibraryCallKit::inline_math_multiplyHigh() {
2103 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2104 return true;
2105 }
2106
2107 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2108 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2109 return true;
2110 }
2111
2112 inline int
2113 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2114 const TypePtr* base_type = TypePtr::NULL_PTR;
2115 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2116 if (base_type == nullptr) {
2117 // Unknown type.
2118 return Type::AnyPtr;
2119 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2120 // Since this is a null+long form, we have to switch to a rawptr.
2121 base = _gvn.transform(new CastX2PNode(offset));
2122 offset = MakeConX(0);
2123 return Type::RawPtr;
2124 } else if (base_type->base() == Type::RawPtr) {
2125 return Type::RawPtr;
2126 } else if (base_type->isa_oopptr()) {
2127 // Base is never null => always a heap address.
2128 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2129 return Type::OopPtr;
2130 }
2131 // Offset is small => always a heap address.
2132 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2133 if (offset_type != nullptr &&
2134 base_type->offset() == 0 && // (should always be?)
2135 offset_type->_lo >= 0 &&
2136 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2137 return Type::OopPtr;
2138 } else if (type == T_OBJECT) {
2139 // off heap access to an oop doesn't make any sense. Has to be on
2140 // heap.
2141 return Type::OopPtr;
2142 }
2143 // Otherwise, it might either be oop+off or null+addr.
2144 return Type::AnyPtr;
2145 } else {
2146 // No information:
2147 return Type::AnyPtr;
2148 }
2149 }
2150
2151 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2152 Node* uncasted_base = base;
2153 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2154 if (kind == Type::RawPtr) {
2155 return basic_plus_adr(top(), uncasted_base, offset);
2156 } else if (kind == Type::AnyPtr) {
2157 assert(base == uncasted_base, "unexpected base change");
2158 if (can_cast) {
2159 if (!_gvn.type(base)->speculative_maybe_null() &&
2160 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2161 // According to profiling, this access is always on
2162 // heap. Casting the base to not null and thus avoiding membars
2163 // around the access should allow better optimizations
2164 Node* null_ctl = top();
2165 base = null_check_oop(base, &null_ctl, true, true, true);
2166 assert(null_ctl->is_top(), "no null control here");
2167 return basic_plus_adr(base, offset);
2168 } else if (_gvn.type(base)->speculative_always_null() &&
2169 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2170 // According to profiling, this access is always off
2171 // heap.
2172 base = null_assert(base);
2173 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2174 offset = MakeConX(0);
2175 return basic_plus_adr(top(), raw_base, offset);
2176 }
2177 }
2178 // We don't know if it's an on heap or off heap access. Fall back
2179 // to raw memory access.
2180 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2181 return basic_plus_adr(top(), raw, offset);
2182 } else {
2183 assert(base == uncasted_base, "unexpected base change");
2184 // We know it's an on heap access so base can't be null
2185 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2186 base = must_be_not_null(base, true);
2187 }
2188 return basic_plus_adr(base, offset);
2189 }
2190 }
2191
2192 //--------------------------inline_number_methods-----------------------------
2193 // inline int Integer.numberOfLeadingZeros(int)
2194 // inline int Long.numberOfLeadingZeros(long)
2195 //
2196 // inline int Integer.numberOfTrailingZeros(int)
2197 // inline int Long.numberOfTrailingZeros(long)
2198 //
2199 // inline int Integer.bitCount(int)
2200 // inline int Long.bitCount(long)
2201 //
2202 // inline char Character.reverseBytes(char)
2203 // inline short Short.reverseBytes(short)
2204 // inline int Integer.reverseBytes(int)
2205 // inline long Long.reverseBytes(long)
2206 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2207 Node* arg = argument(0);
2208 Node* n = nullptr;
2209 switch (id) {
2210 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2211 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2212 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2213 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2214 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2215 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2216 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2217 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2218 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2219 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2220 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2221 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2222 default: fatal_unexpected_iid(id); break;
2223 }
2224 set_result(_gvn.transform(n));
2225 return true;
2226 }
2227
2228 //--------------------------inline_bitshuffle_methods-----------------------------
2229 // inline int Integer.compress(int, int)
2230 // inline int Integer.expand(int, int)
2231 // inline long Long.compress(long, long)
2232 // inline long Long.expand(long, long)
2233 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2234 Node* n = nullptr;
2235 switch (id) {
2236 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2237 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2238 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2239 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2240 default: fatal_unexpected_iid(id); break;
2241 }
2242 set_result(_gvn.transform(n));
2243 return true;
2244 }
2245
2246 //--------------------------inline_number_methods-----------------------------
2247 // inline int Integer.compareUnsigned(int, int)
2248 // inline int Long.compareUnsigned(long, long)
2249 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2250 Node* arg1 = argument(0);
2251 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2252 Node* n = nullptr;
2253 switch (id) {
2254 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2255 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2256 default: fatal_unexpected_iid(id); break;
2257 }
2258 set_result(_gvn.transform(n));
2259 return true;
2260 }
2261
2262 //--------------------------inline_unsigned_divmod_methods-----------------------------
2263 // inline int Integer.divideUnsigned(int, int)
2264 // inline int Integer.remainderUnsigned(int, int)
2265 // inline long Long.divideUnsigned(long, long)
2266 // inline long Long.remainderUnsigned(long, long)
2267 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2268 Node* n = nullptr;
2269 switch (id) {
2270 case vmIntrinsics::_divideUnsigned_i: {
2271 zero_check_int(argument(1));
2272 // Compile-time detect of null-exception
2273 if (stopped()) {
2274 return true; // keep the graph constructed so far
2275 }
2276 n = new UDivINode(control(), argument(0), argument(1));
2277 break;
2278 }
2279 case vmIntrinsics::_divideUnsigned_l: {
2280 zero_check_long(argument(2));
2281 // Compile-time detect of null-exception
2282 if (stopped()) {
2283 return true; // keep the graph constructed so far
2284 }
2285 n = new UDivLNode(control(), argument(0), argument(2));
2286 break;
2287 }
2288 case vmIntrinsics::_remainderUnsigned_i: {
2289 zero_check_int(argument(1));
2290 // Compile-time detect of null-exception
2291 if (stopped()) {
2292 return true; // keep the graph constructed so far
2293 }
2294 n = new UModINode(control(), argument(0), argument(1));
2295 break;
2296 }
2297 case vmIntrinsics::_remainderUnsigned_l: {
2298 zero_check_long(argument(2));
2299 // Compile-time detect of null-exception
2300 if (stopped()) {
2301 return true; // keep the graph constructed so far
2302 }
2303 n = new UModLNode(control(), argument(0), argument(2));
2304 break;
2305 }
2306 default: fatal_unexpected_iid(id); break;
2307 }
2308 set_result(_gvn.transform(n));
2309 return true;
2310 }
2311
2312 //----------------------------inline_unsafe_access----------------------------
2313
2314 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2315 // Attempt to infer a sharper value type from the offset and base type.
2316 ciKlass* sharpened_klass = nullptr;
2317 bool null_free = false;
2318
2319 // See if it is an instance field, with an object type.
2320 if (alias_type->field() != nullptr) {
2321 if (alias_type->field()->type()->is_klass()) {
2322 sharpened_klass = alias_type->field()->type()->as_klass();
2323 null_free = alias_type->field()->is_null_free();
2324 }
2325 }
2326
2327 const TypeOopPtr* result = nullptr;
2328 // See if it is a narrow oop array.
2329 if (adr_type->isa_aryptr()) {
2330 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2331 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2332 null_free = adr_type->is_aryptr()->is_null_free();
2333 if (elem_type != nullptr && elem_type->is_loaded()) {
2334 // Sharpen the value type.
2335 result = elem_type;
2336 }
2337 }
2338 }
2339
2340 // The sharpened class might be unloaded if there is no class loader
2341 // contraint in place.
2342 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2343 // Sharpen the value type.
2344 result = TypeOopPtr::make_from_klass(sharpened_klass);
2345 if (null_free) {
2346 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2347 }
2348 }
2349 if (result != nullptr) {
2350 #ifndef PRODUCT
2351 if (C->print_intrinsics() || C->print_inlining()) {
2352 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2353 tty->print(" sharpened value: "); result->dump(); tty->cr();
2354 }
2355 #endif
2356 }
2357 return result;
2358 }
2359
2360 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2361 switch (kind) {
2362 case Relaxed:
2363 return MO_UNORDERED;
2364 case Opaque:
2365 return MO_RELAXED;
2366 case Acquire:
2367 return MO_ACQUIRE;
2368 case Release:
2369 return MO_RELEASE;
2370 case Volatile:
2371 return MO_SEQ_CST;
2372 default:
2373 ShouldNotReachHere();
2374 return 0;
2375 }
2376 }
2377
2378 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2379 if (callee()->is_static()) return false; // caller must have the capability!
2380 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2381 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2382 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2383 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2384
2385 if (is_reference_type(type)) {
2386 decorators |= ON_UNKNOWN_OOP_REF;
2387 }
2388
2389 if (unaligned) {
2390 decorators |= C2_UNALIGNED;
2391 }
2392
2393 #ifndef PRODUCT
2394 {
2395 ResourceMark rm;
2396 // Check the signatures.
2397 ciSignature* sig = callee()->signature();
2398 #ifdef ASSERT
2399 if (!is_store) {
2400 // Object getReference(Object base, int/long offset), etc.
2401 BasicType rtype = sig->return_type()->basic_type();
2402 assert(rtype == type, "getter must return the expected value");
2403 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2404 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2405 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2406 } else {
2407 // void putReference(Object base, int/long offset, Object x), etc.
2408 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2409 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2410 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2411 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2412 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2413 assert(vtype == type, "putter must accept the expected value");
2414 }
2415 #endif // ASSERT
2416 }
2417 #endif //PRODUCT
2418
2419 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2420
2421 Node* receiver = argument(0); // type: oop
2422
2423 // Build address expression.
2424 Node* heap_base_oop = top();
2425
2426 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2427 Node* base = argument(1); // type: oop
2428 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2429 Node* offset = argument(2); // type: long
2430 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2431 // to be plain byte offsets, which are also the same as those accepted
2432 // by oopDesc::field_addr.
2433 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2434 "fieldOffset must be byte-scaled");
2435
2436 ciInlineKlass* inline_klass = nullptr;
2437 if (is_flat) {
2438 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2439 if (cls == nullptr || cls->const_oop() == nullptr) {
2440 return false;
2441 }
2442 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2443 if (!mirror_type->is_inlinetype()) {
2444 return false;
2445 }
2446 inline_klass = mirror_type->as_inline_klass();
2447 }
2448
2449 if (base->is_InlineType()) {
2450 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2451 InlineTypeNode* vt = base->as_InlineType();
2452 if (offset->is_Con()) {
2453 long off = find_long_con(offset, 0);
2454 ciInlineKlass* vk = vt->type()->inline_klass();
2455 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2456 return false;
2457 }
2458
2459 ciField* field = vk->get_non_flat_field_by_offset(off);
2460 if (field != nullptr) {
2461 BasicType bt = type2field[field->type()->basic_type()];
2462 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2463 bt = T_OBJECT;
2464 }
2465 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2466 Node* value = vt->field_value_by_offset(off, false);
2467 if (value->is_InlineType()) {
2468 value = value->as_InlineType()->adjust_scalarization_depth(this);
2469 }
2470 set_result(value);
2471 return true;
2472 }
2473 }
2474 }
2475 {
2476 // Re-execute the unsafe access if allocation triggers deoptimization.
2477 PreserveReexecuteState preexecs(this);
2478 jvms()->set_should_reexecute(true);
2479 vt = vt->buffer(this);
2480 }
2481 base = vt->get_oop();
2482 }
2483
2484 // 32-bit machines ignore the high half!
2485 offset = ConvL2X(offset);
2486
2487 // Save state and restore on bailout
2488 uint old_sp = sp();
2489 SafePointNode* old_map = clone_map();
2490
2491 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2492 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2493
2494 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2495 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2496 decorators |= IN_NATIVE; // off-heap primitive access
2497 } else {
2498 set_map(old_map);
2499 set_sp(old_sp);
2500 return false; // off-heap oop accesses are not supported
2501 }
2502 } else {
2503 heap_base_oop = base; // on-heap or mixed access
2504 }
2505
2506 // Can base be null? Otherwise, always on-heap access.
2507 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2508
2509 if (!can_access_non_heap) {
2510 decorators |= IN_HEAP;
2511 }
2512
2513 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2514
2515 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2516 if (adr_type == TypePtr::NULL_PTR) {
2517 set_map(old_map);
2518 set_sp(old_sp);
2519 return false; // off-heap access with zero address
2520 }
2521
2522 // Try to categorize the address.
2523 Compile::AliasType* alias_type = C->alias_type(adr_type);
2524 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2525
2526 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2527 alias_type->adr_type() == TypeAryPtr::RANGE) {
2528 set_map(old_map);
2529 set_sp(old_sp);
2530 return false; // not supported
2531 }
2532
2533 bool mismatched = false;
2534 BasicType bt = T_ILLEGAL;
2535 ciField* field = nullptr;
2536 if (adr_type->isa_instptr()) {
2537 const TypeInstPtr* instptr = adr_type->is_instptr();
2538 ciInstanceKlass* k = instptr->instance_klass();
2539 int off = instptr->offset();
2540 if (instptr->const_oop() != nullptr &&
2541 k == ciEnv::current()->Class_klass() &&
2542 instptr->offset() >= (k->size_helper() * wordSize)) {
2543 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2544 field = k->get_field_by_offset(off, true);
2545 } else {
2546 field = k->get_non_flat_field_by_offset(off);
2547 }
2548 if (field != nullptr) {
2549 bt = type2field[field->type()->basic_type()];
2550 }
2551 if (bt != alias_type->basic_type()) {
2552 // Type mismatch. Is it an access to a nested flat field?
2553 field = k->get_field_by_offset(off, false);
2554 if (field != nullptr) {
2555 bt = type2field[field->type()->basic_type()];
2556 }
2557 }
2558 assert(bt == alias_type->basic_type() || is_flat, "should match");
2559 } else {
2560 bt = alias_type->basic_type();
2561 }
2562
2563 if (bt != T_ILLEGAL) {
2564 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2565 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2566 // Alias type doesn't differentiate between byte[] and boolean[]).
2567 // Use address type to get the element type.
2568 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2569 }
2570 if (is_reference_type(bt, true)) {
2571 // accessing an array field with getReference is not a mismatch
2572 bt = T_OBJECT;
2573 }
2574 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2575 // Don't intrinsify mismatched object accesses
2576 set_map(old_map);
2577 set_sp(old_sp);
2578 return false;
2579 }
2580 mismatched = (bt != type);
2581 } else if (alias_type->adr_type()->isa_oopptr()) {
2582 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2583 }
2584
2585 if (is_flat) {
2586 if (adr_type->isa_instptr()) {
2587 if (field == nullptr || field->type() != inline_klass) {
2588 mismatched = true;
2589 }
2590 } else if (adr_type->isa_aryptr()) {
2591 const Type* elem = adr_type->is_aryptr()->elem();
2592 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2593 mismatched = true;
2594 }
2595 } else {
2596 mismatched = true;
2597 }
2598 if (is_store) {
2599 const Type* val_t = _gvn.type(val);
2600 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2601 set_map(old_map);
2602 set_sp(old_sp);
2603 return false;
2604 }
2605 }
2606 }
2607
2608 destruct_map_clone(old_map);
2609 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2610
2611 if (mismatched) {
2612 decorators |= C2_MISMATCHED;
2613 }
2614
2615 // First guess at the value type.
2616 const Type *value_type = Type::get_const_basic_type(type);
2617
2618 // Figure out the memory ordering.
2619 decorators |= mo_decorator_for_access_kind(kind);
2620
2621 if (!is_store) {
2622 if (type == T_OBJECT && !is_flat) {
2623 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2624 if (tjp != nullptr) {
2625 value_type = tjp;
2626 }
2627 }
2628 }
2629
2630 receiver = null_check(receiver);
2631 if (stopped()) {
2632 return true;
2633 }
2634 // Heap pointers get a null-check from the interpreter,
2635 // as a courtesy. However, this is not guaranteed by Unsafe,
2636 // and it is not possible to fully distinguish unintended nulls
2637 // from intended ones in this API.
2638
2639 if (!is_store) {
2640 Node* p = nullptr;
2641 // Try to constant fold a load from a constant field
2642
2643 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2644 // final or stable field
2645 p = make_constant_from_field(field, heap_base_oop);
2646 }
2647
2648 if (p == nullptr) { // Could not constant fold the load
2649 if (is_flat) {
2650 if (adr_type->isa_instptr() && !mismatched) {
2651 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2652 int offset = adr_type->is_instptr()->offset();
2653 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, nullptr, holder, offset, false, -1, decorators);
2654 } else {
2655 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, nullptr, 0, false, -1, decorators);
2656 }
2657 } else {
2658 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2659 const TypeOopPtr* ptr = value_type->make_oopptr();
2660 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2661 // Load a non-flattened inline type from memory
2662 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2663 }
2664 }
2665 // Normalize the value returned by getBoolean in the following cases
2666 if (type == T_BOOLEAN &&
2667 (mismatched ||
2668 heap_base_oop == top() || // - heap_base_oop is null or
2669 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2670 // and the unsafe access is made to large offset
2671 // (i.e., larger than the maximum offset necessary for any
2672 // field access)
2673 ) {
2674 IdealKit ideal = IdealKit(this);
2675 #define __ ideal.
2676 IdealVariable normalized_result(ideal);
2677 __ declarations_done();
2678 __ set(normalized_result, p);
2679 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2680 __ set(normalized_result, ideal.ConI(1));
2681 ideal.end_if();
2682 final_sync(ideal);
2683 p = __ value(normalized_result);
2684 #undef __
2685 }
2686 }
2687 if (type == T_ADDRESS) {
2688 p = gvn().transform(new CastP2XNode(nullptr, p));
2689 p = ConvX2UL(p);
2690 }
2691 // The load node has the control of the preceding MemBarCPUOrder. All
2692 // following nodes will have the control of the MemBarCPUOrder inserted at
2693 // the end of this method. So, pushing the load onto the stack at a later
2694 // point is fine.
2695 set_result(p);
2696 } else {
2697 if (bt == T_ADDRESS) {
2698 // Repackage the long as a pointer.
2699 val = ConvL2X(val);
2700 val = gvn().transform(new CastX2PNode(val));
2701 }
2702 if (is_flat) {
2703 if (adr_type->isa_instptr() && !mismatched) {
2704 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2705 int offset = adr_type->is_instptr()->offset();
2706 val->as_InlineType()->store_flat(this, base, base, nullptr, holder, offset, false, -1, decorators);
2707 } else {
2708 val->as_InlineType()->store_flat(this, base, adr, nullptr, val->bottom_type()->inline_klass(), 0, false, -1, decorators);
2709 }
2710 } else {
2711 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2712 }
2713 }
2714
2715 return true;
2716 }
2717
2718 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2719 Node* receiver = argument(0);
2720 Node* value = argument(1);
2721
2722 const Type* type = gvn().type(value);
2723 if (!type->is_inlinetypeptr()) {
2724 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2725 return false;
2726 }
2727
2728 null_check(receiver);
2729 if (stopped()) {
2730 return true;
2731 }
2732
2733 value = null_check(value);
2734 if (stopped()) {
2735 return true;
2736 }
2737
2738 ciInlineKlass* vk = type->inline_klass();
2739 Node* klass = makecon(TypeKlassPtr::make(vk));
2740 Node* obj = new_instance(klass);
2741 AllocateNode::Ideal_allocation(obj)->_larval = true;
2742
2743 assert(value->is_InlineType(), "must be an InlineTypeNode");
2744 value->as_InlineType()->store(this, obj, obj, vk);
2745
2746 set_result(obj);
2747 return true;
2748 }
2749
2750 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2751 Node* receiver = argument(0);
2752 Node* buffer = argument(1);
2753
2754 const Type* type = gvn().type(buffer);
2755 if (!type->is_inlinetypeptr()) {
2756 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2757 return false;
2758 }
2759
2760 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2761 if (alloc == nullptr) {
2762 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2763 return false;
2764 }
2765
2766 null_check(receiver);
2767 if (stopped()) {
2768 return true;
2769 }
2770
2771 // Unset the larval bit in the object header
2772 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2773 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2774 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2775
2776 // We must ensure that the buffer is properly published
2777 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2778 assert(!type->maybe_null(), "result of an allocation should not be null");
2779 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
2780 return true;
2781 }
2782
2783 //----------------------------inline_unsafe_load_store----------------------------
2784 // This method serves a couple of different customers (depending on LoadStoreKind):
2785 //
2786 // LS_cmp_swap:
2787 //
2788 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2789 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2790 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2791 //
2792 // LS_cmp_swap_weak:
2793 //
2794 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2795 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2796 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2797 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2798 //
2799 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2800 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2801 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2802 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2803 //
2804 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2805 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2806 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2807 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2808 //
2809 // LS_cmp_exchange:
2810 //
2811 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2812 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2813 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2814 //
2815 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2816 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2817 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2818 //
2819 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2820 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2821 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2822 //
2823 // LS_get_add:
2824 //
2825 // int getAndAddInt( Object o, long offset, int delta)
2826 // long getAndAddLong(Object o, long offset, long delta)
2827 //
2828 // LS_get_set:
2829 //
2830 // int getAndSet(Object o, long offset, int newValue)
2831 // long getAndSet(Object o, long offset, long newValue)
2832 // Object getAndSet(Object o, long offset, Object newValue)
2833 //
2834 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2835 // This basic scheme here is the same as inline_unsafe_access, but
2836 // differs in enough details that combining them would make the code
2837 // overly confusing. (This is a true fact! I originally combined
2838 // them, but even I was confused by it!) As much code/comments as
2839 // possible are retained from inline_unsafe_access though to make
2840 // the correspondences clearer. - dl
2841
2842 if (callee()->is_static()) return false; // caller must have the capability!
2843
2844 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2845 decorators |= mo_decorator_for_access_kind(access_kind);
2846
2847 #ifndef PRODUCT
2848 BasicType rtype;
2849 {
2850 ResourceMark rm;
2851 // Check the signatures.
2852 ciSignature* sig = callee()->signature();
2853 rtype = sig->return_type()->basic_type();
2854 switch(kind) {
2855 case LS_get_add:
2856 case LS_get_set: {
2857 // Check the signatures.
2858 #ifdef ASSERT
2859 assert(rtype == type, "get and set must return the expected type");
2860 assert(sig->count() == 3, "get and set has 3 arguments");
2861 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2862 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2863 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2864 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2865 #endif // ASSERT
2866 break;
2867 }
2868 case LS_cmp_swap:
2869 case LS_cmp_swap_weak: {
2870 // Check the signatures.
2871 #ifdef ASSERT
2872 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2873 assert(sig->count() == 4, "CAS has 4 arguments");
2874 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2875 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2876 #endif // ASSERT
2877 break;
2878 }
2879 case LS_cmp_exchange: {
2880 // Check the signatures.
2881 #ifdef ASSERT
2882 assert(rtype == type, "CAS must return the expected type");
2883 assert(sig->count() == 4, "CAS has 4 arguments");
2884 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2885 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2886 #endif // ASSERT
2887 break;
2888 }
2889 default:
2890 ShouldNotReachHere();
2891 }
2892 }
2893 #endif //PRODUCT
2894
2895 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2896
2897 // Get arguments:
2898 Node* receiver = nullptr;
2899 Node* base = nullptr;
2900 Node* offset = nullptr;
2901 Node* oldval = nullptr;
2902 Node* newval = nullptr;
2903 switch(kind) {
2904 case LS_cmp_swap:
2905 case LS_cmp_swap_weak:
2906 case LS_cmp_exchange: {
2907 const bool two_slot_type = type2size[type] == 2;
2908 receiver = argument(0); // type: oop
2909 base = argument(1); // type: oop
2910 offset = argument(2); // type: long
2911 oldval = argument(4); // type: oop, int, or long
2912 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2913 break;
2914 }
2915 case LS_get_add:
2916 case LS_get_set: {
2917 receiver = argument(0); // type: oop
2918 base = argument(1); // type: oop
2919 offset = argument(2); // type: long
2920 oldval = nullptr;
2921 newval = argument(4); // type: oop, int, or long
2922 break;
2923 }
2924 default:
2925 ShouldNotReachHere();
2926 }
2927
2928 // Build field offset expression.
2929 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2930 // to be plain byte offsets, which are also the same as those accepted
2931 // by oopDesc::field_addr.
2932 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2933 // 32-bit machines ignore the high half of long offsets
2934 offset = ConvL2X(offset);
2935 // Save state and restore on bailout
2936 uint old_sp = sp();
2937 SafePointNode* old_map = clone_map();
2938 Node* adr = make_unsafe_address(base, offset,type, false);
2939 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2940
2941 Compile::AliasType* alias_type = C->alias_type(adr_type);
2942 BasicType bt = alias_type->basic_type();
2943 if (bt != T_ILLEGAL &&
2944 (is_reference_type(bt) != (type == T_OBJECT))) {
2945 // Don't intrinsify mismatched object accesses.
2946 set_map(old_map);
2947 set_sp(old_sp);
2948 return false;
2949 }
2950
2951 destruct_map_clone(old_map);
2952
2953 // For CAS, unlike inline_unsafe_access, there seems no point in
2954 // trying to refine types. Just use the coarse types here.
2955 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2956 const Type *value_type = Type::get_const_basic_type(type);
2957
2958 switch (kind) {
2959 case LS_get_set:
2960 case LS_cmp_exchange: {
2961 if (type == T_OBJECT) {
2962 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2963 if (tjp != nullptr) {
2964 value_type = tjp;
2965 }
2966 }
2967 break;
2968 }
2969 case LS_cmp_swap:
2970 case LS_cmp_swap_weak:
2971 case LS_get_add:
2972 break;
2973 default:
2974 ShouldNotReachHere();
2975 }
2976
2977 // Null check receiver.
2978 receiver = null_check(receiver);
2979 if (stopped()) {
2980 return true;
2981 }
2982
2983 int alias_idx = C->get_alias_index(adr_type);
2984
2985 if (is_reference_type(type)) {
2986 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2987
2988 if (oldval != nullptr && oldval->is_InlineType()) {
2989 // Re-execute the unsafe access if allocation triggers deoptimization.
2990 PreserveReexecuteState preexecs(this);
2991 jvms()->set_should_reexecute(true);
2992 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2993 }
2994 if (newval != nullptr && newval->is_InlineType()) {
2995 // Re-execute the unsafe access if allocation triggers deoptimization.
2996 PreserveReexecuteState preexecs(this);
2997 jvms()->set_should_reexecute(true);
2998 newval = newval->as_InlineType()->buffer(this)->get_oop();
2999 }
3000
3001 // Transformation of a value which could be null pointer (CastPP #null)
3002 // could be delayed during Parse (for example, in adjust_map_after_if()).
3003 // Execute transformation here to avoid barrier generation in such case.
3004 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3005 newval = _gvn.makecon(TypePtr::NULL_PTR);
3006
3007 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3008 // Refine the value to a null constant, when it is known to be null
3009 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3010 }
3011 }
3012
3013 Node* result = nullptr;
3014 switch (kind) {
3015 case LS_cmp_exchange: {
3016 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3017 oldval, newval, value_type, type, decorators);
3018 break;
3019 }
3020 case LS_cmp_swap_weak:
3021 decorators |= C2_WEAK_CMPXCHG;
3022 case LS_cmp_swap: {
3023 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3024 oldval, newval, value_type, type, decorators);
3025 break;
3026 }
3027 case LS_get_set: {
3028 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3029 newval, value_type, type, decorators);
3030 break;
3031 }
3032 case LS_get_add: {
3033 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3034 newval, value_type, type, decorators);
3035 break;
3036 }
3037 default:
3038 ShouldNotReachHere();
3039 }
3040
3041 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3042 set_result(result);
3043 return true;
3044 }
3045
3046 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3047 // Regardless of form, don't allow previous ld/st to move down,
3048 // then issue acquire, release, or volatile mem_bar.
3049 insert_mem_bar(Op_MemBarCPUOrder);
3050 switch(id) {
3051 case vmIntrinsics::_loadFence:
3052 insert_mem_bar(Op_LoadFence);
3053 return true;
3054 case vmIntrinsics::_storeFence:
3055 insert_mem_bar(Op_StoreFence);
3056 return true;
3057 case vmIntrinsics::_storeStoreFence:
3058 insert_mem_bar(Op_StoreStoreFence);
3059 return true;
3060 case vmIntrinsics::_fullFence:
3061 insert_mem_bar(Op_MemBarVolatile);
3062 return true;
3063 default:
3064 fatal_unexpected_iid(id);
3065 return false;
3066 }
3067 }
3068
3069 bool LibraryCallKit::inline_onspinwait() {
3070 insert_mem_bar(Op_OnSpinWait);
3071 return true;
3072 }
3073
3074 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3075 if (!kls->is_Con()) {
3076 return true;
3077 }
3078 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3079 if (klsptr == nullptr) {
3080 return true;
3081 }
3082 ciInstanceKlass* ik = klsptr->instance_klass();
3083 // don't need a guard for a klass that is already initialized
3084 return !ik->is_initialized();
3085 }
3086
3087 //----------------------------inline_unsafe_writeback0-------------------------
3088 // public native void Unsafe.writeback0(long address)
3089 bool LibraryCallKit::inline_unsafe_writeback0() {
3090 if (!Matcher::has_match_rule(Op_CacheWB)) {
3091 return false;
3092 }
3093 #ifndef PRODUCT
3094 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3095 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3096 ciSignature* sig = callee()->signature();
3097 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3098 #endif
3099 null_check_receiver(); // null-check, then ignore
3100 Node *addr = argument(1);
3101 addr = new CastX2PNode(addr);
3102 addr = _gvn.transform(addr);
3103 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3104 flush = _gvn.transform(flush);
3105 set_memory(flush, TypeRawPtr::BOTTOM);
3106 return true;
3107 }
3108
3109 //----------------------------inline_unsafe_writeback0-------------------------
3110 // public native void Unsafe.writeback0(long address)
3111 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3112 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3113 return false;
3114 }
3115 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3116 return false;
3117 }
3118 #ifndef PRODUCT
3119 assert(Matcher::has_match_rule(Op_CacheWB),
3120 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3121 : "found match rule for CacheWBPostSync but not CacheWB"));
3122
3123 #endif
3124 null_check_receiver(); // null-check, then ignore
3125 Node *sync;
3126 if (is_pre) {
3127 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3128 } else {
3129 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3130 }
3131 sync = _gvn.transform(sync);
3132 set_memory(sync, TypeRawPtr::BOTTOM);
3133 return true;
3134 }
3135
3136 //----------------------------inline_unsafe_allocate---------------------------
3137 // public native Object Unsafe.allocateInstance(Class<?> cls);
3138 bool LibraryCallKit::inline_unsafe_allocate() {
3139
3140 #if INCLUDE_JVMTI
3141 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3142 return false;
3143 }
3144 #endif //INCLUDE_JVMTI
3145
3146 if (callee()->is_static()) return false; // caller must have the capability!
3147
3148 null_check_receiver(); // null-check, then ignore
3149 Node* cls = null_check(argument(1));
3150 if (stopped()) return true;
3151
3152 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3153 kls = null_check(kls);
3154 if (stopped()) return true; // argument was like int.class
3155
3156 #if INCLUDE_JVMTI
3157 // Don't try to access new allocated obj in the intrinsic.
3158 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3159 // Deoptimize and allocate in interpreter instead.
3160 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3161 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3162 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3163 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3164 {
3165 BuildCutout unless(this, tst, PROB_MAX);
3166 uncommon_trap(Deoptimization::Reason_intrinsic,
3167 Deoptimization::Action_make_not_entrant);
3168 }
3169 if (stopped()) {
3170 return true;
3171 }
3172 #endif //INCLUDE_JVMTI
3173
3174 Node* test = nullptr;
3175 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3176 // Note: The argument might still be an illegal value like
3177 // Serializable.class or Object[].class. The runtime will handle it.
3178 // But we must make an explicit check for initialization.
3179 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3180 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3181 // can generate code to load it as unsigned byte.
3182 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3183 Node* bits = intcon(InstanceKlass::fully_initialized);
3184 test = _gvn.transform(new SubINode(inst, bits));
3185 // The 'test' is non-zero if we need to take a slow path.
3186 }
3187 Node* obj = nullptr;
3188 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3189 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3190 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3191 } else {
3192 obj = new_instance(kls, test);
3193 }
3194 set_result(obj);
3195 return true;
3196 }
3197
3198 //------------------------inline_native_time_funcs--------------
3199 // inline code for System.currentTimeMillis() and System.nanoTime()
3200 // these have the same type and signature
3201 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3202 const TypeFunc* tf = OptoRuntime::void_long_Type();
3203 const TypePtr* no_memory_effects = nullptr;
3204 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3205 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3206 #ifdef ASSERT
3207 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3208 assert(value_top == top(), "second value must be top");
3209 #endif
3210 set_result(value);
3211 return true;
3212 }
3213
3214
3215 #if INCLUDE_JVMTI
3216
3217 // When notifications are disabled then just update the VTMS transition bit and return.
3218 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3219 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3220 if (!DoJVMTIVirtualThreadTransitions) {
3221 return true;
3222 }
3223 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3224 IdealKit ideal(this);
3225
3226 Node* ONE = ideal.ConI(1);
3227 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3228 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3229 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3230
3231 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3232 sync_kit(ideal);
3233 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3234 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3235 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3236 ideal.sync_kit(this);
3237 } ideal.else_(); {
3238 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3239 Node* thread = ideal.thread();
3240 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3241 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3242
3243 sync_kit(ideal);
3244 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3245 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3246
3247 ideal.sync_kit(this);
3248 } ideal.end_if();
3249 final_sync(ideal);
3250
3251 return true;
3252 }
3253
3254 // Always update the is_disable_suspend bit.
3255 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3256 if (!DoJVMTIVirtualThreadTransitions) {
3257 return true;
3258 }
3259 IdealKit ideal(this);
3260
3261 {
3262 // unconditionally update the is_disable_suspend bit in current JavaThread
3263 Node* thread = ideal.thread();
3264 Node* arg = _gvn.transform(argument(0)); // argument for notification
3265 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3266 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3267
3268 sync_kit(ideal);
3269 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3270 ideal.sync_kit(this);
3271 }
3272 final_sync(ideal);
3273
3274 return true;
3275 }
3276
3277 #endif // INCLUDE_JVMTI
3278
3279 #ifdef JFR_HAVE_INTRINSICS
3280
3281 /**
3282 * if oop->klass != null
3283 * // normal class
3284 * epoch = _epoch_state ? 2 : 1
3285 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3286 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3287 * }
3288 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3289 * else
3290 * // primitive class
3291 * if oop->array_klass != null
3292 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3293 * else
3294 * id = LAST_TYPE_ID + 1 // void class path
3295 * if (!signaled)
3296 * signaled = true
3297 */
3298 bool LibraryCallKit::inline_native_classID() {
3299 Node* cls = argument(0);
3300
3301 IdealKit ideal(this);
3302 #define __ ideal.
3303 IdealVariable result(ideal); __ declarations_done();
3304 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3305 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3306 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3307
3308
3309 __ if_then(kls, BoolTest::ne, null()); {
3310 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3311 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3312
3313 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3314 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3315 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3316 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3317 mask = _gvn.transform(new OrLNode(mask, epoch));
3318 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3319
3320 float unlikely = PROB_UNLIKELY(0.999);
3321 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3322 sync_kit(ideal);
3323 make_runtime_call(RC_LEAF,
3324 OptoRuntime::class_id_load_barrier_Type(),
3325 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3326 "class id load barrier",
3327 TypePtr::BOTTOM,
3328 kls);
3329 ideal.sync_kit(this);
3330 } __ end_if();
3331
3332 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3333 } __ else_(); {
3334 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3335 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3336 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3337 __ if_then(array_kls, BoolTest::ne, null()); {
3338 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3339 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3340 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3341 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3342 } __ else_(); {
3343 // void class case
3344 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3345 } __ end_if();
3346
3347 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3348 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3349 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3350 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3351 } __ end_if();
3352 } __ end_if();
3353
3354 final_sync(ideal);
3355 set_result(ideal.value(result));
3356 #undef __
3357 return true;
3358 }
3359
3360 //------------------------inline_native_jvm_commit------------------
3361 bool LibraryCallKit::inline_native_jvm_commit() {
3362 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3363
3364 // Save input memory and i_o state.
3365 Node* input_memory_state = reset_memory();
3366 set_all_memory(input_memory_state);
3367 Node* input_io_state = i_o();
3368
3369 // TLS.
3370 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3371 // Jfr java buffer.
3372 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3373 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3374 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3375
3376 // Load the current value of the notified field in the JfrThreadLocal.
3377 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3378 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3379
3380 // Test for notification.
3381 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3382 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3383 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3384
3385 // True branch, is notified.
3386 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3387 set_control(is_notified);
3388
3389 // Reset notified state.
3390 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3391 Node* notified_reset_memory = reset_memory();
3392
3393 // 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.
3394 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3395 // Convert the machine-word to a long.
3396 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3397
3398 // False branch, not notified.
3399 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3400 set_control(not_notified);
3401 set_all_memory(input_memory_state);
3402
3403 // Arg is the next position as a long.
3404 Node* arg = argument(0);
3405 // Convert long to machine-word.
3406 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3407
3408 // Store the next_position to the underlying jfr java buffer.
3409 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3410
3411 Node* commit_memory = reset_memory();
3412 set_all_memory(commit_memory);
3413
3414 // 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.
3415 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3416 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3417 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3418
3419 // And flags with lease constant.
3420 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3421
3422 // Branch on lease to conditionalize returning the leased java buffer.
3423 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3424 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3425 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3426
3427 // False branch, not a lease.
3428 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3429
3430 // True branch, is lease.
3431 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3432 set_control(is_lease);
3433
3434 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3435 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3436 OptoRuntime::void_void_Type(),
3437 SharedRuntime::jfr_return_lease(),
3438 "return_lease", TypePtr::BOTTOM);
3439 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3440
3441 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3442 record_for_igvn(lease_compare_rgn);
3443 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3444 record_for_igvn(lease_compare_mem);
3445 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3446 record_for_igvn(lease_compare_io);
3447 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3448 record_for_igvn(lease_result_value);
3449
3450 // Update control and phi nodes.
3451 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3452 lease_compare_rgn->init_req(_false_path, not_lease);
3453
3454 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3455 lease_compare_mem->init_req(_false_path, commit_memory);
3456
3457 lease_compare_io->init_req(_true_path, i_o());
3458 lease_compare_io->init_req(_false_path, input_io_state);
3459
3460 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3461 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3462
3463 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3464 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3465 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3466 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3467
3468 // Update control and phi nodes.
3469 result_rgn->init_req(_true_path, is_notified);
3470 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3471
3472 result_mem->init_req(_true_path, notified_reset_memory);
3473 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3474
3475 result_io->init_req(_true_path, input_io_state);
3476 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3477
3478 result_value->init_req(_true_path, current_pos);
3479 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3480
3481 // Set output state.
3482 set_control(_gvn.transform(result_rgn));
3483 set_all_memory(_gvn.transform(result_mem));
3484 set_i_o(_gvn.transform(result_io));
3485 set_result(result_rgn, result_value);
3486 return true;
3487 }
3488
3489 /*
3490 * The intrinsic is a model of this pseudo-code:
3491 *
3492 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3493 * jobject h_event_writer = tl->java_event_writer();
3494 * if (h_event_writer == nullptr) {
3495 * return nullptr;
3496 * }
3497 * oop threadObj = Thread::threadObj();
3498 * oop vthread = java_lang_Thread::vthread(threadObj);
3499 * traceid tid;
3500 * bool pinVirtualThread;
3501 * bool excluded;
3502 * if (vthread != threadObj) { // i.e. current thread is virtual
3503 * tid = java_lang_Thread::tid(vthread);
3504 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3505 * pinVirtualThread = VMContinuations;
3506 * excluded = vthread_epoch_raw & excluded_mask;
3507 * if (!excluded) {
3508 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3509 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3510 * if (vthread_epoch != current_epoch) {
3511 * write_checkpoint();
3512 * }
3513 * }
3514 * } else {
3515 * tid = java_lang_Thread::tid(threadObj);
3516 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3517 * pinVirtualThread = false;
3518 * excluded = thread_epoch_raw & excluded_mask;
3519 * }
3520 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3521 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3522 * if (tid_in_event_writer != tid) {
3523 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3524 * setField(event_writer, "excluded", excluded);
3525 * setField(event_writer, "threadID", tid);
3526 * }
3527 * return event_writer
3528 */
3529 bool LibraryCallKit::inline_native_getEventWriter() {
3530 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3531
3532 // Save input memory and i_o state.
3533 Node* input_memory_state = reset_memory();
3534 set_all_memory(input_memory_state);
3535 Node* input_io_state = i_o();
3536
3537 // The most significant bit of the u2 is used to denote thread exclusion
3538 Node* excluded_shift = _gvn.intcon(15);
3539 Node* excluded_mask = _gvn.intcon(1 << 15);
3540 // The epoch generation is the range [1-32767]
3541 Node* epoch_mask = _gvn.intcon(32767);
3542
3543 // TLS
3544 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3545
3546 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3547 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3548
3549 // Load the eventwriter jobject handle.
3550 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3551
3552 // Null check the jobject handle.
3553 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3554 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3555 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3556
3557 // False path, jobj is null.
3558 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3559
3560 // True path, jobj is not null.
3561 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3562
3563 set_control(jobj_is_not_null);
3564
3565 // Load the threadObj for the CarrierThread.
3566 Node* threadObj = generate_current_thread(tls_ptr);
3567
3568 // Load the vthread.
3569 Node* vthread = generate_virtual_thread(tls_ptr);
3570
3571 // If vthread != threadObj, this is a virtual thread.
3572 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3573 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3574 IfNode* iff_vthread_not_equal_threadObj =
3575 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3576
3577 // False branch, fallback to threadObj.
3578 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3579 set_control(vthread_equal_threadObj);
3580
3581 // Load the tid field from the vthread object.
3582 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3583
3584 // Load the raw epoch value from the threadObj.
3585 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3586 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3587 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3588 TypeInt::CHAR, T_CHAR,
3589 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3590
3591 // Mask off the excluded information from the epoch.
3592 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3593
3594 // True branch, this is a virtual thread.
3595 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3596 set_control(vthread_not_equal_threadObj);
3597
3598 // Load the tid field from the vthread object.
3599 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3600
3601 // Continuation support determines if a virtual thread should be pinned.
3602 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3603 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3604
3605 // Load the raw epoch value from the vthread.
3606 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3607 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3608 TypeInt::CHAR, T_CHAR,
3609 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3610
3611 // Mask off the excluded information from the epoch.
3612 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3613
3614 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3615 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3616 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3617 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3618
3619 // False branch, vthread is excluded, no need to write epoch info.
3620 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3621
3622 // True branch, vthread is included, update epoch info.
3623 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3624 set_control(included);
3625
3626 // Get epoch value.
3627 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3628
3629 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3630 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3631 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3632
3633 // Compare the epoch in the vthread to the current epoch generation.
3634 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3635 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3636 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3637
3638 // False path, epoch is equal, checkpoint information is valid.
3639 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3640
3641 // True path, epoch is not equal, write a checkpoint for the vthread.
3642 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3643
3644 set_control(epoch_is_not_equal);
3645
3646 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3647 // The call also updates the native thread local thread id and the vthread with the current epoch.
3648 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3649 OptoRuntime::jfr_write_checkpoint_Type(),
3650 SharedRuntime::jfr_write_checkpoint(),
3651 "write_checkpoint", TypePtr::BOTTOM);
3652 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3653
3654 // vthread epoch != current epoch
3655 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3656 record_for_igvn(epoch_compare_rgn);
3657 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3658 record_for_igvn(epoch_compare_mem);
3659 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3660 record_for_igvn(epoch_compare_io);
3661
3662 // Update control and phi nodes.
3663 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3664 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3665 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3666 epoch_compare_mem->init_req(_false_path, input_memory_state);
3667 epoch_compare_io->init_req(_true_path, i_o());
3668 epoch_compare_io->init_req(_false_path, input_io_state);
3669
3670 // excluded != true
3671 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3672 record_for_igvn(exclude_compare_rgn);
3673 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3674 record_for_igvn(exclude_compare_mem);
3675 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3676 record_for_igvn(exclude_compare_io);
3677
3678 // Update control and phi nodes.
3679 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3680 exclude_compare_rgn->init_req(_false_path, excluded);
3681 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3682 exclude_compare_mem->init_req(_false_path, input_memory_state);
3683 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3684 exclude_compare_io->init_req(_false_path, input_io_state);
3685
3686 // vthread != threadObj
3687 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3688 record_for_igvn(vthread_compare_rgn);
3689 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3690 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3691 record_for_igvn(vthread_compare_io);
3692 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3693 record_for_igvn(tid);
3694 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3695 record_for_igvn(exclusion);
3696 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3697 record_for_igvn(pinVirtualThread);
3698
3699 // Update control and phi nodes.
3700 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3701 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3702 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3703 vthread_compare_mem->init_req(_false_path, input_memory_state);
3704 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3705 vthread_compare_io->init_req(_false_path, input_io_state);
3706 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3707 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3708 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3709 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3710 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3711 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3712
3713 // Update branch state.
3714 set_control(_gvn.transform(vthread_compare_rgn));
3715 set_all_memory(_gvn.transform(vthread_compare_mem));
3716 set_i_o(_gvn.transform(vthread_compare_io));
3717
3718 // Load the event writer oop by dereferencing the jobject handle.
3719 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3720 assert(klass_EventWriter->is_loaded(), "invariant");
3721 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3722 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3723 const TypeOopPtr* const xtype = aklass->as_instance_type();
3724 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3725 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3726
3727 // Load the current thread id from the event writer object.
3728 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3729 // Get the field offset to, conditionally, store an updated tid value later.
3730 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3731 // Get the field offset to, conditionally, store an updated exclusion value later.
3732 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3733 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3734 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3735
3736 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3737 record_for_igvn(event_writer_tid_compare_rgn);
3738 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3739 record_for_igvn(event_writer_tid_compare_mem);
3740 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3741 record_for_igvn(event_writer_tid_compare_io);
3742
3743 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3744 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3745 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3746 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3747
3748 // False path, tids are the same.
3749 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3750
3751 // True path, tid is not equal, need to update the tid in the event writer.
3752 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3753 record_for_igvn(tid_is_not_equal);
3754
3755 // Store the pin state to the event writer.
3756 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3757
3758 // Store the exclusion state to the event writer.
3759 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3760 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3761
3762 // Store the tid to the event writer.
3763 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3764
3765 // Update control and phi nodes.
3766 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3767 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3768 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3769 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3770 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3771 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3772
3773 // Result of top level CFG, Memory, IO and Value.
3774 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3775 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3776 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3777 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3778
3779 // Result control.
3780 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3781 result_rgn->init_req(_false_path, jobj_is_null);
3782
3783 // Result memory.
3784 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3785 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3786
3787 // Result IO.
3788 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3789 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3790
3791 // Result value.
3792 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3793 result_value->init_req(_false_path, null()); // return null
3794
3795 // Set output state.
3796 set_control(_gvn.transform(result_rgn));
3797 set_all_memory(_gvn.transform(result_mem));
3798 set_i_o(_gvn.transform(result_io));
3799 set_result(result_rgn, result_value);
3800 return true;
3801 }
3802
3803 /*
3804 * The intrinsic is a model of this pseudo-code:
3805 *
3806 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3807 * if (carrierThread != thread) { // is virtual thread
3808 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3809 * bool excluded = vthread_epoch_raw & excluded_mask;
3810 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3811 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3812 * if (!excluded) {
3813 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3814 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3815 * }
3816 * Atomic::release_store(&tl->_vthread, true);
3817 * return;
3818 * }
3819 * Atomic::release_store(&tl->_vthread, false);
3820 */
3821 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3822 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3823
3824 Node* input_memory_state = reset_memory();
3825 set_all_memory(input_memory_state);
3826
3827 // The most significant bit of the u2 is used to denote thread exclusion
3828 Node* excluded_mask = _gvn.intcon(1 << 15);
3829 // The epoch generation is the range [1-32767]
3830 Node* epoch_mask = _gvn.intcon(32767);
3831
3832 Node* const carrierThread = generate_current_thread(jt);
3833 // If thread != carrierThread, this is a virtual thread.
3834 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3835 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3836 IfNode* iff_thread_not_equal_carrierThread =
3837 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3838
3839 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3840
3841 // False branch, is carrierThread.
3842 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3843 // Store release
3844 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3845
3846 set_all_memory(input_memory_state);
3847
3848 // True branch, is virtual thread.
3849 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3850 set_control(thread_not_equal_carrierThread);
3851
3852 // Load the raw epoch value from the vthread.
3853 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3854 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3855 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3856
3857 // Mask off the excluded information from the epoch.
3858 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3859
3860 // Load the tid field from the thread.
3861 Node* tid = load_field_from_object(thread, "tid", "J");
3862
3863 // Store the vthread tid to the jfr thread local.
3864 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3865 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3866
3867 // Branch is_excluded to conditionalize updating the epoch .
3868 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3869 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3870 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3871
3872 // True branch, vthread is excluded, no need to write epoch info.
3873 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3874 set_control(excluded);
3875 Node* vthread_is_excluded = _gvn.intcon(1);
3876
3877 // False branch, vthread is included, update epoch info.
3878 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3879 set_control(included);
3880 Node* vthread_is_included = _gvn.intcon(0);
3881
3882 // Get epoch value.
3883 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3884
3885 // Store the vthread epoch to the jfr thread local.
3886 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3887 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3888
3889 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3890 record_for_igvn(excluded_rgn);
3891 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3892 record_for_igvn(excluded_mem);
3893 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3894 record_for_igvn(exclusion);
3895
3896 // Merge the excluded control and memory.
3897 excluded_rgn->init_req(_true_path, excluded);
3898 excluded_rgn->init_req(_false_path, included);
3899 excluded_mem->init_req(_true_path, tid_memory);
3900 excluded_mem->init_req(_false_path, included_memory);
3901 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3902 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3903
3904 // Set intermediate state.
3905 set_control(_gvn.transform(excluded_rgn));
3906 set_all_memory(excluded_mem);
3907
3908 // Store the vthread exclusion state to the jfr thread local.
3909 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3910 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3911
3912 // Store release
3913 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3914
3915 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3916 record_for_igvn(thread_compare_rgn);
3917 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3918 record_for_igvn(thread_compare_mem);
3919 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3920 record_for_igvn(vthread);
3921
3922 // Merge the thread_compare control and memory.
3923 thread_compare_rgn->init_req(_true_path, control());
3924 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3925 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3926 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3927
3928 // Set output state.
3929 set_control(_gvn.transform(thread_compare_rgn));
3930 set_all_memory(_gvn.transform(thread_compare_mem));
3931 }
3932
3933 #endif // JFR_HAVE_INTRINSICS
3934
3935 //------------------------inline_native_currentCarrierThread------------------
3936 bool LibraryCallKit::inline_native_currentCarrierThread() {
3937 Node* junk = nullptr;
3938 set_result(generate_current_thread(junk));
3939 return true;
3940 }
3941
3942 //------------------------inline_native_currentThread------------------
3943 bool LibraryCallKit::inline_native_currentThread() {
3944 Node* junk = nullptr;
3945 set_result(generate_virtual_thread(junk));
3946 return true;
3947 }
3948
3949 //------------------------inline_native_setVthread------------------
3950 bool LibraryCallKit::inline_native_setCurrentThread() {
3951 assert(C->method()->changes_current_thread(),
3952 "method changes current Thread but is not annotated ChangesCurrentThread");
3953 Node* arr = argument(1);
3954 Node* thread = _gvn.transform(new ThreadLocalNode());
3955 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3956 Node* thread_obj_handle
3957 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3958 thread_obj_handle = _gvn.transform(thread_obj_handle);
3959 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3960 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3961
3962 // Change the _monitor_owner_id of the JavaThread
3963 Node* tid = load_field_from_object(arr, "tid", "J");
3964 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3965 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3966
3967 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3968 return true;
3969 }
3970
3971 const Type* LibraryCallKit::scopedValueCache_type() {
3972 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3973 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3974 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3975
3976 // Because we create the scopedValue cache lazily we have to make the
3977 // type of the result BotPTR.
3978 bool xk = etype->klass_is_exact();
3979 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3980 return objects_type;
3981 }
3982
3983 Node* LibraryCallKit::scopedValueCache_helper() {
3984 Node* thread = _gvn.transform(new ThreadLocalNode());
3985 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3986 // We cannot use immutable_memory() because we might flip onto a
3987 // different carrier thread, at which point we'll need to use that
3988 // carrier thread's cache.
3989 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3990 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3991 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3992 }
3993
3994 //------------------------inline_native_scopedValueCache------------------
3995 bool LibraryCallKit::inline_native_scopedValueCache() {
3996 Node* cache_obj_handle = scopedValueCache_helper();
3997 const Type* objects_type = scopedValueCache_type();
3998 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3999
4000 return true;
4001 }
4002
4003 //------------------------inline_native_setScopedValueCache------------------
4004 bool LibraryCallKit::inline_native_setScopedValueCache() {
4005 Node* arr = argument(0);
4006 Node* cache_obj_handle = scopedValueCache_helper();
4007 const Type* objects_type = scopedValueCache_type();
4008
4009 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4010 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4011
4012 return true;
4013 }
4014
4015 //------------------------inline_native_Continuation_pin and unpin-----------
4016
4017 // Shared implementation routine for both pin and unpin.
4018 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4019 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4020
4021 // Save input memory.
4022 Node* input_memory_state = reset_memory();
4023 set_all_memory(input_memory_state);
4024
4025 // TLS
4026 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4027 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4028 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4029
4030 // Null check the last continuation object.
4031 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4032 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4033 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4034
4035 // False path, last continuation is null.
4036 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4037
4038 // True path, last continuation is not null.
4039 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4040
4041 set_control(continuation_is_not_null);
4042
4043 // Load the pin count from the last continuation.
4044 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4045 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4046
4047 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4048 Node* pin_count_rhs;
4049 if (unpin) {
4050 pin_count_rhs = _gvn.intcon(0);
4051 } else {
4052 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4053 }
4054 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4055 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4056 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4057
4058 // True branch, pin count over/underflow.
4059 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4060 {
4061 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4062 // which will throw IllegalStateException for pin count over/underflow.
4063 // No memory changed so far - we can use memory create by reset_memory()
4064 // at the beginning of this intrinsic. No need to call reset_memory() again.
4065 PreserveJVMState pjvms(this);
4066 set_control(pin_count_over_underflow);
4067 uncommon_trap(Deoptimization::Reason_intrinsic,
4068 Deoptimization::Action_none);
4069 assert(stopped(), "invariant");
4070 }
4071
4072 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4073 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4074 set_control(valid_pin_count);
4075
4076 Node* next_pin_count;
4077 if (unpin) {
4078 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4079 } else {
4080 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4081 }
4082
4083 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4084
4085 // Result of top level CFG and Memory.
4086 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4087 record_for_igvn(result_rgn);
4088 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4089 record_for_igvn(result_mem);
4090
4091 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4092 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4093 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4094 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4095
4096 // Set output state.
4097 set_control(_gvn.transform(result_rgn));
4098 set_all_memory(_gvn.transform(result_mem));
4099
4100 return true;
4101 }
4102
4103 //-----------------------load_klass_from_mirror_common-------------------------
4104 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4105 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4106 // and branch to the given path on the region.
4107 // If never_see_null, take an uncommon trap on null, so we can optimistically
4108 // compile for the non-null case.
4109 // If the region is null, force never_see_null = true.
4110 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4111 bool never_see_null,
4112 RegionNode* region,
4113 int null_path,
4114 int offset) {
4115 if (region == nullptr) never_see_null = true;
4116 Node* p = basic_plus_adr(mirror, offset);
4117 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4118 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4119 Node* null_ctl = top();
4120 kls = null_check_oop(kls, &null_ctl, never_see_null);
4121 if (region != nullptr) {
4122 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4123 region->init_req(null_path, null_ctl);
4124 } else {
4125 assert(null_ctl == top(), "no loose ends");
4126 }
4127 return kls;
4128 }
4129
4130 //--------------------(inline_native_Class_query helpers)---------------------
4131 // Use this for JVM_ACC_INTERFACE.
4132 // Fall through if (mods & mask) == bits, take the guard otherwise.
4133 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4134 ByteSize offset, const Type* type, BasicType bt) {
4135 // Branch around if the given klass has the given modifier bit set.
4136 // Like generate_guard, adds a new path onto the region.
4137 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4138 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4139 Node* mask = intcon(modifier_mask);
4140 Node* bits = intcon(modifier_bits);
4141 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4142 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4143 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4144 return generate_fair_guard(bol, region);
4145 }
4146
4147 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4148 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4149 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4150 }
4151
4152 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4153 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4154 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4155 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4156 }
4157
4158 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4159 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4160 }
4161
4162 //-------------------------inline_native_Class_query-------------------
4163 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4164 const Type* return_type = TypeInt::BOOL;
4165 Node* prim_return_value = top(); // what happens if it's a primitive class?
4166 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4167 bool expect_prim = false; // most of these guys expect to work on refs
4168
4169 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4170
4171 Node* mirror = argument(0);
4172 Node* obj = top();
4173
4174 switch (id) {
4175 case vmIntrinsics::_isInstance:
4176 // nothing is an instance of a primitive type
4177 prim_return_value = intcon(0);
4178 obj = argument(1);
4179 break;
4180 case vmIntrinsics::_isHidden:
4181 prim_return_value = intcon(0);
4182 break;
4183 case vmIntrinsics::_getSuperclass:
4184 prim_return_value = null();
4185 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4186 break;
4187 case vmIntrinsics::_getClassAccessFlags:
4188 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4189 return_type = TypeInt::CHAR;
4190 break;
4191 default:
4192 fatal_unexpected_iid(id);
4193 break;
4194 }
4195
4196 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4197 if (mirror_con == nullptr) return false; // cannot happen?
4198
4199 #ifndef PRODUCT
4200 if (C->print_intrinsics() || C->print_inlining()) {
4201 ciType* k = mirror_con->java_mirror_type();
4202 if (k) {
4203 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4204 k->print_name();
4205 tty->cr();
4206 }
4207 }
4208 #endif
4209
4210 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4211 RegionNode* region = new RegionNode(PATH_LIMIT);
4212 record_for_igvn(region);
4213 PhiNode* phi = new PhiNode(region, return_type);
4214
4215 // The mirror will never be null of Reflection.getClassAccessFlags, however
4216 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4217 // if it is. See bug 4774291.
4218
4219 // For Reflection.getClassAccessFlags(), the null check occurs in
4220 // the wrong place; see inline_unsafe_access(), above, for a similar
4221 // situation.
4222 mirror = null_check(mirror);
4223 // If mirror or obj is dead, only null-path is taken.
4224 if (stopped()) return true;
4225
4226 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4227
4228 // Now load the mirror's klass metaobject, and null-check it.
4229 // Side-effects region with the control path if the klass is null.
4230 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4231 // If kls is null, we have a primitive mirror.
4232 phi->init_req(_prim_path, prim_return_value);
4233 if (stopped()) { set_result(region, phi); return true; }
4234 bool safe_for_replace = (region->in(_prim_path) == top());
4235
4236 Node* p; // handy temp
4237 Node* null_ctl;
4238
4239 // Now that we have the non-null klass, we can perform the real query.
4240 // For constant classes, the query will constant-fold in LoadNode::Value.
4241 Node* query_value = top();
4242 switch (id) {
4243 case vmIntrinsics::_isInstance:
4244 // nothing is an instance of a primitive type
4245 query_value = gen_instanceof(obj, kls, safe_for_replace);
4246 break;
4247
4248 case vmIntrinsics::_isHidden:
4249 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4250 if (generate_hidden_class_guard(kls, region) != nullptr)
4251 // A guard was added. If the guard is taken, it was an hidden class.
4252 phi->add_req(intcon(1));
4253 // If we fall through, it's a plain class.
4254 query_value = intcon(0);
4255 break;
4256
4257
4258 case vmIntrinsics::_getSuperclass:
4259 // The rules here are somewhat unfortunate, but we can still do better
4260 // with random logic than with a JNI call.
4261 // Interfaces store null or Object as _super, but must report null.
4262 // Arrays store an intermediate super as _super, but must report Object.
4263 // Other types can report the actual _super.
4264 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4265 if (generate_interface_guard(kls, region) != nullptr)
4266 // A guard was added. If the guard is taken, it was an interface.
4267 phi->add_req(null());
4268 if (generate_array_guard(kls, region) != nullptr)
4269 // A guard was added. If the guard is taken, it was an array.
4270 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4271 // If we fall through, it's a plain class. Get its _super.
4272 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4273 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4274 null_ctl = top();
4275 kls = null_check_oop(kls, &null_ctl);
4276 if (null_ctl != top()) {
4277 // If the guard is taken, Object.superClass is null (both klass and mirror).
4278 region->add_req(null_ctl);
4279 phi ->add_req(null());
4280 }
4281 if (!stopped()) {
4282 query_value = load_mirror_from_klass(kls);
4283 }
4284 break;
4285
4286 case vmIntrinsics::_getClassAccessFlags:
4287 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4288 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4289 break;
4290
4291 default:
4292 fatal_unexpected_iid(id);
4293 break;
4294 }
4295
4296 // Fall-through is the normal case of a query to a real class.
4297 phi->init_req(1, query_value);
4298 region->init_req(1, control());
4299
4300 C->set_has_split_ifs(true); // Has chance for split-if optimization
4301 set_result(region, phi);
4302 return true;
4303 }
4304
4305
4306 //-------------------------inline_Class_cast-------------------
4307 bool LibraryCallKit::inline_Class_cast() {
4308 Node* mirror = argument(0); // Class
4309 Node* obj = argument(1);
4310 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4311 if (mirror_con == nullptr) {
4312 return false; // dead path (mirror->is_top()).
4313 }
4314 if (obj == nullptr || obj->is_top()) {
4315 return false; // dead path
4316 }
4317 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4318
4319 // First, see if Class.cast() can be folded statically.
4320 // java_mirror_type() returns non-null for compile-time Class constants.
4321 bool is_null_free_array = false;
4322 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4323 if (tm != nullptr && tm->is_klass() &&
4324 tp != nullptr) {
4325 if (!tp->is_loaded()) {
4326 // Don't use intrinsic when class is not loaded.
4327 return false;
4328 } else {
4329 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4330 if (is_null_free_array) {
4331 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4332 }
4333 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4334 if (static_res == Compile::SSC_always_true) {
4335 // isInstance() is true - fold the code.
4336 set_result(obj);
4337 return true;
4338 } else if (static_res == Compile::SSC_always_false) {
4339 // Don't use intrinsic, have to throw ClassCastException.
4340 // If the reference is null, the non-intrinsic bytecode will
4341 // be optimized appropriately.
4342 return false;
4343 }
4344 }
4345 }
4346
4347 // Bailout intrinsic and do normal inlining if exception path is frequent.
4348 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4349 return false;
4350 }
4351
4352 // Generate dynamic checks.
4353 // Class.cast() is java implementation of _checkcast bytecode.
4354 // Do checkcast (Parse::do_checkcast()) optimizations here.
4355
4356 mirror = null_check(mirror);
4357 // If mirror is dead, only null-path is taken.
4358 if (stopped()) {
4359 return true;
4360 }
4361
4362 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4363 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4364 RegionNode* region = new RegionNode(PATH_LIMIT);
4365 record_for_igvn(region);
4366
4367 // Now load the mirror's klass metaobject, and null-check it.
4368 // If kls is null, we have a primitive mirror and
4369 // nothing is an instance of a primitive type.
4370 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4371
4372 Node* res = top();
4373 Node* io = i_o();
4374 Node* mem = merged_memory();
4375 if (!stopped()) {
4376
4377 Node* bad_type_ctrl = top();
4378 // Do checkcast optimizations.
4379 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4380 region->init_req(_bad_type_path, bad_type_ctrl);
4381 }
4382 if (region->in(_prim_path) != top() ||
4383 region->in(_bad_type_path) != top() ||
4384 region->in(_npe_path) != top()) {
4385 // Let Interpreter throw ClassCastException.
4386 PreserveJVMState pjvms(this);
4387 set_control(_gvn.transform(region));
4388 // Set IO and memory because gen_checkcast may override them when buffering inline types
4389 set_i_o(io);
4390 set_all_memory(mem);
4391 uncommon_trap(Deoptimization::Reason_intrinsic,
4392 Deoptimization::Action_maybe_recompile);
4393 }
4394 if (!stopped()) {
4395 set_result(res);
4396 }
4397 return true;
4398 }
4399
4400
4401 //--------------------------inline_native_subtype_check------------------------
4402 // This intrinsic takes the JNI calls out of the heart of
4403 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4404 bool LibraryCallKit::inline_native_subtype_check() {
4405 // Pull both arguments off the stack.
4406 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4407 args[0] = argument(0);
4408 args[1] = argument(1);
4409 Node* klasses[2]; // corresponding Klasses: superk, subk
4410 klasses[0] = klasses[1] = top();
4411
4412 enum {
4413 // A full decision tree on {superc is prim, subc is prim}:
4414 _prim_0_path = 1, // {P,N} => false
4415 // {P,P} & superc!=subc => false
4416 _prim_same_path, // {P,P} & superc==subc => true
4417 _prim_1_path, // {N,P} => false
4418 _ref_subtype_path, // {N,N} & subtype check wins => true
4419 _both_ref_path, // {N,N} & subtype check loses => false
4420 PATH_LIMIT
4421 };
4422
4423 RegionNode* region = new RegionNode(PATH_LIMIT);
4424 RegionNode* prim_region = new RegionNode(2);
4425 Node* phi = new PhiNode(region, TypeInt::BOOL);
4426 record_for_igvn(region);
4427 record_for_igvn(prim_region);
4428
4429 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4430 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4431 int class_klass_offset = java_lang_Class::klass_offset();
4432
4433 // First null-check both mirrors and load each mirror's klass metaobject.
4434 int which_arg;
4435 for (which_arg = 0; which_arg <= 1; which_arg++) {
4436 Node* arg = args[which_arg];
4437 arg = null_check(arg);
4438 if (stopped()) break;
4439 args[which_arg] = arg;
4440
4441 Node* p = basic_plus_adr(arg, class_klass_offset);
4442 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4443 klasses[which_arg] = _gvn.transform(kls);
4444 }
4445
4446 // Having loaded both klasses, test each for null.
4447 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4448 for (which_arg = 0; which_arg <= 1; which_arg++) {
4449 Node* kls = klasses[which_arg];
4450 Node* null_ctl = top();
4451 kls = null_check_oop(kls, &null_ctl, never_see_null);
4452 if (which_arg == 0) {
4453 prim_region->init_req(1, null_ctl);
4454 } else {
4455 region->init_req(_prim_1_path, null_ctl);
4456 }
4457 if (stopped()) break;
4458 klasses[which_arg] = kls;
4459 }
4460
4461 if (!stopped()) {
4462 // now we have two reference types, in klasses[0..1]
4463 Node* subk = klasses[1]; // the argument to isAssignableFrom
4464 Node* superk = klasses[0]; // the receiver
4465 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4466 region->set_req(_ref_subtype_path, control());
4467 }
4468
4469 // If both operands are primitive (both klasses null), then
4470 // we must return true when they are identical primitives.
4471 // It is convenient to test this after the first null klass check.
4472 // This path is also used if superc is a value mirror.
4473 set_control(_gvn.transform(prim_region));
4474 if (!stopped()) {
4475 // Since superc is primitive, make a guard for the superc==subc case.
4476 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4477 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4478 generate_fair_guard(bol_eq, region);
4479 if (region->req() == PATH_LIMIT+1) {
4480 // A guard was added. If the added guard is taken, superc==subc.
4481 region->swap_edges(PATH_LIMIT, _prim_same_path);
4482 region->del_req(PATH_LIMIT);
4483 }
4484 region->set_req(_prim_0_path, control()); // Not equal after all.
4485 }
4486
4487 // these are the only paths that produce 'true':
4488 phi->set_req(_prim_same_path, intcon(1));
4489 phi->set_req(_ref_subtype_path, intcon(1));
4490
4491 // pull together the cases:
4492 assert(region->req() == PATH_LIMIT, "sane region");
4493 for (uint i = 1; i < region->req(); i++) {
4494 Node* ctl = region->in(i);
4495 if (ctl == nullptr || ctl == top()) {
4496 region->set_req(i, top());
4497 phi ->set_req(i, top());
4498 } else if (phi->in(i) == nullptr) {
4499 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4500 }
4501 }
4502
4503 set_control(_gvn.transform(region));
4504 set_result(_gvn.transform(phi));
4505 return true;
4506 }
4507
4508 //---------------------generate_array_guard_common------------------------
4509 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4510
4511 if (stopped()) {
4512 return nullptr;
4513 }
4514
4515 // Like generate_guard, adds a new path onto the region.
4516 jint layout_con = 0;
4517 Node* layout_val = get_layout_helper(kls, layout_con);
4518 if (layout_val == nullptr) {
4519 bool query = 0;
4520 switch(kind) {
4521 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4522 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4523 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4524 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4525 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4526 default:
4527 ShouldNotReachHere();
4528 }
4529 if (!query) {
4530 return nullptr; // never a branch
4531 } else { // always a branch
4532 Node* always_branch = control();
4533 if (region != nullptr)
4534 region->add_req(always_branch);
4535 set_control(top());
4536 return always_branch;
4537 }
4538 }
4539 unsigned int value = 0;
4540 BoolTest::mask btest = BoolTest::illegal;
4541 switch(kind) {
4542 case ObjectArray:
4543 case NonObjectArray: {
4544 value = Klass::_lh_array_tag_obj_value;
4545 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4546 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4547 break;
4548 }
4549 case TypeArray: {
4550 value = Klass::_lh_array_tag_type_value;
4551 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4552 btest = BoolTest::eq;
4553 break;
4554 }
4555 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4556 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4557 default:
4558 ShouldNotReachHere();
4559 }
4560 // Now test the correct condition.
4561 jint nval = (jint)value;
4562 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4563 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4564 Node* ctrl = generate_fair_guard(bol, region);
4565 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4566 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4567 // Keep track of the fact that 'obj' is an array to prevent
4568 // array specific accesses from floating above the guard.
4569 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4570 }
4571 return ctrl;
4572 }
4573
4574 // public static native Object[] newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4575 // public static native Object[] newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4576 // public static native Object[] newNullableAtomicArray(Class<?> componentType, int length);
4577 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4578 assert(null_free || atomic, "nullable implies atomic");
4579 Node* componentType = argument(0);
4580 Node* length = argument(1);
4581 Node* init_val = null_free ? argument(2) : nullptr;
4582
4583 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4584 if (tp != nullptr) {
4585 ciInstanceKlass* ik = tp->instance_klass();
4586 if (ik == C->env()->Class_klass()) {
4587 ciType* t = tp->java_mirror_type();
4588 if (t != nullptr && t->is_inlinetype()) {
4589 ciInlineKlass* vk = t->as_inline_klass();
4590 bool flat = vk->maybe_flat_in_array();
4591 if (flat && atomic) {
4592 // Only flat if we have a corresponding atomic layout
4593 flat = null_free ? vk->has_atomic_layout() : vk->has_nullable_atomic_layout();
4594 }
4595 // TODO 8350865 refactor
4596 if (flat && !atomic) {
4597 flat = vk->has_non_atomic_layout();
4598 }
4599
4600 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4601 if (UseZGC && null_free && !flat) {
4602 return false;
4603 }
4604
4605 ciArrayKlass* array_klass = ciArrayKlass::make(t, flat, null_free, atomic);
4606 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4607 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4608 if (null_free) {
4609 if (init_val->is_InlineType()) {
4610 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4611 // Zeroing is enough because the init value is the all-zero value
4612 init_val = nullptr;
4613 } else {
4614 init_val = init_val->as_InlineType()->buffer(this);
4615 }
4616 }
4617 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4618 }
4619 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4620 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4621 assert(arytype->is_null_free() == null_free, "inconsistency");
4622 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4623 assert(arytype->is_flat() == flat, "inconsistency");
4624 assert(arytype->is_aryptr()->is_not_flat() == !flat, "inconsistency");
4625 set_result(obj);
4626 return true;
4627 }
4628 }
4629 }
4630 }
4631 return false;
4632 }
4633
4634 //-----------------------inline_native_newArray--------------------------
4635 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4636 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4637 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4638 Node* mirror;
4639 Node* count_val;
4640 if (uninitialized) {
4641 null_check_receiver();
4642 mirror = argument(1);
4643 count_val = argument(2);
4644 } else {
4645 mirror = argument(0);
4646 count_val = argument(1);
4647 }
4648
4649 mirror = null_check(mirror);
4650 // If mirror or obj is dead, only null-path is taken.
4651 if (stopped()) return true;
4652
4653 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4654 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4655 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4656 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4657 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4658
4659 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4660 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4661 result_reg, _slow_path);
4662 Node* normal_ctl = control();
4663 Node* no_array_ctl = result_reg->in(_slow_path);
4664
4665 // Generate code for the slow case. We make a call to newArray().
4666 set_control(no_array_ctl);
4667 if (!stopped()) {
4668 // Either the input type is void.class, or else the
4669 // array klass has not yet been cached. Either the
4670 // ensuing call will throw an exception, or else it
4671 // will cache the array klass for next time.
4672 PreserveJVMState pjvms(this);
4673 CallJavaNode* slow_call = nullptr;
4674 if (uninitialized) {
4675 // Generate optimized virtual call (holder class 'Unsafe' is final)
4676 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4677 } else {
4678 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4679 }
4680 Node* slow_result = set_results_for_java_call(slow_call);
4681 // this->control() comes from set_results_for_java_call
4682 result_reg->set_req(_slow_path, control());
4683 result_val->set_req(_slow_path, slow_result);
4684 result_io ->set_req(_slow_path, i_o());
4685 result_mem->set_req(_slow_path, reset_memory());
4686 }
4687
4688 set_control(normal_ctl);
4689 if (!stopped()) {
4690 // Normal case: The array type has been cached in the java.lang.Class.
4691 // The following call works fine even if the array type is polymorphic.
4692 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4693 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4694 result_reg->init_req(_normal_path, control());
4695 result_val->init_req(_normal_path, obj);
4696 result_io ->init_req(_normal_path, i_o());
4697 result_mem->init_req(_normal_path, reset_memory());
4698
4699 if (uninitialized) {
4700 // Mark the allocation so that zeroing is skipped
4701 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4702 alloc->maybe_set_complete(&_gvn);
4703 }
4704 }
4705
4706 // Return the combined state.
4707 set_i_o( _gvn.transform(result_io) );
4708 set_all_memory( _gvn.transform(result_mem));
4709
4710 C->set_has_split_ifs(true); // Has chance for split-if optimization
4711 set_result(result_reg, result_val);
4712 return true;
4713 }
4714
4715 //----------------------inline_native_getLength--------------------------
4716 // public static native int java.lang.reflect.Array.getLength(Object array);
4717 bool LibraryCallKit::inline_native_getLength() {
4718 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4719
4720 Node* array = null_check(argument(0));
4721 // If array is dead, only null-path is taken.
4722 if (stopped()) return true;
4723
4724 // Deoptimize if it is a non-array.
4725 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4726
4727 if (non_array != nullptr) {
4728 PreserveJVMState pjvms(this);
4729 set_control(non_array);
4730 uncommon_trap(Deoptimization::Reason_intrinsic,
4731 Deoptimization::Action_maybe_recompile);
4732 }
4733
4734 // If control is dead, only non-array-path is taken.
4735 if (stopped()) return true;
4736
4737 // The works fine even if the array type is polymorphic.
4738 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4739 Node* result = load_array_length(array);
4740
4741 C->set_has_split_ifs(true); // Has chance for split-if optimization
4742 set_result(result);
4743 return true;
4744 }
4745
4746 //------------------------inline_array_copyOf----------------------------
4747 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4748 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4749 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4750 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4751
4752 // Get the arguments.
4753 Node* original = argument(0);
4754 Node* start = is_copyOfRange? argument(1): intcon(0);
4755 Node* end = is_copyOfRange? argument(2): argument(1);
4756 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4757
4758 Node* newcopy = nullptr;
4759
4760 // Set the original stack and the reexecute bit for the interpreter to reexecute
4761 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4762 { PreserveReexecuteState preexecs(this);
4763 jvms()->set_should_reexecute(true);
4764
4765 array_type_mirror = null_check(array_type_mirror);
4766 original = null_check(original);
4767
4768 // Check if a null path was taken unconditionally.
4769 if (stopped()) return true;
4770
4771 Node* orig_length = load_array_length(original);
4772
4773 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4774 klass_node = null_check(klass_node);
4775
4776 RegionNode* bailout = new RegionNode(1);
4777 record_for_igvn(bailout);
4778
4779 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4780 // Bail out if that is so.
4781 // Inline type array may have object field that would require a
4782 // write barrier. Conservatively, go to slow path.
4783 // TODO 8251971: Optimize for the case when flat src/dst are later found
4784 // to not contain oops (i.e., move this check to the macro expansion phase).
4785 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4786 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4787 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4788 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4789 // Can src array be flat and contain oops?
4790 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4791 // Can dest array be flat and contain oops?
4792 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4793 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4794 if (not_objArray != nullptr) {
4795 // Improve the klass node's type from the new optimistic assumption:
4796 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4797 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4798 Node* cast = new CastPPNode(control(), klass_node, akls);
4799 klass_node = _gvn.transform(cast);
4800 }
4801
4802 // Bail out if either start or end is negative.
4803 generate_negative_guard(start, bailout, &start);
4804 generate_negative_guard(end, bailout, &end);
4805
4806 Node* length = end;
4807 if (_gvn.type(start) != TypeInt::ZERO) {
4808 length = _gvn.transform(new SubINode(end, start));
4809 }
4810
4811 // Bail out if length is negative (i.e., if start > end).
4812 // Without this the new_array would throw
4813 // NegativeArraySizeException but IllegalArgumentException is what
4814 // should be thrown
4815 generate_negative_guard(length, bailout, &length);
4816
4817 // Handle inline type arrays
4818 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4819 if (!stopped()) {
4820 // TODO JDK-8329224
4821 if (!orig_t->is_null_free()) {
4822 // Not statically known to be null free, add a check
4823 generate_fair_guard(null_free_array_test(original), bailout);
4824 }
4825 orig_t = _gvn.type(original)->isa_aryptr();
4826 if (orig_t != nullptr && orig_t->is_flat()) {
4827 // Src is flat, check that dest is flat as well
4828 if (exclude_flat) {
4829 // Dest can't be flat, bail out
4830 bailout->add_req(control());
4831 set_control(top());
4832 } else {
4833 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4834 }
4835 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
4836 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4837 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4838 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4839 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4840 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4841 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4842 if (orig_t != nullptr) {
4843 orig_t = orig_t->cast_to_not_flat();
4844 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4845 }
4846 }
4847 if (!can_validate) {
4848 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4849 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4850 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4851 generate_fair_guard(flat_array_test(klass_node), bailout);
4852 generate_fair_guard(null_free_array_test(original), bailout);
4853 }
4854 }
4855
4856 // Bail out if start is larger than the original length
4857 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4858 generate_negative_guard(orig_tail, bailout, &orig_tail);
4859
4860 if (bailout->req() > 1) {
4861 PreserveJVMState pjvms(this);
4862 set_control(_gvn.transform(bailout));
4863 uncommon_trap(Deoptimization::Reason_intrinsic,
4864 Deoptimization::Action_maybe_recompile);
4865 }
4866
4867 if (!stopped()) {
4868 // How many elements will we copy from the original?
4869 // The answer is MinI(orig_tail, length).
4870 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4871
4872 // Generate a direct call to the right arraycopy function(s).
4873 // We know the copy is disjoint but we might not know if the
4874 // oop stores need checking.
4875 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4876 // This will fail a store-check if x contains any non-nulls.
4877
4878 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4879 // loads/stores but it is legal only if we're sure the
4880 // Arrays.copyOf would succeed. So we need all input arguments
4881 // to the copyOf to be validated, including that the copy to the
4882 // new array won't trigger an ArrayStoreException. That subtype
4883 // check can be optimized if we know something on the type of
4884 // the input array from type speculation.
4885 if (_gvn.type(klass_node)->singleton()) {
4886 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4887 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4888
4889 int test = C->static_subtype_check(superk, subk);
4890 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4891 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4892 if (t_original->speculative_type() != nullptr) {
4893 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4894 }
4895 }
4896 }
4897
4898 bool validated = false;
4899 // Reason_class_check rather than Reason_intrinsic because we
4900 // want to intrinsify even if this traps.
4901 if (can_validate) {
4902 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4903
4904 if (not_subtype_ctrl != top()) {
4905 PreserveJVMState pjvms(this);
4906 set_control(not_subtype_ctrl);
4907 uncommon_trap(Deoptimization::Reason_class_check,
4908 Deoptimization::Action_make_not_entrant);
4909 assert(stopped(), "Should be stopped");
4910 }
4911 validated = true;
4912 }
4913
4914 if (!stopped()) {
4915 newcopy = new_array(klass_node, length, 0); // no arguments to push
4916
4917 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4918 load_object_klass(original), klass_node);
4919 if (!is_copyOfRange) {
4920 ac->set_copyof(validated);
4921 } else {
4922 ac->set_copyofrange(validated);
4923 }
4924 Node* n = _gvn.transform(ac);
4925 if (n == ac) {
4926 ac->connect_outputs(this);
4927 } else {
4928 assert(validated, "shouldn't transform if all arguments not validated");
4929 set_all_memory(n);
4930 }
4931 }
4932 }
4933 } // original reexecute is set back here
4934
4935 C->set_has_split_ifs(true); // Has chance for split-if optimization
4936 if (!stopped()) {
4937 set_result(newcopy);
4938 }
4939 return true;
4940 }
4941
4942
4943 //----------------------generate_virtual_guard---------------------------
4944 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4945 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4946 RegionNode* slow_region) {
4947 ciMethod* method = callee();
4948 int vtable_index = method->vtable_index();
4949 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4950 "bad index %d", vtable_index);
4951 // Get the Method* out of the appropriate vtable entry.
4952 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4953 vtable_index*vtableEntry::size_in_bytes() +
4954 in_bytes(vtableEntry::method_offset());
4955 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4956 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4957
4958 // Compare the target method with the expected method (e.g., Object.hashCode).
4959 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4960
4961 Node* native_call = makecon(native_call_addr);
4962 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4963 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4964
4965 return generate_slow_guard(test_native, slow_region);
4966 }
4967
4968 //-----------------------generate_method_call----------------------------
4969 // Use generate_method_call to make a slow-call to the real
4970 // method if the fast path fails. An alternative would be to
4971 // use a stub like OptoRuntime::slow_arraycopy_Java.
4972 // This only works for expanding the current library call,
4973 // not another intrinsic. (E.g., don't use this for making an
4974 // arraycopy call inside of the copyOf intrinsic.)
4975 CallJavaNode*
4976 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4977 // When compiling the intrinsic method itself, do not use this technique.
4978 guarantee(callee() != C->method(), "cannot make slow-call to self");
4979
4980 ciMethod* method = callee();
4981 // ensure the JVMS we have will be correct for this call
4982 guarantee(method_id == method->intrinsic_id(), "must match");
4983
4984 const TypeFunc* tf = TypeFunc::make(method);
4985 if (res_not_null) {
4986 assert(tf->return_type() == T_OBJECT, "");
4987 const TypeTuple* range = tf->range_cc();
4988 const Type** fields = TypeTuple::fields(range->cnt());
4989 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4990 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4991 tf = TypeFunc::make(tf->domain_cc(), new_range);
4992 }
4993 CallJavaNode* slow_call;
4994 if (is_static) {
4995 assert(!is_virtual, "");
4996 slow_call = new CallStaticJavaNode(C, tf,
4997 SharedRuntime::get_resolve_static_call_stub(), method);
4998 } else if (is_virtual) {
4999 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5000 int vtable_index = Method::invalid_vtable_index;
5001 if (UseInlineCaches) {
5002 // Suppress the vtable call
5003 } else {
5004 // hashCode and clone are not a miranda methods,
5005 // so the vtable index is fixed.
5006 // No need to use the linkResolver to get it.
5007 vtable_index = method->vtable_index();
5008 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5009 "bad index %d", vtable_index);
5010 }
5011 slow_call = new CallDynamicJavaNode(tf,
5012 SharedRuntime::get_resolve_virtual_call_stub(),
5013 method, vtable_index);
5014 } else { // neither virtual nor static: opt_virtual
5015 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5016 slow_call = new CallStaticJavaNode(C, tf,
5017 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5018 slow_call->set_optimized_virtual(true);
5019 }
5020 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5021 // To be able to issue a direct call (optimized virtual or virtual)
5022 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5023 // about the method being invoked should be attached to the call site to
5024 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5025 slow_call->set_override_symbolic_info(true);
5026 }
5027 set_arguments_for_java_call(slow_call);
5028 set_edges_for_java_call(slow_call);
5029 return slow_call;
5030 }
5031
5032
5033 /**
5034 * Build special case code for calls to hashCode on an object. This call may
5035 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5036 * slightly different code.
5037 */
5038 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5039 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5040 assert(!(is_virtual && is_static), "either virtual, special, or static");
5041
5042 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5043
5044 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5045 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5046 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5047 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5048 Node* obj = argument(0);
5049
5050 // Don't intrinsify hashcode on inline types for now.
5051 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
5052 if (gvn().type(obj)->is_inlinetypeptr()) {
5053 return false;
5054 }
5055
5056 if (!is_static) {
5057 // Check for hashing null object
5058 obj = null_check_receiver();
5059 if (stopped()) return true; // unconditionally null
5060 result_reg->init_req(_null_path, top());
5061 result_val->init_req(_null_path, top());
5062 } else {
5063 // Do a null check, and return zero if null.
5064 // System.identityHashCode(null) == 0
5065 Node* null_ctl = top();
5066 obj = null_check_oop(obj, &null_ctl);
5067 result_reg->init_req(_null_path, null_ctl);
5068 result_val->init_req(_null_path, _gvn.intcon(0));
5069 }
5070
5071 // Unconditionally null? Then return right away.
5072 if (stopped()) {
5073 set_control( result_reg->in(_null_path));
5074 if (!stopped())
5075 set_result(result_val->in(_null_path));
5076 return true;
5077 }
5078
5079 // We only go to the fast case code if we pass a number of guards. The
5080 // paths which do not pass are accumulated in the slow_region.
5081 RegionNode* slow_region = new RegionNode(1);
5082 record_for_igvn(slow_region);
5083
5084 // If this is a virtual call, we generate a funny guard. We pull out
5085 // the vtable entry corresponding to hashCode() from the target object.
5086 // If the target method which we are calling happens to be the native
5087 // Object hashCode() method, we pass the guard. We do not need this
5088 // guard for non-virtual calls -- the caller is known to be the native
5089 // Object hashCode().
5090 if (is_virtual) {
5091 // After null check, get the object's klass.
5092 Node* obj_klass = load_object_klass(obj);
5093 generate_virtual_guard(obj_klass, slow_region);
5094 }
5095
5096 // Get the header out of the object, use LoadMarkNode when available
5097 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5098 // The control of the load must be null. Otherwise, the load can move before
5099 // the null check after castPP removal.
5100 Node* no_ctrl = nullptr;
5101 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5102
5103 if (!UseObjectMonitorTable) {
5104 // Test the header to see if it is safe to read w.r.t. locking.
5105 // This also serves as guard against inline types
5106 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5107 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5108 if (LockingMode == LM_LIGHTWEIGHT) {
5109 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5110 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5111 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5112
5113 generate_slow_guard(test_monitor, slow_region);
5114 } else {
5115 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5116 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5117 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5118
5119 generate_slow_guard(test_not_unlocked, slow_region);
5120 }
5121 }
5122
5123 // Get the hash value and check to see that it has been properly assigned.
5124 // We depend on hash_mask being at most 32 bits and avoid the use of
5125 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5126 // vm: see markWord.hpp.
5127 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5128 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5129 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5130 // This hack lets the hash bits live anywhere in the mark object now, as long
5131 // as the shift drops the relevant bits into the low 32 bits. Note that
5132 // Java spec says that HashCode is an int so there's no point in capturing
5133 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5134 hshifted_header = ConvX2I(hshifted_header);
5135 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5136
5137 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5138 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5139 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5140
5141 generate_slow_guard(test_assigned, slow_region);
5142
5143 Node* init_mem = reset_memory();
5144 // fill in the rest of the null path:
5145 result_io ->init_req(_null_path, i_o());
5146 result_mem->init_req(_null_path, init_mem);
5147
5148 result_val->init_req(_fast_path, hash_val);
5149 result_reg->init_req(_fast_path, control());
5150 result_io ->init_req(_fast_path, i_o());
5151 result_mem->init_req(_fast_path, init_mem);
5152
5153 // Generate code for the slow case. We make a call to hashCode().
5154 set_control(_gvn.transform(slow_region));
5155 if (!stopped()) {
5156 // No need for PreserveJVMState, because we're using up the present state.
5157 set_all_memory(init_mem);
5158 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5159 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5160 Node* slow_result = set_results_for_java_call(slow_call);
5161 // this->control() comes from set_results_for_java_call
5162 result_reg->init_req(_slow_path, control());
5163 result_val->init_req(_slow_path, slow_result);
5164 result_io ->set_req(_slow_path, i_o());
5165 result_mem ->set_req(_slow_path, reset_memory());
5166 }
5167
5168 // Return the combined state.
5169 set_i_o( _gvn.transform(result_io) );
5170 set_all_memory( _gvn.transform(result_mem));
5171
5172 set_result(result_reg, result_val);
5173 return true;
5174 }
5175
5176 //---------------------------inline_native_getClass----------------------------
5177 // public final native Class<?> java.lang.Object.getClass();
5178 //
5179 // Build special case code for calls to getClass on an object.
5180 bool LibraryCallKit::inline_native_getClass() {
5181 Node* obj = argument(0);
5182 if (obj->is_InlineType()) {
5183 const Type* t = _gvn.type(obj);
5184 if (t->maybe_null()) {
5185 null_check(obj);
5186 }
5187 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5188 return true;
5189 }
5190 obj = null_check_receiver();
5191 if (stopped()) return true;
5192 set_result(load_mirror_from_klass(load_object_klass(obj)));
5193 return true;
5194 }
5195
5196 //-----------------inline_native_Reflection_getCallerClass---------------------
5197 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5198 //
5199 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5200 //
5201 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5202 // in that it must skip particular security frames and checks for
5203 // caller sensitive methods.
5204 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5205 #ifndef PRODUCT
5206 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5207 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5208 }
5209 #endif
5210
5211 if (!jvms()->has_method()) {
5212 #ifndef PRODUCT
5213 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5214 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5215 }
5216 #endif
5217 return false;
5218 }
5219
5220 // Walk back up the JVM state to find the caller at the required
5221 // depth.
5222 JVMState* caller_jvms = jvms();
5223
5224 // Cf. JVM_GetCallerClass
5225 // NOTE: Start the loop at depth 1 because the current JVM state does
5226 // not include the Reflection.getCallerClass() frame.
5227 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5228 ciMethod* m = caller_jvms->method();
5229 switch (n) {
5230 case 0:
5231 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5232 break;
5233 case 1:
5234 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5235 if (!m->caller_sensitive()) {
5236 #ifndef PRODUCT
5237 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5238 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5239 }
5240 #endif
5241 return false; // bail-out; let JVM_GetCallerClass do the work
5242 }
5243 break;
5244 default:
5245 if (!m->is_ignored_by_security_stack_walk()) {
5246 // We have reached the desired frame; return the holder class.
5247 // Acquire method holder as java.lang.Class and push as constant.
5248 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5249 ciInstance* caller_mirror = caller_klass->java_mirror();
5250 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5251
5252 #ifndef PRODUCT
5253 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5254 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());
5255 tty->print_cr(" JVM state at this point:");
5256 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5257 ciMethod* m = jvms()->of_depth(i)->method();
5258 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5259 }
5260 }
5261 #endif
5262 return true;
5263 }
5264 break;
5265 }
5266 }
5267
5268 #ifndef PRODUCT
5269 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5270 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5271 tty->print_cr(" JVM state at this point:");
5272 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5273 ciMethod* m = jvms()->of_depth(i)->method();
5274 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5275 }
5276 }
5277 #endif
5278
5279 return false; // bail-out; let JVM_GetCallerClass do the work
5280 }
5281
5282 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5283 Node* arg = argument(0);
5284 Node* result = nullptr;
5285
5286 switch (id) {
5287 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5288 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5289 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5290 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5291 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5292 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5293
5294 case vmIntrinsics::_doubleToLongBits: {
5295 // two paths (plus control) merge in a wood
5296 RegionNode *r = new RegionNode(3);
5297 Node *phi = new PhiNode(r, TypeLong::LONG);
5298
5299 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5300 // Build the boolean node
5301 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5302
5303 // Branch either way.
5304 // NaN case is less traveled, which makes all the difference.
5305 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5306 Node *opt_isnan = _gvn.transform(ifisnan);
5307 assert( opt_isnan->is_If(), "Expect an IfNode");
5308 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5309 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5310
5311 set_control(iftrue);
5312
5313 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5314 Node *slow_result = longcon(nan_bits); // return NaN
5315 phi->init_req(1, _gvn.transform( slow_result ));
5316 r->init_req(1, iftrue);
5317
5318 // Else fall through
5319 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5320 set_control(iffalse);
5321
5322 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5323 r->init_req(2, iffalse);
5324
5325 // Post merge
5326 set_control(_gvn.transform(r));
5327 record_for_igvn(r);
5328
5329 C->set_has_split_ifs(true); // Has chance for split-if optimization
5330 result = phi;
5331 assert(result->bottom_type()->isa_long(), "must be");
5332 break;
5333 }
5334
5335 case vmIntrinsics::_floatToIntBits: {
5336 // two paths (plus control) merge in a wood
5337 RegionNode *r = new RegionNode(3);
5338 Node *phi = new PhiNode(r, TypeInt::INT);
5339
5340 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5341 // Build the boolean node
5342 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5343
5344 // Branch either way.
5345 // NaN case is less traveled, which makes all the difference.
5346 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5347 Node *opt_isnan = _gvn.transform(ifisnan);
5348 assert( opt_isnan->is_If(), "Expect an IfNode");
5349 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5350 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5351
5352 set_control(iftrue);
5353
5354 static const jint nan_bits = 0x7fc00000;
5355 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5356 phi->init_req(1, _gvn.transform( slow_result ));
5357 r->init_req(1, iftrue);
5358
5359 // Else fall through
5360 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5361 set_control(iffalse);
5362
5363 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5364 r->init_req(2, iffalse);
5365
5366 // Post merge
5367 set_control(_gvn.transform(r));
5368 record_for_igvn(r);
5369
5370 C->set_has_split_ifs(true); // Has chance for split-if optimization
5371 result = phi;
5372 assert(result->bottom_type()->isa_int(), "must be");
5373 break;
5374 }
5375
5376 default:
5377 fatal_unexpected_iid(id);
5378 break;
5379 }
5380 set_result(_gvn.transform(result));
5381 return true;
5382 }
5383
5384 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5385 Node* arg = argument(0);
5386 Node* result = nullptr;
5387
5388 switch (id) {
5389 case vmIntrinsics::_floatIsInfinite:
5390 result = new IsInfiniteFNode(arg);
5391 break;
5392 case vmIntrinsics::_floatIsFinite:
5393 result = new IsFiniteFNode(arg);
5394 break;
5395 case vmIntrinsics::_doubleIsInfinite:
5396 result = new IsInfiniteDNode(arg);
5397 break;
5398 case vmIntrinsics::_doubleIsFinite:
5399 result = new IsFiniteDNode(arg);
5400 break;
5401 default:
5402 fatal_unexpected_iid(id);
5403 break;
5404 }
5405 set_result(_gvn.transform(result));
5406 return true;
5407 }
5408
5409 //----------------------inline_unsafe_copyMemory-------------------------
5410 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5411
5412 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5413 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5414 const Type* base_t = gvn.type(base);
5415
5416 bool in_native = (base_t == TypePtr::NULL_PTR);
5417 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5418 bool is_mixed = !in_heap && !in_native;
5419
5420 if (is_mixed) {
5421 return true; // mixed accesses can touch both on-heap and off-heap memory
5422 }
5423 if (in_heap) {
5424 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5425 if (!is_prim_array) {
5426 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5427 // there's not enough type information available to determine proper memory slice for it.
5428 return true;
5429 }
5430 }
5431 return false;
5432 }
5433
5434 bool LibraryCallKit::inline_unsafe_copyMemory() {
5435 if (callee()->is_static()) return false; // caller must have the capability!
5436 null_check_receiver(); // null-check receiver
5437 if (stopped()) return true;
5438
5439 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5440
5441 Node* src_base = argument(1); // type: oop
5442 Node* src_off = ConvL2X(argument(2)); // type: long
5443 Node* dst_base = argument(4); // type: oop
5444 Node* dst_off = ConvL2X(argument(5)); // type: long
5445 Node* size = ConvL2X(argument(7)); // type: long
5446
5447 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5448 "fieldOffset must be byte-scaled");
5449
5450 Node* src_addr = make_unsafe_address(src_base, src_off);
5451 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5452
5453 Node* thread = _gvn.transform(new ThreadLocalNode());
5454 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5455 BasicType doing_unsafe_access_bt = T_BYTE;
5456 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5457
5458 // update volatile field
5459 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5460
5461 int flags = RC_LEAF | RC_NO_FP;
5462
5463 const TypePtr* dst_type = TypePtr::BOTTOM;
5464
5465 // Adjust memory effects of the runtime call based on input values.
5466 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5467 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5468 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5469
5470 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5471 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5472 flags |= RC_NARROW_MEM; // narrow in memory
5473 }
5474 }
5475
5476 // Call it. Note that the length argument is not scaled.
5477 make_runtime_call(flags,
5478 OptoRuntime::fast_arraycopy_Type(),
5479 StubRoutines::unsafe_arraycopy(),
5480 "unsafe_arraycopy",
5481 dst_type,
5482 src_addr, dst_addr, size XTOP);
5483
5484 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5485
5486 return true;
5487 }
5488
5489 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5490 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5491 bool LibraryCallKit::inline_unsafe_setMemory() {
5492 if (callee()->is_static()) return false; // caller must have the capability!
5493 null_check_receiver(); // null-check receiver
5494 if (stopped()) return true;
5495
5496 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5497
5498 Node* dst_base = argument(1); // type: oop
5499 Node* dst_off = ConvL2X(argument(2)); // type: long
5500 Node* size = ConvL2X(argument(4)); // type: long
5501 Node* byte = argument(6); // type: byte
5502
5503 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5504 "fieldOffset must be byte-scaled");
5505
5506 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5507
5508 Node* thread = _gvn.transform(new ThreadLocalNode());
5509 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5510 BasicType doing_unsafe_access_bt = T_BYTE;
5511 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5512
5513 // update volatile field
5514 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5515
5516 int flags = RC_LEAF | RC_NO_FP;
5517
5518 const TypePtr* dst_type = TypePtr::BOTTOM;
5519
5520 // Adjust memory effects of the runtime call based on input values.
5521 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5522 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5523
5524 flags |= RC_NARROW_MEM; // narrow in memory
5525 }
5526
5527 // Call it. Note that the length argument is not scaled.
5528 make_runtime_call(flags,
5529 OptoRuntime::unsafe_setmemory_Type(),
5530 StubRoutines::unsafe_setmemory(),
5531 "unsafe_setmemory",
5532 dst_type,
5533 dst_addr, size XTOP, byte);
5534
5535 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5536
5537 return true;
5538 }
5539
5540 #undef XTOP
5541
5542 //----------------------inline_unsafe_isFlatArray------------------------
5543 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5544 // This intrinsic exploits assumptions made by the native implementation
5545 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5546 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5547 Node* cls = argument(1);
5548 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5549 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p,
5550 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5551 Node* result = flat_array_test(kls);
5552 set_result(result);
5553 return true;
5554 }
5555
5556 //------------------------clone_coping-----------------------------------
5557 // Helper function for inline_native_clone.
5558 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5559 assert(obj_size != nullptr, "");
5560 Node* raw_obj = alloc_obj->in(1);
5561 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5562
5563 AllocateNode* alloc = nullptr;
5564 if (ReduceBulkZeroing &&
5565 // If we are implementing an array clone without knowing its source type
5566 // (can happen when compiling the array-guarded branch of a reflective
5567 // Object.clone() invocation), initialize the array within the allocation.
5568 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5569 // to a runtime clone call that assumes fully initialized source arrays.
5570 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5571 // We will be completely responsible for initializing this object -
5572 // mark Initialize node as complete.
5573 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5574 // The object was just allocated - there should be no any stores!
5575 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5576 // Mark as complete_with_arraycopy so that on AllocateNode
5577 // expansion, we know this AllocateNode is initialized by an array
5578 // copy and a StoreStore barrier exists after the array copy.
5579 alloc->initialization()->set_complete_with_arraycopy();
5580 }
5581
5582 Node* size = _gvn.transform(obj_size);
5583 access_clone(obj, alloc_obj, size, is_array);
5584
5585 // Do not let reads from the cloned object float above the arraycopy.
5586 if (alloc != nullptr) {
5587 // Do not let stores that initialize this object be reordered with
5588 // a subsequent store that would make this object accessible by
5589 // other threads.
5590 // Record what AllocateNode this StoreStore protects so that
5591 // escape analysis can go from the MemBarStoreStoreNode to the
5592 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5593 // based on the escape status of the AllocateNode.
5594 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5595 } else {
5596 insert_mem_bar(Op_MemBarCPUOrder);
5597 }
5598 }
5599
5600 //------------------------inline_native_clone----------------------------
5601 // protected native Object java.lang.Object.clone();
5602 //
5603 // Here are the simple edge cases:
5604 // null receiver => normal trap
5605 // virtual and clone was overridden => slow path to out-of-line clone
5606 // not cloneable or finalizer => slow path to out-of-line Object.clone
5607 //
5608 // The general case has two steps, allocation and copying.
5609 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5610 //
5611 // Copying also has two cases, oop arrays and everything else.
5612 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5613 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5614 //
5615 // These steps fold up nicely if and when the cloned object's klass
5616 // can be sharply typed as an object array, a type array, or an instance.
5617 //
5618 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5619 PhiNode* result_val;
5620
5621 // Set the reexecute bit for the interpreter to reexecute
5622 // the bytecode that invokes Object.clone if deoptimization happens.
5623 { PreserveReexecuteState preexecs(this);
5624 jvms()->set_should_reexecute(true);
5625
5626 Node* obj = argument(0);
5627 obj = null_check_receiver();
5628 if (stopped()) return true;
5629
5630 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5631 if (obj_type->is_inlinetypeptr()) {
5632 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5633 // no identity.
5634 set_result(obj);
5635 return true;
5636 }
5637
5638 // If we are going to clone an instance, we need its exact type to
5639 // know the number and types of fields to convert the clone to
5640 // loads/stores. Maybe a speculative type can help us.
5641 if (!obj_type->klass_is_exact() &&
5642 obj_type->speculative_type() != nullptr &&
5643 obj_type->speculative_type()->is_instance_klass() &&
5644 !obj_type->speculative_type()->is_inlinetype()) {
5645 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5646 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5647 !spec_ik->has_injected_fields()) {
5648 if (!obj_type->isa_instptr() ||
5649 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5650 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5651 }
5652 }
5653 }
5654
5655 // Conservatively insert a memory barrier on all memory slices.
5656 // Do not let writes into the original float below the clone.
5657 insert_mem_bar(Op_MemBarCPUOrder);
5658
5659 // paths into result_reg:
5660 enum {
5661 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5662 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5663 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5664 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5665 PATH_LIMIT
5666 };
5667 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5668 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5669 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5670 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5671 record_for_igvn(result_reg);
5672
5673 // TODO 8350865 For arrays, this might be folded and then not account for atomic arrays
5674 Node* obj_klass = load_object_klass(obj);
5675 // We only go to the fast case code if we pass a number of guards.
5676 // The paths which do not pass are accumulated in the slow_region.
5677 RegionNode* slow_region = new RegionNode(1);
5678 record_for_igvn(slow_region);
5679
5680 Node* array_obj = obj;
5681 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5682 if (array_ctl != nullptr) {
5683 // It's an array.
5684 PreserveJVMState pjvms(this);
5685 set_control(array_ctl);
5686
5687 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5688 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5689 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5690 obj_type->can_be_inline_array() &&
5691 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5692 // Flat inline type array may have object field that would require a
5693 // write barrier. Conservatively, go to slow path.
5694 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5695 }
5696
5697 if (!stopped()) {
5698 Node* obj_length = load_array_length(array_obj);
5699 Node* array_size = nullptr; // Size of the array without object alignment padding.
5700 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5701
5702 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5703 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5704 // If it is an oop array, it requires very special treatment,
5705 // because gc barriers are required when accessing the array.
5706 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5707 if (is_obja != nullptr) {
5708 PreserveJVMState pjvms2(this);
5709 set_control(is_obja);
5710 // Generate a direct call to the right arraycopy function(s).
5711 // Clones are always tightly coupled.
5712 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5713 ac->set_clone_oop_array();
5714 Node* n = _gvn.transform(ac);
5715 assert(n == ac, "cannot disappear");
5716 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5717
5718 result_reg->init_req(_objArray_path, control());
5719 result_val->init_req(_objArray_path, alloc_obj);
5720 result_i_o ->set_req(_objArray_path, i_o());
5721 result_mem ->set_req(_objArray_path, reset_memory());
5722 }
5723 }
5724 // Otherwise, there are no barriers to worry about.
5725 // (We can dispense with card marks if we know the allocation
5726 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5727 // causes the non-eden paths to take compensating steps to
5728 // simulate a fresh allocation, so that no further
5729 // card marks are required in compiled code to initialize
5730 // the object.)
5731
5732 if (!stopped()) {
5733 copy_to_clone(obj, alloc_obj, array_size, true);
5734
5735 // Present the results of the copy.
5736 result_reg->init_req(_array_path, control());
5737 result_val->init_req(_array_path, alloc_obj);
5738 result_i_o ->set_req(_array_path, i_o());
5739 result_mem ->set_req(_array_path, reset_memory());
5740 }
5741 }
5742 }
5743
5744 if (!stopped()) {
5745 // It's an instance (we did array above). Make the slow-path tests.
5746 // If this is a virtual call, we generate a funny guard. We grab
5747 // the vtable entry corresponding to clone() from the target object.
5748 // If the target method which we are calling happens to be the
5749 // Object clone() method, we pass the guard. We do not need this
5750 // guard for non-virtual calls; the caller is known to be the native
5751 // Object clone().
5752 if (is_virtual) {
5753 generate_virtual_guard(obj_klass, slow_region);
5754 }
5755
5756 // The object must be easily cloneable and must not have a finalizer.
5757 // Both of these conditions may be checked in a single test.
5758 // We could optimize the test further, but we don't care.
5759 generate_misc_flags_guard(obj_klass,
5760 // Test both conditions:
5761 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5762 // Must be cloneable but not finalizer:
5763 KlassFlags::_misc_is_cloneable_fast,
5764 slow_region);
5765 }
5766
5767 if (!stopped()) {
5768 // It's an instance, and it passed the slow-path tests.
5769 PreserveJVMState pjvms(this);
5770 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5771 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5772 // is reexecuted if deoptimization occurs and there could be problems when merging
5773 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5774 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5775
5776 copy_to_clone(obj, alloc_obj, obj_size, false);
5777
5778 // Present the results of the slow call.
5779 result_reg->init_req(_instance_path, control());
5780 result_val->init_req(_instance_path, alloc_obj);
5781 result_i_o ->set_req(_instance_path, i_o());
5782 result_mem ->set_req(_instance_path, reset_memory());
5783 }
5784
5785 // Generate code for the slow case. We make a call to clone().
5786 set_control(_gvn.transform(slow_region));
5787 if (!stopped()) {
5788 PreserveJVMState pjvms(this);
5789 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5790 // We need to deoptimize on exception (see comment above)
5791 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5792 // this->control() comes from set_results_for_java_call
5793 result_reg->init_req(_slow_path, control());
5794 result_val->init_req(_slow_path, slow_result);
5795 result_i_o ->set_req(_slow_path, i_o());
5796 result_mem ->set_req(_slow_path, reset_memory());
5797 }
5798
5799 // Return the combined state.
5800 set_control( _gvn.transform(result_reg));
5801 set_i_o( _gvn.transform(result_i_o));
5802 set_all_memory( _gvn.transform(result_mem));
5803 } // original reexecute is set back here
5804
5805 set_result(_gvn.transform(result_val));
5806 return true;
5807 }
5808
5809 // If we have a tightly coupled allocation, the arraycopy may take care
5810 // of the array initialization. If one of the guards we insert between
5811 // the allocation and the arraycopy causes a deoptimization, an
5812 // uninitialized array will escape the compiled method. To prevent that
5813 // we set the JVM state for uncommon traps between the allocation and
5814 // the arraycopy to the state before the allocation so, in case of
5815 // deoptimization, we'll reexecute the allocation and the
5816 // initialization.
5817 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5818 if (alloc != nullptr) {
5819 ciMethod* trap_method = alloc->jvms()->method();
5820 int trap_bci = alloc->jvms()->bci();
5821
5822 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5823 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5824 // Make sure there's no store between the allocation and the
5825 // arraycopy otherwise visible side effects could be rexecuted
5826 // in case of deoptimization and cause incorrect execution.
5827 bool no_interfering_store = true;
5828 Node* mem = alloc->in(TypeFunc::Memory);
5829 if (mem->is_MergeMem()) {
5830 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5831 Node* n = mms.memory();
5832 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5833 assert(n->is_Store(), "what else?");
5834 no_interfering_store = false;
5835 break;
5836 }
5837 }
5838 } else {
5839 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5840 Node* n = mms.memory();
5841 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5842 assert(n->is_Store(), "what else?");
5843 no_interfering_store = false;
5844 break;
5845 }
5846 }
5847 }
5848
5849 if (no_interfering_store) {
5850 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5851
5852 JVMState* saved_jvms = jvms();
5853 saved_reexecute_sp = _reexecute_sp;
5854
5855 set_jvms(sfpt->jvms());
5856 _reexecute_sp = jvms()->sp();
5857
5858 return saved_jvms;
5859 }
5860 }
5861 }
5862 return nullptr;
5863 }
5864
5865 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5866 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5867 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5868 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5869 uint size = alloc->req();
5870 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5871 old_jvms->set_map(sfpt);
5872 for (uint i = 0; i < size; i++) {
5873 sfpt->init_req(i, alloc->in(i));
5874 }
5875 int adjustment = 1;
5876 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5877 if (ary_klass_ptr->is_null_free()) {
5878 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
5879 // also requires the componentType and initVal on stack for re-execution.
5880 // Re-create and push the componentType.
5881 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5882 ciInstance* instance = klass->component_mirror_instance();
5883 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5884 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5885 adjustment++;
5886 }
5887 // re-push array length for deoptimization
5888 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5889 if (ary_klass_ptr->is_null_free()) {
5890 // Re-create and push the initVal.
5891 Node* init_val = alloc->in(AllocateNode::InitValue);
5892 if (init_val == nullptr) {
5893 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
5894 } else if (UseCompressedOops) {
5895 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
5896 }
5897 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
5898 adjustment++;
5899 }
5900 old_jvms->set_sp(old_jvms->sp() + adjustment);
5901 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5902 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5903 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5904 old_jvms->set_should_reexecute(true);
5905
5906 sfpt->set_i_o(map()->i_o());
5907 sfpt->set_memory(map()->memory());
5908 sfpt->set_control(map()->control());
5909 return sfpt;
5910 }
5911
5912 // In case of a deoptimization, we restart execution at the
5913 // allocation, allocating a new array. We would leave an uninitialized
5914 // array in the heap that GCs wouldn't expect. Move the allocation
5915 // after the traps so we don't allocate the array if we
5916 // deoptimize. This is possible because tightly_coupled_allocation()
5917 // guarantees there's no observer of the allocated array at this point
5918 // and the control flow is simple enough.
5919 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5920 int saved_reexecute_sp, uint new_idx) {
5921 if (saved_jvms_before_guards != nullptr && !stopped()) {
5922 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5923
5924 assert(alloc != nullptr, "only with a tightly coupled allocation");
5925 // restore JVM state to the state at the arraycopy
5926 saved_jvms_before_guards->map()->set_control(map()->control());
5927 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5928 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5929 // If we've improved the types of some nodes (null check) while
5930 // emitting the guards, propagate them to the current state
5931 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5932 set_jvms(saved_jvms_before_guards);
5933 _reexecute_sp = saved_reexecute_sp;
5934
5935 // Remove the allocation from above the guards
5936 CallProjections* callprojs = alloc->extract_projections(true);
5937 InitializeNode* init = alloc->initialization();
5938 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5939 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5940 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5941
5942 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5943 // the allocation (i.e. is only valid if the allocation succeeds):
5944 // 1) replace CastIINode with AllocateArrayNode's length here
5945 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5946 //
5947 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5948 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5949 Node* init_control = init->proj_out(TypeFunc::Control);
5950 Node* alloc_length = alloc->Ideal_length();
5951 #ifdef ASSERT
5952 Node* prev_cast = nullptr;
5953 #endif
5954 for (uint i = 0; i < init_control->outcnt(); i++) {
5955 Node* init_out = init_control->raw_out(i);
5956 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5957 #ifdef ASSERT
5958 if (prev_cast == nullptr) {
5959 prev_cast = init_out;
5960 } else {
5961 if (prev_cast->cmp(*init_out) == false) {
5962 prev_cast->dump();
5963 init_out->dump();
5964 assert(false, "not equal CastIINode");
5965 }
5966 }
5967 #endif
5968 C->gvn_replace_by(init_out, alloc_length);
5969 }
5970 }
5971 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5972
5973 // move the allocation here (after the guards)
5974 _gvn.hash_delete(alloc);
5975 alloc->set_req(TypeFunc::Control, control());
5976 alloc->set_req(TypeFunc::I_O, i_o());
5977 Node *mem = reset_memory();
5978 set_all_memory(mem);
5979 alloc->set_req(TypeFunc::Memory, mem);
5980 set_control(init->proj_out_or_null(TypeFunc::Control));
5981 set_i_o(callprojs->fallthrough_ioproj);
5982
5983 // Update memory as done in GraphKit::set_output_for_allocation()
5984 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5985 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5986 if (ary_type->isa_aryptr() && length_type != nullptr) {
5987 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5988 }
5989 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5990 int elemidx = C->get_alias_index(telemref);
5991 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5992 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5993
5994 Node* allocx = _gvn.transform(alloc);
5995 assert(allocx == alloc, "where has the allocation gone?");
5996 assert(dest->is_CheckCastPP(), "not an allocation result?");
5997
5998 _gvn.hash_delete(dest);
5999 dest->set_req(0, control());
6000 Node* destx = _gvn.transform(dest);
6001 assert(destx == dest, "where has the allocation result gone?");
6002
6003 array_ideal_length(alloc, ary_type, true);
6004 }
6005 }
6006
6007 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6008 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6009 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6010 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6011 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6012 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6013 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6014 JVMState* saved_jvms_before_guards) {
6015 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6016 // There is at least one unrelated uncommon trap which needs to be replaced.
6017 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6018
6019 JVMState* saved_jvms = jvms();
6020 const int saved_reexecute_sp = _reexecute_sp;
6021 set_jvms(sfpt->jvms());
6022 _reexecute_sp = jvms()->sp();
6023
6024 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6025
6026 // Restore state
6027 set_jvms(saved_jvms);
6028 _reexecute_sp = saved_reexecute_sp;
6029 }
6030 }
6031
6032 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6033 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6034 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6035 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6036 while (if_proj->is_IfProj()) {
6037 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6038 if (uncommon_trap != nullptr) {
6039 create_new_uncommon_trap(uncommon_trap);
6040 }
6041 assert(if_proj->in(0)->is_If(), "must be If");
6042 if_proj = if_proj->in(0)->in(0);
6043 }
6044 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6045 "must have reached control projection of init node");
6046 }
6047
6048 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6049 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6050 assert(trap_request != 0, "no valid UCT trap request");
6051 PreserveJVMState pjvms(this);
6052 set_control(uncommon_trap_call->in(0));
6053 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6054 Deoptimization::trap_request_action(trap_request));
6055 assert(stopped(), "Should be stopped");
6056 _gvn.hash_delete(uncommon_trap_call);
6057 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6058 }
6059
6060 // Common checks for array sorting intrinsics arguments.
6061 // Returns `true` if checks passed.
6062 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6063 // check address of the class
6064 if (elementType == nullptr || elementType->is_top()) {
6065 return false; // dead path
6066 }
6067 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6068 if (elem_klass == nullptr) {
6069 return false; // dead path
6070 }
6071 // java_mirror_type() returns non-null for compile-time Class constants only
6072 ciType* elem_type = elem_klass->java_mirror_type();
6073 if (elem_type == nullptr) {
6074 return false;
6075 }
6076 bt = elem_type->basic_type();
6077 // Disable the intrinsic if the CPU does not support SIMD sort
6078 if (!Matcher::supports_simd_sort(bt)) {
6079 return false;
6080 }
6081 // check address of the array
6082 if (obj == nullptr || obj->is_top()) {
6083 return false; // dead path
6084 }
6085 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6086 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6087 return false; // failed input validation
6088 }
6089 return true;
6090 }
6091
6092 //------------------------------inline_array_partition-----------------------
6093 bool LibraryCallKit::inline_array_partition() {
6094 address stubAddr = StubRoutines::select_array_partition_function();
6095 if (stubAddr == nullptr) {
6096 return false; // Intrinsic's stub is not implemented on this platform
6097 }
6098 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6099
6100 // no receiver because it is a static method
6101 Node* elementType = argument(0);
6102 Node* obj = argument(1);
6103 Node* offset = argument(2); // long
6104 Node* fromIndex = argument(4);
6105 Node* toIndex = argument(5);
6106 Node* indexPivot1 = argument(6);
6107 Node* indexPivot2 = argument(7);
6108 // PartitionOperation: argument(8) is ignored
6109
6110 Node* pivotIndices = nullptr;
6111 BasicType bt = T_ILLEGAL;
6112
6113 if (!check_array_sort_arguments(elementType, obj, bt)) {
6114 return false;
6115 }
6116 null_check(obj);
6117 // If obj is dead, only null-path is taken.
6118 if (stopped()) {
6119 return true;
6120 }
6121 // Set the original stack and the reexecute bit for the interpreter to reexecute
6122 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6123 { PreserveReexecuteState preexecs(this);
6124 jvms()->set_should_reexecute(true);
6125
6126 Node* obj_adr = make_unsafe_address(obj, offset);
6127
6128 // create the pivotIndices array of type int and size = 2
6129 Node* size = intcon(2);
6130 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6131 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6132 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6133 guarantee(alloc != nullptr, "created above");
6134 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6135
6136 // pass the basic type enum to the stub
6137 Node* elemType = intcon(bt);
6138
6139 // Call the stub
6140 const char *stubName = "array_partition_stub";
6141 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6142 stubAddr, stubName, TypePtr::BOTTOM,
6143 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6144 indexPivot1, indexPivot2);
6145
6146 } // original reexecute is set back here
6147
6148 if (!stopped()) {
6149 set_result(pivotIndices);
6150 }
6151
6152 return true;
6153 }
6154
6155
6156 //------------------------------inline_array_sort-----------------------
6157 bool LibraryCallKit::inline_array_sort() {
6158 address stubAddr = StubRoutines::select_arraysort_function();
6159 if (stubAddr == nullptr) {
6160 return false; // Intrinsic's stub is not implemented on this platform
6161 }
6162 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6163
6164 // no receiver because it is a static method
6165 Node* elementType = argument(0);
6166 Node* obj = argument(1);
6167 Node* offset = argument(2); // long
6168 Node* fromIndex = argument(4);
6169 Node* toIndex = argument(5);
6170 // SortOperation: argument(6) is ignored
6171
6172 BasicType bt = T_ILLEGAL;
6173
6174 if (!check_array_sort_arguments(elementType, obj, bt)) {
6175 return false;
6176 }
6177 null_check(obj);
6178 // If obj is dead, only null-path is taken.
6179 if (stopped()) {
6180 return true;
6181 }
6182 Node* obj_adr = make_unsafe_address(obj, offset);
6183
6184 // pass the basic type enum to the stub
6185 Node* elemType = intcon(bt);
6186
6187 // Call the stub.
6188 const char *stubName = "arraysort_stub";
6189 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6190 stubAddr, stubName, TypePtr::BOTTOM,
6191 obj_adr, elemType, fromIndex, toIndex);
6192
6193 return true;
6194 }
6195
6196
6197 //------------------------------inline_arraycopy-----------------------
6198 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6199 // Object dest, int destPos,
6200 // int length);
6201 bool LibraryCallKit::inline_arraycopy() {
6202 // Get the arguments.
6203 Node* src = argument(0); // type: oop
6204 Node* src_offset = argument(1); // type: int
6205 Node* dest = argument(2); // type: oop
6206 Node* dest_offset = argument(3); // type: int
6207 Node* length = argument(4); // type: int
6208
6209 uint new_idx = C->unique();
6210
6211 // Check for allocation before we add nodes that would confuse
6212 // tightly_coupled_allocation()
6213 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6214
6215 int saved_reexecute_sp = -1;
6216 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6217 // See arraycopy_restore_alloc_state() comment
6218 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6219 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6220 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6221 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6222
6223 // The following tests must be performed
6224 // (1) src and dest are arrays.
6225 // (2) src and dest arrays must have elements of the same BasicType
6226 // (3) src and dest must not be null.
6227 // (4) src_offset must not be negative.
6228 // (5) dest_offset must not be negative.
6229 // (6) length must not be negative.
6230 // (7) src_offset + length must not exceed length of src.
6231 // (8) dest_offset + length must not exceed length of dest.
6232 // (9) each element of an oop array must be assignable
6233
6234 // (3) src and dest must not be null.
6235 // always do this here because we need the JVM state for uncommon traps
6236 Node* null_ctl = top();
6237 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6238 assert(null_ctl->is_top(), "no null control here");
6239 dest = null_check(dest, T_ARRAY);
6240
6241 if (!can_emit_guards) {
6242 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6243 // guards but the arraycopy node could still take advantage of a
6244 // tightly allocated allocation. tightly_coupled_allocation() is
6245 // called again to make sure it takes the null check above into
6246 // account: the null check is mandatory and if it caused an
6247 // uncommon trap to be emitted then the allocation can't be
6248 // considered tightly coupled in this context.
6249 alloc = tightly_coupled_allocation(dest);
6250 }
6251
6252 bool validated = false;
6253
6254 const Type* src_type = _gvn.type(src);
6255 const Type* dest_type = _gvn.type(dest);
6256 const TypeAryPtr* top_src = src_type->isa_aryptr();
6257 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6258
6259 // Do we have the type of src?
6260 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6261 // Do we have the type of dest?
6262 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6263 // Is the type for src from speculation?
6264 bool src_spec = false;
6265 // Is the type for dest from speculation?
6266 bool dest_spec = false;
6267
6268 if ((!has_src || !has_dest) && can_emit_guards) {
6269 // We don't have sufficient type information, let's see if
6270 // speculative types can help. We need to have types for both src
6271 // and dest so that it pays off.
6272
6273 // Do we already have or could we have type information for src
6274 bool could_have_src = has_src;
6275 // Do we already have or could we have type information for dest
6276 bool could_have_dest = has_dest;
6277
6278 ciKlass* src_k = nullptr;
6279 if (!has_src) {
6280 src_k = src_type->speculative_type_not_null();
6281 if (src_k != nullptr && src_k->is_array_klass()) {
6282 could_have_src = true;
6283 }
6284 }
6285
6286 ciKlass* dest_k = nullptr;
6287 if (!has_dest) {
6288 dest_k = dest_type->speculative_type_not_null();
6289 if (dest_k != nullptr && dest_k->is_array_klass()) {
6290 could_have_dest = true;
6291 }
6292 }
6293
6294 if (could_have_src && could_have_dest) {
6295 // This is going to pay off so emit the required guards
6296 if (!has_src) {
6297 src = maybe_cast_profiled_obj(src, src_k, true);
6298 src_type = _gvn.type(src);
6299 top_src = src_type->isa_aryptr();
6300 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6301 src_spec = true;
6302 }
6303 if (!has_dest) {
6304 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6305 dest_type = _gvn.type(dest);
6306 top_dest = dest_type->isa_aryptr();
6307 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6308 dest_spec = true;
6309 }
6310 }
6311 }
6312
6313 if (has_src && has_dest && can_emit_guards) {
6314 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6315 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6316 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6317 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6318
6319 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6320 // If both arrays are object arrays then having the exact types
6321 // for both will remove the need for a subtype check at runtime
6322 // before the call and may make it possible to pick a faster copy
6323 // routine (without a subtype check on every element)
6324 // Do we have the exact type of src?
6325 bool could_have_src = src_spec;
6326 // Do we have the exact type of dest?
6327 bool could_have_dest = dest_spec;
6328 ciKlass* src_k = nullptr;
6329 ciKlass* dest_k = nullptr;
6330 if (!src_spec) {
6331 src_k = src_type->speculative_type_not_null();
6332 if (src_k != nullptr && src_k->is_array_klass()) {
6333 could_have_src = true;
6334 }
6335 }
6336 if (!dest_spec) {
6337 dest_k = dest_type->speculative_type_not_null();
6338 if (dest_k != nullptr && dest_k->is_array_klass()) {
6339 could_have_dest = true;
6340 }
6341 }
6342 if (could_have_src && could_have_dest) {
6343 // If we can have both exact types, emit the missing guards
6344 if (could_have_src && !src_spec) {
6345 src = maybe_cast_profiled_obj(src, src_k, true);
6346 src_type = _gvn.type(src);
6347 top_src = src_type->isa_aryptr();
6348 }
6349 if (could_have_dest && !dest_spec) {
6350 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6351 dest_type = _gvn.type(dest);
6352 top_dest = dest_type->isa_aryptr();
6353 }
6354 }
6355 }
6356 }
6357
6358 ciMethod* trap_method = method();
6359 int trap_bci = bci();
6360 if (saved_jvms_before_guards != nullptr) {
6361 trap_method = alloc->jvms()->method();
6362 trap_bci = alloc->jvms()->bci();
6363 }
6364
6365 bool negative_length_guard_generated = false;
6366
6367 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6368 can_emit_guards && !src->is_top() && !dest->is_top()) {
6369 // validate arguments: enables transformation the ArrayCopyNode
6370 validated = true;
6371
6372 RegionNode* slow_region = new RegionNode(1);
6373 record_for_igvn(slow_region);
6374
6375 // (1) src and dest are arrays.
6376 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6377 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6378
6379 // (2) src and dest arrays must have elements of the same BasicType
6380 // done at macro expansion or at Ideal transformation time
6381
6382 // (4) src_offset must not be negative.
6383 generate_negative_guard(src_offset, slow_region);
6384
6385 // (5) dest_offset must not be negative.
6386 generate_negative_guard(dest_offset, slow_region);
6387
6388 // (7) src_offset + length must not exceed length of src.
6389 generate_limit_guard(src_offset, length,
6390 load_array_length(src),
6391 slow_region);
6392
6393 // (8) dest_offset + length must not exceed length of dest.
6394 generate_limit_guard(dest_offset, length,
6395 load_array_length(dest),
6396 slow_region);
6397
6398 // (6) length must not be negative.
6399 // This is also checked in generate_arraycopy() during macro expansion, but
6400 // we also have to check it here for the case where the ArrayCopyNode will
6401 // be eliminated by Escape Analysis.
6402 if (EliminateAllocations) {
6403 generate_negative_guard(length, slow_region);
6404 negative_length_guard_generated = true;
6405 }
6406
6407 // (9) each element of an oop array must be assignable
6408 Node* dest_klass = load_object_klass(dest);
6409 if (src != dest) {
6410 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6411 slow_region->add_req(not_subtype_ctrl);
6412 }
6413
6414 // TODO 8350865 Fix below logic. Also handle atomicity.
6415 generate_fair_guard(flat_array_test(src), slow_region);
6416 generate_fair_guard(flat_array_test(dest), slow_region);
6417
6418 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6419 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6420 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6421 src_type = _gvn.type(src);
6422 top_src = src_type->isa_aryptr();
6423
6424 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6425 if (!stopped() && UseArrayFlattening) {
6426 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6427 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6428 if (top_src != nullptr && top_src->is_flat()) {
6429 // Src is flat, check that dest is flat as well
6430 if (top_dest != nullptr && !top_dest->is_flat()) {
6431 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6432 // Since dest is flat and src <: dest, dest must have the same type as src.
6433 top_dest = top_src->cast_to_exactness(false);
6434 assert(top_dest->is_flat(), "dest must be flat");
6435 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6436 }
6437 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6438 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6439 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6440 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6441 generate_fair_guard(flat_array_test(src), slow_region);
6442 if (top_src != nullptr) {
6443 top_src = top_src->cast_to_not_flat();
6444 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6445 }
6446 }
6447 }
6448
6449 {
6450 PreserveJVMState pjvms(this);
6451 set_control(_gvn.transform(slow_region));
6452 uncommon_trap(Deoptimization::Reason_intrinsic,
6453 Deoptimization::Action_make_not_entrant);
6454 assert(stopped(), "Should be stopped");
6455 }
6456 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6457 }
6458
6459 if (stopped()) {
6460 return true;
6461 }
6462
6463 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6464 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6465 // so the compiler has a chance to eliminate them: during macro expansion,
6466 // we have to set their control (CastPP nodes are eliminated).
6467 load_object_klass(src), load_object_klass(dest),
6468 load_array_length(src), load_array_length(dest));
6469
6470 ac->set_arraycopy(validated);
6471
6472 Node* n = _gvn.transform(ac);
6473 if (n == ac) {
6474 ac->connect_outputs(this);
6475 } else {
6476 assert(validated, "shouldn't transform if all arguments not validated");
6477 set_all_memory(n);
6478 }
6479 clear_upper_avx();
6480
6481
6482 return true;
6483 }
6484
6485
6486 // Helper function which determines if an arraycopy immediately follows
6487 // an allocation, with no intervening tests or other escapes for the object.
6488 AllocateArrayNode*
6489 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6490 if (stopped()) return nullptr; // no fast path
6491 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6492
6493 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6494 if (alloc == nullptr) return nullptr;
6495
6496 Node* rawmem = memory(Compile::AliasIdxRaw);
6497 // Is the allocation's memory state untouched?
6498 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6499 // Bail out if there have been raw-memory effects since the allocation.
6500 // (Example: There might have been a call or safepoint.)
6501 return nullptr;
6502 }
6503 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6504 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6505 return nullptr;
6506 }
6507
6508 // There must be no unexpected observers of this allocation.
6509 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6510 Node* obs = ptr->fast_out(i);
6511 if (obs != this->map()) {
6512 return nullptr;
6513 }
6514 }
6515
6516 // This arraycopy must unconditionally follow the allocation of the ptr.
6517 Node* alloc_ctl = ptr->in(0);
6518 Node* ctl = control();
6519 while (ctl != alloc_ctl) {
6520 // There may be guards which feed into the slow_region.
6521 // Any other control flow means that we might not get a chance
6522 // to finish initializing the allocated object.
6523 // Various low-level checks bottom out in uncommon traps. These
6524 // are considered safe since we've already checked above that
6525 // there is no unexpected observer of this allocation.
6526 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6527 assert(ctl->in(0)->is_If(), "must be If");
6528 ctl = ctl->in(0)->in(0);
6529 } else {
6530 return nullptr;
6531 }
6532 }
6533
6534 // If we get this far, we have an allocation which immediately
6535 // precedes the arraycopy, and we can take over zeroing the new object.
6536 // The arraycopy will finish the initialization, and provide
6537 // a new control state to which we will anchor the destination pointer.
6538
6539 return alloc;
6540 }
6541
6542 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6543 if (node->is_IfProj()) {
6544 Node* other_proj = node->as_IfProj()->other_if_proj();
6545 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6546 Node* obs = other_proj->fast_out(j);
6547 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6548 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6549 return obs->as_CallStaticJava();
6550 }
6551 }
6552 }
6553 return nullptr;
6554 }
6555
6556 //-------------inline_encodeISOArray-----------------------------------
6557 // encode char[] to byte[] in ISO_8859_1 or ASCII
6558 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6559 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6560 // no receiver since it is static method
6561 Node *src = argument(0);
6562 Node *src_offset = argument(1);
6563 Node *dst = argument(2);
6564 Node *dst_offset = argument(3);
6565 Node *length = argument(4);
6566
6567 src = must_be_not_null(src, true);
6568 dst = must_be_not_null(dst, true);
6569
6570 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6571 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6572 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6573 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6574 // failed array check
6575 return false;
6576 }
6577
6578 // Figure out the size and type of the elements we will be copying.
6579 BasicType src_elem = src_type->elem()->array_element_basic_type();
6580 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6581 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6582 return false;
6583 }
6584
6585 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6586 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6587 // 'src_start' points to src array + scaled offset
6588 // 'dst_start' points to dst array + scaled offset
6589
6590 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6591 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6592 enc = _gvn.transform(enc);
6593 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6594 set_memory(res_mem, mtype);
6595 set_result(enc);
6596 clear_upper_avx();
6597
6598 return true;
6599 }
6600
6601 //-------------inline_multiplyToLen-----------------------------------
6602 bool LibraryCallKit::inline_multiplyToLen() {
6603 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6604
6605 address stubAddr = StubRoutines::multiplyToLen();
6606 if (stubAddr == nullptr) {
6607 return false; // Intrinsic's stub is not implemented on this platform
6608 }
6609 const char* stubName = "multiplyToLen";
6610
6611 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6612
6613 // no receiver because it is a static method
6614 Node* x = argument(0);
6615 Node* xlen = argument(1);
6616 Node* y = argument(2);
6617 Node* ylen = argument(3);
6618 Node* z = argument(4);
6619
6620 x = must_be_not_null(x, true);
6621 y = must_be_not_null(y, true);
6622
6623 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6624 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6625 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6626 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6627 // failed array check
6628 return false;
6629 }
6630
6631 BasicType x_elem = x_type->elem()->array_element_basic_type();
6632 BasicType y_elem = y_type->elem()->array_element_basic_type();
6633 if (x_elem != T_INT || y_elem != T_INT) {
6634 return false;
6635 }
6636
6637 Node* x_start = array_element_address(x, intcon(0), x_elem);
6638 Node* y_start = array_element_address(y, intcon(0), y_elem);
6639 // 'x_start' points to x array + scaled xlen
6640 // 'y_start' points to y array + scaled ylen
6641
6642 Node* z_start = array_element_address(z, intcon(0), T_INT);
6643
6644 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6645 OptoRuntime::multiplyToLen_Type(),
6646 stubAddr, stubName, TypePtr::BOTTOM,
6647 x_start, xlen, y_start, ylen, z_start);
6648
6649 C->set_has_split_ifs(true); // Has chance for split-if optimization
6650 set_result(z);
6651 return true;
6652 }
6653
6654 //-------------inline_squareToLen------------------------------------
6655 bool LibraryCallKit::inline_squareToLen() {
6656 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6657
6658 address stubAddr = StubRoutines::squareToLen();
6659 if (stubAddr == nullptr) {
6660 return false; // Intrinsic's stub is not implemented on this platform
6661 }
6662 const char* stubName = "squareToLen";
6663
6664 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6665
6666 Node* x = argument(0);
6667 Node* len = argument(1);
6668 Node* z = argument(2);
6669 Node* zlen = argument(3);
6670
6671 x = must_be_not_null(x, true);
6672 z = must_be_not_null(z, true);
6673
6674 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6675 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6676 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6677 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6678 // failed array check
6679 return false;
6680 }
6681
6682 BasicType x_elem = x_type->elem()->array_element_basic_type();
6683 BasicType z_elem = z_type->elem()->array_element_basic_type();
6684 if (x_elem != T_INT || z_elem != T_INT) {
6685 return false;
6686 }
6687
6688
6689 Node* x_start = array_element_address(x, intcon(0), x_elem);
6690 Node* z_start = array_element_address(z, intcon(0), z_elem);
6691
6692 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6693 OptoRuntime::squareToLen_Type(),
6694 stubAddr, stubName, TypePtr::BOTTOM,
6695 x_start, len, z_start, zlen);
6696
6697 set_result(z);
6698 return true;
6699 }
6700
6701 //-------------inline_mulAdd------------------------------------------
6702 bool LibraryCallKit::inline_mulAdd() {
6703 assert(UseMulAddIntrinsic, "not implemented on this platform");
6704
6705 address stubAddr = StubRoutines::mulAdd();
6706 if (stubAddr == nullptr) {
6707 return false; // Intrinsic's stub is not implemented on this platform
6708 }
6709 const char* stubName = "mulAdd";
6710
6711 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6712
6713 Node* out = argument(0);
6714 Node* in = argument(1);
6715 Node* offset = argument(2);
6716 Node* len = argument(3);
6717 Node* k = argument(4);
6718
6719 in = must_be_not_null(in, true);
6720 out = must_be_not_null(out, true);
6721
6722 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6723 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6724 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6725 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6726 // failed array check
6727 return false;
6728 }
6729
6730 BasicType out_elem = out_type->elem()->array_element_basic_type();
6731 BasicType in_elem = in_type->elem()->array_element_basic_type();
6732 if (out_elem != T_INT || in_elem != T_INT) {
6733 return false;
6734 }
6735
6736 Node* outlen = load_array_length(out);
6737 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6738 Node* out_start = array_element_address(out, intcon(0), out_elem);
6739 Node* in_start = array_element_address(in, intcon(0), in_elem);
6740
6741 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6742 OptoRuntime::mulAdd_Type(),
6743 stubAddr, stubName, TypePtr::BOTTOM,
6744 out_start,in_start, new_offset, len, k);
6745 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6746 set_result(result);
6747 return true;
6748 }
6749
6750 //-------------inline_montgomeryMultiply-----------------------------------
6751 bool LibraryCallKit::inline_montgomeryMultiply() {
6752 address stubAddr = StubRoutines::montgomeryMultiply();
6753 if (stubAddr == nullptr) {
6754 return false; // Intrinsic's stub is not implemented on this platform
6755 }
6756
6757 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6758 const char* stubName = "montgomery_multiply";
6759
6760 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6761
6762 Node* a = argument(0);
6763 Node* b = argument(1);
6764 Node* n = argument(2);
6765 Node* len = argument(3);
6766 Node* inv = argument(4);
6767 Node* m = argument(6);
6768
6769 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6770 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6771 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6772 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6773 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6774 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6775 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6776 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6777 // failed array check
6778 return false;
6779 }
6780
6781 BasicType a_elem = a_type->elem()->array_element_basic_type();
6782 BasicType b_elem = b_type->elem()->array_element_basic_type();
6783 BasicType n_elem = n_type->elem()->array_element_basic_type();
6784 BasicType m_elem = m_type->elem()->array_element_basic_type();
6785 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6786 return false;
6787 }
6788
6789 // Make the call
6790 {
6791 Node* a_start = array_element_address(a, intcon(0), a_elem);
6792 Node* b_start = array_element_address(b, intcon(0), b_elem);
6793 Node* n_start = array_element_address(n, intcon(0), n_elem);
6794 Node* m_start = array_element_address(m, intcon(0), m_elem);
6795
6796 Node* call = make_runtime_call(RC_LEAF,
6797 OptoRuntime::montgomeryMultiply_Type(),
6798 stubAddr, stubName, TypePtr::BOTTOM,
6799 a_start, b_start, n_start, len, inv, top(),
6800 m_start);
6801 set_result(m);
6802 }
6803
6804 return true;
6805 }
6806
6807 bool LibraryCallKit::inline_montgomerySquare() {
6808 address stubAddr = StubRoutines::montgomerySquare();
6809 if (stubAddr == nullptr) {
6810 return false; // Intrinsic's stub is not implemented on this platform
6811 }
6812
6813 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6814 const char* stubName = "montgomery_square";
6815
6816 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6817
6818 Node* a = argument(0);
6819 Node* n = argument(1);
6820 Node* len = argument(2);
6821 Node* inv = argument(3);
6822 Node* m = argument(5);
6823
6824 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6825 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6826 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6827 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6828 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6829 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6830 // failed array check
6831 return false;
6832 }
6833
6834 BasicType a_elem = a_type->elem()->array_element_basic_type();
6835 BasicType n_elem = n_type->elem()->array_element_basic_type();
6836 BasicType m_elem = m_type->elem()->array_element_basic_type();
6837 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6838 return false;
6839 }
6840
6841 // Make the call
6842 {
6843 Node* a_start = array_element_address(a, intcon(0), a_elem);
6844 Node* n_start = array_element_address(n, intcon(0), n_elem);
6845 Node* m_start = array_element_address(m, intcon(0), m_elem);
6846
6847 Node* call = make_runtime_call(RC_LEAF,
6848 OptoRuntime::montgomerySquare_Type(),
6849 stubAddr, stubName, TypePtr::BOTTOM,
6850 a_start, n_start, len, inv, top(),
6851 m_start);
6852 set_result(m);
6853 }
6854
6855 return true;
6856 }
6857
6858 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6859 address stubAddr = nullptr;
6860 const char* stubName = nullptr;
6861
6862 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6863 if (stubAddr == nullptr) {
6864 return false; // Intrinsic's stub is not implemented on this platform
6865 }
6866
6867 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6868
6869 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6870
6871 Node* newArr = argument(0);
6872 Node* oldArr = argument(1);
6873 Node* newIdx = argument(2);
6874 Node* shiftCount = argument(3);
6875 Node* numIter = argument(4);
6876
6877 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6878 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6879 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6880 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6881 return false;
6882 }
6883
6884 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6885 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6886 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6887 return false;
6888 }
6889
6890 // Make the call
6891 {
6892 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6893 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6894
6895 Node* call = make_runtime_call(RC_LEAF,
6896 OptoRuntime::bigIntegerShift_Type(),
6897 stubAddr,
6898 stubName,
6899 TypePtr::BOTTOM,
6900 newArr_start,
6901 oldArr_start,
6902 newIdx,
6903 shiftCount,
6904 numIter);
6905 }
6906
6907 return true;
6908 }
6909
6910 //-------------inline_vectorizedMismatch------------------------------
6911 bool LibraryCallKit::inline_vectorizedMismatch() {
6912 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6913
6914 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6915 Node* obja = argument(0); // Object
6916 Node* aoffset = argument(1); // long
6917 Node* objb = argument(3); // Object
6918 Node* boffset = argument(4); // long
6919 Node* length = argument(6); // int
6920 Node* scale = argument(7); // int
6921
6922 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6923 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6924 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6925 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6926 scale == top()) {
6927 return false; // failed input validation
6928 }
6929
6930 Node* obja_adr = make_unsafe_address(obja, aoffset);
6931 Node* objb_adr = make_unsafe_address(objb, boffset);
6932
6933 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6934 //
6935 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6936 // if (length <= inline_limit) {
6937 // inline_path:
6938 // vmask = VectorMaskGen length
6939 // vload1 = LoadVectorMasked obja, vmask
6940 // vload2 = LoadVectorMasked objb, vmask
6941 // result1 = VectorCmpMasked vload1, vload2, vmask
6942 // } else {
6943 // call_stub_path:
6944 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6945 // }
6946 // exit_block:
6947 // return Phi(result1, result2);
6948 //
6949 enum { inline_path = 1, // input is small enough to process it all at once
6950 stub_path = 2, // input is too large; call into the VM
6951 PATH_LIMIT = 3
6952 };
6953
6954 Node* exit_block = new RegionNode(PATH_LIMIT);
6955 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6956 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6957
6958 Node* call_stub_path = control();
6959
6960 BasicType elem_bt = T_ILLEGAL;
6961
6962 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6963 if (scale_t->is_con()) {
6964 switch (scale_t->get_con()) {
6965 case 0: elem_bt = T_BYTE; break;
6966 case 1: elem_bt = T_SHORT; break;
6967 case 2: elem_bt = T_INT; break;
6968 case 3: elem_bt = T_LONG; break;
6969
6970 default: elem_bt = T_ILLEGAL; break; // not supported
6971 }
6972 }
6973
6974 int inline_limit = 0;
6975 bool do_partial_inline = false;
6976
6977 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6978 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6979 do_partial_inline = inline_limit >= 16;
6980 }
6981
6982 if (do_partial_inline) {
6983 assert(elem_bt != T_ILLEGAL, "sanity");
6984
6985 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6986 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6987 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6988
6989 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6990 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6991 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6992
6993 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6994
6995 if (!stopped()) {
6996 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6997
6998 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6999 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7000 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7001 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7002
7003 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7004 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7005 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7006 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7007
7008 exit_block->init_req(inline_path, control());
7009 memory_phi->init_req(inline_path, map()->memory());
7010 result_phi->init_req(inline_path, result);
7011
7012 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7013 clear_upper_avx();
7014 }
7015 }
7016 }
7017
7018 if (call_stub_path != nullptr) {
7019 set_control(call_stub_path);
7020
7021 Node* call = make_runtime_call(RC_LEAF,
7022 OptoRuntime::vectorizedMismatch_Type(),
7023 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7024 obja_adr, objb_adr, length, scale);
7025
7026 exit_block->init_req(stub_path, control());
7027 memory_phi->init_req(stub_path, map()->memory());
7028 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7029 }
7030
7031 exit_block = _gvn.transform(exit_block);
7032 memory_phi = _gvn.transform(memory_phi);
7033 result_phi = _gvn.transform(result_phi);
7034
7035 set_control(exit_block);
7036 set_all_memory(memory_phi);
7037 set_result(result_phi);
7038
7039 return true;
7040 }
7041
7042 //------------------------------inline_vectorizedHashcode----------------------------
7043 bool LibraryCallKit::inline_vectorizedHashCode() {
7044 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7045
7046 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7047 Node* array = argument(0);
7048 Node* offset = argument(1);
7049 Node* length = argument(2);
7050 Node* initialValue = argument(3);
7051 Node* basic_type = argument(4);
7052
7053 if (basic_type == top()) {
7054 return false; // failed input validation
7055 }
7056
7057 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7058 if (!basic_type_t->is_con()) {
7059 return false; // Only intrinsify if mode argument is constant
7060 }
7061
7062 array = must_be_not_null(array, true);
7063
7064 BasicType bt = (BasicType)basic_type_t->get_con();
7065
7066 // Resolve address of first element
7067 Node* array_start = array_element_address(array, offset, bt);
7068
7069 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7070 array_start, length, initialValue, basic_type)));
7071 clear_upper_avx();
7072
7073 return true;
7074 }
7075
7076 /**
7077 * Calculate CRC32 for byte.
7078 * int java.util.zip.CRC32.update(int crc, int b)
7079 */
7080 bool LibraryCallKit::inline_updateCRC32() {
7081 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7082 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7083 // no receiver since it is static method
7084 Node* crc = argument(0); // type: int
7085 Node* b = argument(1); // type: int
7086
7087 /*
7088 * int c = ~ crc;
7089 * b = timesXtoThe32[(b ^ c) & 0xFF];
7090 * b = b ^ (c >>> 8);
7091 * crc = ~b;
7092 */
7093
7094 Node* M1 = intcon(-1);
7095 crc = _gvn.transform(new XorINode(crc, M1));
7096 Node* result = _gvn.transform(new XorINode(crc, b));
7097 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7098
7099 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7100 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7101 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7102 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7103
7104 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7105 result = _gvn.transform(new XorINode(crc, result));
7106 result = _gvn.transform(new XorINode(result, M1));
7107 set_result(result);
7108 return true;
7109 }
7110
7111 /**
7112 * Calculate CRC32 for byte[] array.
7113 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7114 */
7115 bool LibraryCallKit::inline_updateBytesCRC32() {
7116 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7117 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7118 // no receiver since it is static method
7119 Node* crc = argument(0); // type: int
7120 Node* src = argument(1); // type: oop
7121 Node* offset = argument(2); // type: int
7122 Node* length = argument(3); // type: int
7123
7124 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7125 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7126 // failed array check
7127 return false;
7128 }
7129
7130 // Figure out the size and type of the elements we will be copying.
7131 BasicType src_elem = src_type->elem()->array_element_basic_type();
7132 if (src_elem != T_BYTE) {
7133 return false;
7134 }
7135
7136 // 'src_start' points to src array + scaled offset
7137 src = must_be_not_null(src, true);
7138 Node* src_start = array_element_address(src, offset, src_elem);
7139
7140 // We assume that range check is done by caller.
7141 // TODO: generate range check (offset+length < src.length) in debug VM.
7142
7143 // Call the stub.
7144 address stubAddr = StubRoutines::updateBytesCRC32();
7145 const char *stubName = "updateBytesCRC32";
7146
7147 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7148 stubAddr, stubName, TypePtr::BOTTOM,
7149 crc, src_start, length);
7150 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7151 set_result(result);
7152 return true;
7153 }
7154
7155 /**
7156 * Calculate CRC32 for ByteBuffer.
7157 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7158 */
7159 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7160 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7161 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7162 // no receiver since it is static method
7163 Node* crc = argument(0); // type: int
7164 Node* src = argument(1); // type: long
7165 Node* offset = argument(3); // type: int
7166 Node* length = argument(4); // type: int
7167
7168 src = ConvL2X(src); // adjust Java long to machine word
7169 Node* base = _gvn.transform(new CastX2PNode(src));
7170 offset = ConvI2X(offset);
7171
7172 // 'src_start' points to src array + scaled offset
7173 Node* src_start = basic_plus_adr(top(), base, offset);
7174
7175 // Call the stub.
7176 address stubAddr = StubRoutines::updateBytesCRC32();
7177 const char *stubName = "updateBytesCRC32";
7178
7179 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7180 stubAddr, stubName, TypePtr::BOTTOM,
7181 crc, src_start, length);
7182 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7183 set_result(result);
7184 return true;
7185 }
7186
7187 //------------------------------get_table_from_crc32c_class-----------------------
7188 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7189 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7190 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7191
7192 return table;
7193 }
7194
7195 //------------------------------inline_updateBytesCRC32C-----------------------
7196 //
7197 // Calculate CRC32C for byte[] array.
7198 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7199 //
7200 bool LibraryCallKit::inline_updateBytesCRC32C() {
7201 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7202 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7203 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7204 // no receiver since it is a static method
7205 Node* crc = argument(0); // type: int
7206 Node* src = argument(1); // type: oop
7207 Node* offset = argument(2); // type: int
7208 Node* end = argument(3); // type: int
7209
7210 Node* length = _gvn.transform(new SubINode(end, offset));
7211
7212 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7213 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7214 // failed array check
7215 return false;
7216 }
7217
7218 // Figure out the size and type of the elements we will be copying.
7219 BasicType src_elem = src_type->elem()->array_element_basic_type();
7220 if (src_elem != T_BYTE) {
7221 return false;
7222 }
7223
7224 // 'src_start' points to src array + scaled offset
7225 src = must_be_not_null(src, true);
7226 Node* src_start = array_element_address(src, offset, src_elem);
7227
7228 // static final int[] byteTable in class CRC32C
7229 Node* table = get_table_from_crc32c_class(callee()->holder());
7230 table = must_be_not_null(table, true);
7231 Node* table_start = array_element_address(table, intcon(0), T_INT);
7232
7233 // We assume that range check is done by caller.
7234 // TODO: generate range check (offset+length < src.length) in debug VM.
7235
7236 // Call the stub.
7237 address stubAddr = StubRoutines::updateBytesCRC32C();
7238 const char *stubName = "updateBytesCRC32C";
7239
7240 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7241 stubAddr, stubName, TypePtr::BOTTOM,
7242 crc, src_start, length, table_start);
7243 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7244 set_result(result);
7245 return true;
7246 }
7247
7248 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7249 //
7250 // Calculate CRC32C for DirectByteBuffer.
7251 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7252 //
7253 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7254 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7255 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7256 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7257 // no receiver since it is a static method
7258 Node* crc = argument(0); // type: int
7259 Node* src = argument(1); // type: long
7260 Node* offset = argument(3); // type: int
7261 Node* end = argument(4); // type: int
7262
7263 Node* length = _gvn.transform(new SubINode(end, offset));
7264
7265 src = ConvL2X(src); // adjust Java long to machine word
7266 Node* base = _gvn.transform(new CastX2PNode(src));
7267 offset = ConvI2X(offset);
7268
7269 // 'src_start' points to src array + scaled offset
7270 Node* src_start = basic_plus_adr(top(), base, offset);
7271
7272 // static final int[] byteTable in class CRC32C
7273 Node* table = get_table_from_crc32c_class(callee()->holder());
7274 table = must_be_not_null(table, true);
7275 Node* table_start = array_element_address(table, intcon(0), T_INT);
7276
7277 // Call the stub.
7278 address stubAddr = StubRoutines::updateBytesCRC32C();
7279 const char *stubName = "updateBytesCRC32C";
7280
7281 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7282 stubAddr, stubName, TypePtr::BOTTOM,
7283 crc, src_start, length, table_start);
7284 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7285 set_result(result);
7286 return true;
7287 }
7288
7289 //------------------------------inline_updateBytesAdler32----------------------
7290 //
7291 // Calculate Adler32 checksum for byte[] array.
7292 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7293 //
7294 bool LibraryCallKit::inline_updateBytesAdler32() {
7295 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7296 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7297 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7298 // no receiver since it is static method
7299 Node* crc = argument(0); // type: int
7300 Node* src = argument(1); // type: oop
7301 Node* offset = argument(2); // type: int
7302 Node* length = argument(3); // type: int
7303
7304 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7305 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7306 // failed array check
7307 return false;
7308 }
7309
7310 // Figure out the size and type of the elements we will be copying.
7311 BasicType src_elem = src_type->elem()->array_element_basic_type();
7312 if (src_elem != T_BYTE) {
7313 return false;
7314 }
7315
7316 // 'src_start' points to src array + scaled offset
7317 Node* src_start = array_element_address(src, offset, src_elem);
7318
7319 // We assume that range check is done by caller.
7320 // TODO: generate range check (offset+length < src.length) in debug VM.
7321
7322 // Call the stub.
7323 address stubAddr = StubRoutines::updateBytesAdler32();
7324 const char *stubName = "updateBytesAdler32";
7325
7326 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7327 stubAddr, stubName, TypePtr::BOTTOM,
7328 crc, src_start, length);
7329 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7330 set_result(result);
7331 return true;
7332 }
7333
7334 //------------------------------inline_updateByteBufferAdler32---------------
7335 //
7336 // Calculate Adler32 checksum for DirectByteBuffer.
7337 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7338 //
7339 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7340 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7341 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7342 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7343 // no receiver since it is static method
7344 Node* crc = argument(0); // type: int
7345 Node* src = argument(1); // type: long
7346 Node* offset = argument(3); // type: int
7347 Node* length = argument(4); // type: int
7348
7349 src = ConvL2X(src); // adjust Java long to machine word
7350 Node* base = _gvn.transform(new CastX2PNode(src));
7351 offset = ConvI2X(offset);
7352
7353 // 'src_start' points to src array + scaled offset
7354 Node* src_start = basic_plus_adr(top(), base, offset);
7355
7356 // Call the stub.
7357 address stubAddr = StubRoutines::updateBytesAdler32();
7358 const char *stubName = "updateBytesAdler32";
7359
7360 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7361 stubAddr, stubName, TypePtr::BOTTOM,
7362 crc, src_start, length);
7363
7364 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7365 set_result(result);
7366 return true;
7367 }
7368
7369 //----------------------------inline_reference_get----------------------------
7370 // public T java.lang.ref.Reference.get();
7371 bool LibraryCallKit::inline_reference_get() {
7372 const int referent_offset = java_lang_ref_Reference::referent_offset();
7373
7374 // Get the argument:
7375 Node* reference_obj = null_check_receiver();
7376 if (stopped()) return true;
7377
7378 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7379 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7380 decorators, /*is_static*/ false, nullptr);
7381 if (result == nullptr) return false;
7382
7383 // Add memory barrier to prevent commoning reads from this field
7384 // across safepoint since GC can change its value.
7385 insert_mem_bar(Op_MemBarCPUOrder);
7386
7387 set_result(result);
7388 return true;
7389 }
7390
7391 //----------------------------inline_reference_refersTo0----------------------------
7392 // bool java.lang.ref.Reference.refersTo0();
7393 // bool java.lang.ref.PhantomReference.refersTo0();
7394 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7395 // Get arguments:
7396 Node* reference_obj = null_check_receiver();
7397 Node* other_obj = argument(1);
7398 if (stopped()) return true;
7399
7400 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7401 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7402 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7403 decorators, /*is_static*/ false, nullptr);
7404 if (referent == nullptr) return false;
7405
7406 // Add memory barrier to prevent commoning reads from this field
7407 // across safepoint since GC can change its value.
7408 insert_mem_bar(Op_MemBarCPUOrder);
7409
7410 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7411 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7412 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7413
7414 RegionNode* region = new RegionNode(3);
7415 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7416
7417 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7418 region->init_req(1, if_true);
7419 phi->init_req(1, intcon(1));
7420
7421 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7422 region->init_req(2, if_false);
7423 phi->init_req(2, intcon(0));
7424
7425 set_control(_gvn.transform(region));
7426 record_for_igvn(region);
7427 set_result(_gvn.transform(phi));
7428 return true;
7429 }
7430
7431 //----------------------------inline_reference_clear0----------------------------
7432 // void java.lang.ref.Reference.clear0();
7433 // void java.lang.ref.PhantomReference.clear0();
7434 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7435 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7436
7437 // Get arguments
7438 Node* reference_obj = null_check_receiver();
7439 if (stopped()) return true;
7440
7441 // Common access parameters
7442 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7443 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7444 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7445 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7446 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7447
7448 Node* referent = access_load_at(reference_obj,
7449 referent_field_addr,
7450 referent_field_addr_type,
7451 val_type,
7452 T_OBJECT,
7453 decorators);
7454
7455 IdealKit ideal(this);
7456 #define __ ideal.
7457 __ if_then(referent, BoolTest::ne, null());
7458 sync_kit(ideal);
7459 access_store_at(reference_obj,
7460 referent_field_addr,
7461 referent_field_addr_type,
7462 null(),
7463 val_type,
7464 T_OBJECT,
7465 decorators);
7466 __ sync_kit(this);
7467 __ end_if();
7468 final_sync(ideal);
7469 #undef __
7470
7471 return true;
7472 }
7473
7474 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7475 DecoratorSet decorators, bool is_static,
7476 ciInstanceKlass* fromKls) {
7477 if (fromKls == nullptr) {
7478 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7479 assert(tinst != nullptr, "obj is null");
7480 assert(tinst->is_loaded(), "obj is not loaded");
7481 fromKls = tinst->instance_klass();
7482 } else {
7483 assert(is_static, "only for static field access");
7484 }
7485 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7486 ciSymbol::make(fieldTypeString),
7487 is_static);
7488
7489 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7490 if (field == nullptr) return (Node *) nullptr;
7491
7492 if (is_static) {
7493 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7494 fromObj = makecon(tip);
7495 }
7496
7497 // Next code copied from Parse::do_get_xxx():
7498
7499 // Compute address and memory type.
7500 int offset = field->offset_in_bytes();
7501 bool is_vol = field->is_volatile();
7502 ciType* field_klass = field->type();
7503 assert(field_klass->is_loaded(), "should be loaded");
7504 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7505 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7506 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7507 "slice of address and input slice don't match");
7508 BasicType bt = field->layout_type();
7509
7510 // Build the resultant type of the load
7511 const Type *type;
7512 if (bt == T_OBJECT) {
7513 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7514 } else {
7515 type = Type::get_const_basic_type(bt);
7516 }
7517
7518 if (is_vol) {
7519 decorators |= MO_SEQ_CST;
7520 }
7521
7522 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7523 }
7524
7525 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7526 bool is_exact /* true */, bool is_static /* false */,
7527 ciInstanceKlass * fromKls /* nullptr */) {
7528 if (fromKls == nullptr) {
7529 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7530 assert(tinst != nullptr, "obj is null");
7531 assert(tinst->is_loaded(), "obj is not loaded");
7532 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7533 fromKls = tinst->instance_klass();
7534 }
7535 else {
7536 assert(is_static, "only for static field access");
7537 }
7538 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7539 ciSymbol::make(fieldTypeString),
7540 is_static);
7541
7542 assert(field != nullptr, "undefined field");
7543 assert(!field->is_volatile(), "not defined for volatile fields");
7544
7545 if (is_static) {
7546 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7547 fromObj = makecon(tip);
7548 }
7549
7550 // Next code copied from Parse::do_get_xxx():
7551
7552 // Compute address and memory type.
7553 int offset = field->offset_in_bytes();
7554 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7555
7556 return adr;
7557 }
7558
7559 //------------------------------inline_aescrypt_Block-----------------------
7560 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7561 address stubAddr = nullptr;
7562 const char *stubName;
7563 assert(UseAES, "need AES instruction support");
7564
7565 switch(id) {
7566 case vmIntrinsics::_aescrypt_encryptBlock:
7567 stubAddr = StubRoutines::aescrypt_encryptBlock();
7568 stubName = "aescrypt_encryptBlock";
7569 break;
7570 case vmIntrinsics::_aescrypt_decryptBlock:
7571 stubAddr = StubRoutines::aescrypt_decryptBlock();
7572 stubName = "aescrypt_decryptBlock";
7573 break;
7574 default:
7575 break;
7576 }
7577 if (stubAddr == nullptr) return false;
7578
7579 Node* aescrypt_object = argument(0);
7580 Node* src = argument(1);
7581 Node* src_offset = argument(2);
7582 Node* dest = argument(3);
7583 Node* dest_offset = argument(4);
7584
7585 src = must_be_not_null(src, true);
7586 dest = must_be_not_null(dest, true);
7587
7588 // (1) src and dest are arrays.
7589 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7590 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7591 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7592 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7593
7594 // for the quick and dirty code we will skip all the checks.
7595 // we are just trying to get the call to be generated.
7596 Node* src_start = src;
7597 Node* dest_start = dest;
7598 if (src_offset != nullptr || dest_offset != nullptr) {
7599 assert(src_offset != nullptr && dest_offset != nullptr, "");
7600 src_start = array_element_address(src, src_offset, T_BYTE);
7601 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7602 }
7603
7604 // now need to get the start of its expanded key array
7605 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7606 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7607 if (k_start == nullptr) return false;
7608
7609 // Call the stub.
7610 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7611 stubAddr, stubName, TypePtr::BOTTOM,
7612 src_start, dest_start, k_start);
7613
7614 return true;
7615 }
7616
7617 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7618 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7619 address stubAddr = nullptr;
7620 const char *stubName = nullptr;
7621
7622 assert(UseAES, "need AES instruction support");
7623
7624 switch(id) {
7625 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7626 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7627 stubName = "cipherBlockChaining_encryptAESCrypt";
7628 break;
7629 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7630 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7631 stubName = "cipherBlockChaining_decryptAESCrypt";
7632 break;
7633 default:
7634 break;
7635 }
7636 if (stubAddr == nullptr) return false;
7637
7638 Node* cipherBlockChaining_object = argument(0);
7639 Node* src = argument(1);
7640 Node* src_offset = argument(2);
7641 Node* len = argument(3);
7642 Node* dest = argument(4);
7643 Node* dest_offset = argument(5);
7644
7645 src = must_be_not_null(src, false);
7646 dest = must_be_not_null(dest, false);
7647
7648 // (1) src and dest are arrays.
7649 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7650 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7651 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7652 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7653
7654 // checks are the responsibility of the caller
7655 Node* src_start = src;
7656 Node* dest_start = dest;
7657 if (src_offset != nullptr || dest_offset != nullptr) {
7658 assert(src_offset != nullptr && dest_offset != nullptr, "");
7659 src_start = array_element_address(src, src_offset, T_BYTE);
7660 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7661 }
7662
7663 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7664 // (because of the predicated logic executed earlier).
7665 // so we cast it here safely.
7666 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7667
7668 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7669 if (embeddedCipherObj == nullptr) return false;
7670
7671 // cast it to what we know it will be at runtime
7672 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7673 assert(tinst != nullptr, "CBC obj is null");
7674 assert(tinst->is_loaded(), "CBC obj is not loaded");
7675 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7676 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7677
7678 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7679 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7680 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7681 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7682 aescrypt_object = _gvn.transform(aescrypt_object);
7683
7684 // we need to get the start of the aescrypt_object's expanded key array
7685 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7686 if (k_start == nullptr) return false;
7687
7688 // similarly, get the start address of the r vector
7689 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7690 if (objRvec == nullptr) return false;
7691 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7692
7693 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7694 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7695 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7696 stubAddr, stubName, TypePtr::BOTTOM,
7697 src_start, dest_start, k_start, r_start, len);
7698
7699 // return cipher length (int)
7700 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7701 set_result(retvalue);
7702 return true;
7703 }
7704
7705 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7706 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7707 address stubAddr = nullptr;
7708 const char *stubName = nullptr;
7709
7710 assert(UseAES, "need AES instruction support");
7711
7712 switch (id) {
7713 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7714 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7715 stubName = "electronicCodeBook_encryptAESCrypt";
7716 break;
7717 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7718 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7719 stubName = "electronicCodeBook_decryptAESCrypt";
7720 break;
7721 default:
7722 break;
7723 }
7724
7725 if (stubAddr == nullptr) return false;
7726
7727 Node* electronicCodeBook_object = argument(0);
7728 Node* src = argument(1);
7729 Node* src_offset = argument(2);
7730 Node* len = argument(3);
7731 Node* dest = argument(4);
7732 Node* dest_offset = argument(5);
7733
7734 // (1) src and dest are arrays.
7735 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7736 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7737 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7738 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7739
7740 // checks are the responsibility of the caller
7741 Node* src_start = src;
7742 Node* dest_start = dest;
7743 if (src_offset != nullptr || dest_offset != nullptr) {
7744 assert(src_offset != nullptr && dest_offset != nullptr, "");
7745 src_start = array_element_address(src, src_offset, T_BYTE);
7746 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7747 }
7748
7749 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7750 // (because of the predicated logic executed earlier).
7751 // so we cast it here safely.
7752 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7753
7754 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7755 if (embeddedCipherObj == nullptr) return false;
7756
7757 // cast it to what we know it will be at runtime
7758 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7759 assert(tinst != nullptr, "ECB obj is null");
7760 assert(tinst->is_loaded(), "ECB obj is not loaded");
7761 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7762 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7763
7764 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7765 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7766 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7767 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7768 aescrypt_object = _gvn.transform(aescrypt_object);
7769
7770 // we need to get the start of the aescrypt_object's expanded key array
7771 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7772 if (k_start == nullptr) return false;
7773
7774 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7775 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7776 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7777 stubAddr, stubName, TypePtr::BOTTOM,
7778 src_start, dest_start, k_start, len);
7779
7780 // return cipher length (int)
7781 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7782 set_result(retvalue);
7783 return true;
7784 }
7785
7786 //------------------------------inline_counterMode_AESCrypt-----------------------
7787 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7788 assert(UseAES, "need AES instruction support");
7789 if (!UseAESCTRIntrinsics) return false;
7790
7791 address stubAddr = nullptr;
7792 const char *stubName = nullptr;
7793 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7794 stubAddr = StubRoutines::counterMode_AESCrypt();
7795 stubName = "counterMode_AESCrypt";
7796 }
7797 if (stubAddr == nullptr) return false;
7798
7799 Node* counterMode_object = argument(0);
7800 Node* src = argument(1);
7801 Node* src_offset = argument(2);
7802 Node* len = argument(3);
7803 Node* dest = argument(4);
7804 Node* dest_offset = argument(5);
7805
7806 // (1) src and dest are arrays.
7807 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7808 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7809 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7810 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7811
7812 // checks are the responsibility of the caller
7813 Node* src_start = src;
7814 Node* dest_start = dest;
7815 if (src_offset != nullptr || dest_offset != nullptr) {
7816 assert(src_offset != nullptr && dest_offset != nullptr, "");
7817 src_start = array_element_address(src, src_offset, T_BYTE);
7818 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7819 }
7820
7821 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7822 // (because of the predicated logic executed earlier).
7823 // so we cast it here safely.
7824 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7825 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7826 if (embeddedCipherObj == nullptr) return false;
7827 // cast it to what we know it will be at runtime
7828 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7829 assert(tinst != nullptr, "CTR obj is null");
7830 assert(tinst->is_loaded(), "CTR obj is not loaded");
7831 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7832 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7833 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7834 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7835 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7836 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7837 aescrypt_object = _gvn.transform(aescrypt_object);
7838 // we need to get the start of the aescrypt_object's expanded key array
7839 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7840 if (k_start == nullptr) return false;
7841 // similarly, get the start address of the r vector
7842 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7843 if (obj_counter == nullptr) return false;
7844 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7845
7846 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7847 if (saved_encCounter == nullptr) return false;
7848 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7849 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7850
7851 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7852 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7853 OptoRuntime::counterMode_aescrypt_Type(),
7854 stubAddr, stubName, TypePtr::BOTTOM,
7855 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7856
7857 // return cipher length (int)
7858 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7859 set_result(retvalue);
7860 return true;
7861 }
7862
7863 //------------------------------get_key_start_from_aescrypt_object-----------------------
7864 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7865 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7866 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7867 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7868 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7869 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7870 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7871 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7872 if (objSessionK == nullptr) {
7873 return (Node *) nullptr;
7874 }
7875 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7876 #else
7877 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7878 #endif // PPC64
7879 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7880 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7881
7882 // now have the array, need to get the start address of the K array
7883 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7884 return k_start;
7885 }
7886
7887 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7888 // Return node representing slow path of predicate check.
7889 // the pseudo code we want to emulate with this predicate is:
7890 // for encryption:
7891 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7892 // for decryption:
7893 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7894 // note cipher==plain is more conservative than the original java code but that's OK
7895 //
7896 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7897 // The receiver was checked for null already.
7898 Node* objCBC = argument(0);
7899
7900 Node* src = argument(1);
7901 Node* dest = argument(4);
7902
7903 // Load embeddedCipher field of CipherBlockChaining object.
7904 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7905
7906 // get AESCrypt klass for instanceOf check
7907 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7908 // will have same classloader as CipherBlockChaining object
7909 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7910 assert(tinst != nullptr, "CBCobj is null");
7911 assert(tinst->is_loaded(), "CBCobj is not loaded");
7912
7913 // we want to do an instanceof comparison against the AESCrypt class
7914 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7915 if (!klass_AESCrypt->is_loaded()) {
7916 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7917 Node* ctrl = control();
7918 set_control(top()); // no regular fast path
7919 return ctrl;
7920 }
7921
7922 src = must_be_not_null(src, true);
7923 dest = must_be_not_null(dest, true);
7924
7925 // Resolve oops to stable for CmpP below.
7926 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7927
7928 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7929 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7930 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7931
7932 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7933
7934 // for encryption, we are done
7935 if (!decrypting)
7936 return instof_false; // even if it is null
7937
7938 // for decryption, we need to add a further check to avoid
7939 // taking the intrinsic path when cipher and plain are the same
7940 // see the original java code for why.
7941 RegionNode* region = new RegionNode(3);
7942 region->init_req(1, instof_false);
7943
7944 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7945 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7946 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7947 region->init_req(2, src_dest_conjoint);
7948
7949 record_for_igvn(region);
7950 return _gvn.transform(region);
7951 }
7952
7953 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7954 // Return node representing slow path of predicate check.
7955 // the pseudo code we want to emulate with this predicate is:
7956 // for encryption:
7957 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7958 // for decryption:
7959 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7960 // note cipher==plain is more conservative than the original java code but that's OK
7961 //
7962 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7963 // The receiver was checked for null already.
7964 Node* objECB = argument(0);
7965
7966 // Load embeddedCipher field of ElectronicCodeBook object.
7967 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7968
7969 // get AESCrypt klass for instanceOf check
7970 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7971 // will have same classloader as ElectronicCodeBook object
7972 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7973 assert(tinst != nullptr, "ECBobj is null");
7974 assert(tinst->is_loaded(), "ECBobj is not loaded");
7975
7976 // we want to do an instanceof comparison against the AESCrypt class
7977 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7978 if (!klass_AESCrypt->is_loaded()) {
7979 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7980 Node* ctrl = control();
7981 set_control(top()); // no regular fast path
7982 return ctrl;
7983 }
7984 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7985
7986 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7987 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7988 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7989
7990 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7991
7992 // for encryption, we are done
7993 if (!decrypting)
7994 return instof_false; // even if it is null
7995
7996 // for decryption, we need to add a further check to avoid
7997 // taking the intrinsic path when cipher and plain are the same
7998 // see the original java code for why.
7999 RegionNode* region = new RegionNode(3);
8000 region->init_req(1, instof_false);
8001 Node* src = argument(1);
8002 Node* dest = argument(4);
8003 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8004 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8005 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8006 region->init_req(2, src_dest_conjoint);
8007
8008 record_for_igvn(region);
8009 return _gvn.transform(region);
8010 }
8011
8012 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8013 // Return node representing slow path of predicate check.
8014 // the pseudo code we want to emulate with this predicate is:
8015 // for encryption:
8016 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8017 // for decryption:
8018 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8019 // note cipher==plain is more conservative than the original java code but that's OK
8020 //
8021
8022 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8023 // The receiver was checked for null already.
8024 Node* objCTR = argument(0);
8025
8026 // Load embeddedCipher field of CipherBlockChaining object.
8027 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8028
8029 // get AESCrypt klass for instanceOf check
8030 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8031 // will have same classloader as CipherBlockChaining object
8032 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8033 assert(tinst != nullptr, "CTRobj is null");
8034 assert(tinst->is_loaded(), "CTRobj is not loaded");
8035
8036 // we want to do an instanceof comparison against the AESCrypt class
8037 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8038 if (!klass_AESCrypt->is_loaded()) {
8039 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8040 Node* ctrl = control();
8041 set_control(top()); // no regular fast path
8042 return ctrl;
8043 }
8044
8045 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8046 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8047 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8048 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8049 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8050
8051 return instof_false; // even if it is null
8052 }
8053
8054 //------------------------------inline_ghash_processBlocks
8055 bool LibraryCallKit::inline_ghash_processBlocks() {
8056 address stubAddr;
8057 const char *stubName;
8058 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8059
8060 stubAddr = StubRoutines::ghash_processBlocks();
8061 stubName = "ghash_processBlocks";
8062
8063 Node* data = argument(0);
8064 Node* offset = argument(1);
8065 Node* len = argument(2);
8066 Node* state = argument(3);
8067 Node* subkeyH = argument(4);
8068
8069 state = must_be_not_null(state, true);
8070 subkeyH = must_be_not_null(subkeyH, true);
8071 data = must_be_not_null(data, true);
8072
8073 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8074 assert(state_start, "state is null");
8075 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8076 assert(subkeyH_start, "subkeyH is null");
8077 Node* data_start = array_element_address(data, offset, T_BYTE);
8078 assert(data_start, "data is null");
8079
8080 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8081 OptoRuntime::ghash_processBlocks_Type(),
8082 stubAddr, stubName, TypePtr::BOTTOM,
8083 state_start, subkeyH_start, data_start, len);
8084 return true;
8085 }
8086
8087 //------------------------------inline_chacha20Block
8088 bool LibraryCallKit::inline_chacha20Block() {
8089 address stubAddr;
8090 const char *stubName;
8091 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8092
8093 stubAddr = StubRoutines::chacha20Block();
8094 stubName = "chacha20Block";
8095
8096 Node* state = argument(0);
8097 Node* result = argument(1);
8098
8099 state = must_be_not_null(state, true);
8100 result = must_be_not_null(result, true);
8101
8102 Node* state_start = array_element_address(state, intcon(0), T_INT);
8103 assert(state_start, "state is null");
8104 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8105 assert(result_start, "result is null");
8106
8107 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8108 OptoRuntime::chacha20Block_Type(),
8109 stubAddr, stubName, TypePtr::BOTTOM,
8110 state_start, result_start);
8111 // return key stream length (int)
8112 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8113 set_result(retvalue);
8114 return true;
8115 }
8116
8117 //------------------------------inline_kyberNtt
8118 bool LibraryCallKit::inline_kyberNtt() {
8119 address stubAddr;
8120 const char *stubName;
8121 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8122 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8123
8124 stubAddr = StubRoutines::kyberNtt();
8125 stubName = "kyberNtt";
8126 if (!stubAddr) return false;
8127
8128 Node* coeffs = argument(0);
8129 Node* ntt_zetas = argument(1);
8130
8131 coeffs = must_be_not_null(coeffs, true);
8132 ntt_zetas = must_be_not_null(ntt_zetas, true);
8133
8134 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8135 assert(coeffs_start, "coeffs is null");
8136 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8137 assert(ntt_zetas_start, "ntt_zetas is null");
8138 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8139 OptoRuntime::kyberNtt_Type(),
8140 stubAddr, stubName, TypePtr::BOTTOM,
8141 coeffs_start, ntt_zetas_start);
8142 // return an int
8143 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8144 set_result(retvalue);
8145 return true;
8146 }
8147
8148 //------------------------------inline_kyberInverseNtt
8149 bool LibraryCallKit::inline_kyberInverseNtt() {
8150 address stubAddr;
8151 const char *stubName;
8152 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8153 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8154
8155 stubAddr = StubRoutines::kyberInverseNtt();
8156 stubName = "kyberInverseNtt";
8157 if (!stubAddr) return false;
8158
8159 Node* coeffs = argument(0);
8160 Node* zetas = argument(1);
8161
8162 coeffs = must_be_not_null(coeffs, true);
8163 zetas = must_be_not_null(zetas, true);
8164
8165 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8166 assert(coeffs_start, "coeffs is null");
8167 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8168 assert(zetas_start, "inverseNtt_zetas is null");
8169 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8170 OptoRuntime::kyberInverseNtt_Type(),
8171 stubAddr, stubName, TypePtr::BOTTOM,
8172 coeffs_start, zetas_start);
8173
8174 // return an int
8175 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8176 set_result(retvalue);
8177 return true;
8178 }
8179
8180 //------------------------------inline_kyberNttMult
8181 bool LibraryCallKit::inline_kyberNttMult() {
8182 address stubAddr;
8183 const char *stubName;
8184 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8185 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8186
8187 stubAddr = StubRoutines::kyberNttMult();
8188 stubName = "kyberNttMult";
8189 if (!stubAddr) return false;
8190
8191 Node* result = argument(0);
8192 Node* ntta = argument(1);
8193 Node* nttb = argument(2);
8194 Node* zetas = argument(3);
8195
8196 result = must_be_not_null(result, true);
8197 ntta = must_be_not_null(ntta, true);
8198 nttb = must_be_not_null(nttb, true);
8199 zetas = must_be_not_null(zetas, true);
8200 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8201 assert(result_start, "result is null");
8202 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8203 assert(ntta_start, "ntta is null");
8204 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8205 assert(nttb_start, "nttb is null");
8206 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8207 assert(zetas_start, "nttMult_zetas is null");
8208 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8209 OptoRuntime::kyberNttMult_Type(),
8210 stubAddr, stubName, TypePtr::BOTTOM,
8211 result_start, ntta_start, nttb_start,
8212 zetas_start);
8213
8214 // return an int
8215 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8216 set_result(retvalue);
8217
8218 return true;
8219 }
8220
8221 //------------------------------inline_kyberAddPoly_2
8222 bool LibraryCallKit::inline_kyberAddPoly_2() {
8223 address stubAddr;
8224 const char *stubName;
8225 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8226 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8227
8228 stubAddr = StubRoutines::kyberAddPoly_2();
8229 stubName = "kyberAddPoly_2";
8230 if (!stubAddr) return false;
8231
8232 Node* result = argument(0);
8233 Node* a = argument(1);
8234 Node* b = argument(2);
8235
8236 result = must_be_not_null(result, true);
8237 a = must_be_not_null(a, true);
8238 b = must_be_not_null(b, true);
8239
8240 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8241 assert(result_start, "result is null");
8242 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8243 assert(a_start, "a is null");
8244 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8245 assert(b_start, "b is null");
8246 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8247 OptoRuntime::kyberAddPoly_2_Type(),
8248 stubAddr, stubName, TypePtr::BOTTOM,
8249 result_start, a_start, b_start);
8250 // return an int
8251 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8252 set_result(retvalue);
8253 return true;
8254 }
8255
8256 //------------------------------inline_kyberAddPoly_3
8257 bool LibraryCallKit::inline_kyberAddPoly_3() {
8258 address stubAddr;
8259 const char *stubName;
8260 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8261 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8262
8263 stubAddr = StubRoutines::kyberAddPoly_3();
8264 stubName = "kyberAddPoly_3";
8265 if (!stubAddr) return false;
8266
8267 Node* result = argument(0);
8268 Node* a = argument(1);
8269 Node* b = argument(2);
8270 Node* c = argument(3);
8271
8272 result = must_be_not_null(result, true);
8273 a = must_be_not_null(a, true);
8274 b = must_be_not_null(b, true);
8275 c = must_be_not_null(c, true);
8276
8277 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8278 assert(result_start, "result is null");
8279 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8280 assert(a_start, "a is null");
8281 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8282 assert(b_start, "b is null");
8283 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8284 assert(c_start, "c is null");
8285 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8286 OptoRuntime::kyberAddPoly_3_Type(),
8287 stubAddr, stubName, TypePtr::BOTTOM,
8288 result_start, a_start, b_start, c_start);
8289 // return an int
8290 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8291 set_result(retvalue);
8292 return true;
8293 }
8294
8295 //------------------------------inline_kyber12To16
8296 bool LibraryCallKit::inline_kyber12To16() {
8297 address stubAddr;
8298 const char *stubName;
8299 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8300 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8301
8302 stubAddr = StubRoutines::kyber12To16();
8303 stubName = "kyber12To16";
8304 if (!stubAddr) return false;
8305
8306 Node* condensed = argument(0);
8307 Node* condensedOffs = argument(1);
8308 Node* parsed = argument(2);
8309 Node* parsedLength = argument(3);
8310
8311 condensed = must_be_not_null(condensed, true);
8312 parsed = must_be_not_null(parsed, true);
8313
8314 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8315 assert(condensed_start, "condensed is null");
8316 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8317 assert(parsed_start, "parsed is null");
8318 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8319 OptoRuntime::kyber12To16_Type(),
8320 stubAddr, stubName, TypePtr::BOTTOM,
8321 condensed_start, condensedOffs, parsed_start, parsedLength);
8322 // return an int
8323 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8324 set_result(retvalue);
8325 return true;
8326
8327 }
8328
8329 //------------------------------inline_kyberBarrettReduce
8330 bool LibraryCallKit::inline_kyberBarrettReduce() {
8331 address stubAddr;
8332 const char *stubName;
8333 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8334 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8335
8336 stubAddr = StubRoutines::kyberBarrettReduce();
8337 stubName = "kyberBarrettReduce";
8338 if (!stubAddr) return false;
8339
8340 Node* coeffs = argument(0);
8341
8342 coeffs = must_be_not_null(coeffs, true);
8343
8344 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8345 assert(coeffs_start, "coeffs is null");
8346 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8347 OptoRuntime::kyberBarrettReduce_Type(),
8348 stubAddr, stubName, TypePtr::BOTTOM,
8349 coeffs_start);
8350 // return an int
8351 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8352 set_result(retvalue);
8353 return true;
8354 }
8355
8356 //------------------------------inline_dilithiumAlmostNtt
8357 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8358 address stubAddr;
8359 const char *stubName;
8360 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8361 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8362
8363 stubAddr = StubRoutines::dilithiumAlmostNtt();
8364 stubName = "dilithiumAlmostNtt";
8365 if (!stubAddr) return false;
8366
8367 Node* coeffs = argument(0);
8368 Node* ntt_zetas = argument(1);
8369
8370 coeffs = must_be_not_null(coeffs, true);
8371 ntt_zetas = must_be_not_null(ntt_zetas, true);
8372
8373 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8374 assert(coeffs_start, "coeffs is null");
8375 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8376 assert(ntt_zetas_start, "ntt_zetas is null");
8377 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8378 OptoRuntime::dilithiumAlmostNtt_Type(),
8379 stubAddr, stubName, TypePtr::BOTTOM,
8380 coeffs_start, ntt_zetas_start);
8381 // return an int
8382 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8383 set_result(retvalue);
8384 return true;
8385 }
8386
8387 //------------------------------inline_dilithiumAlmostInverseNtt
8388 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8389 address stubAddr;
8390 const char *stubName;
8391 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8392 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8393
8394 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8395 stubName = "dilithiumAlmostInverseNtt";
8396 if (!stubAddr) return false;
8397
8398 Node* coeffs = argument(0);
8399 Node* zetas = argument(1);
8400
8401 coeffs = must_be_not_null(coeffs, true);
8402 zetas = must_be_not_null(zetas, true);
8403
8404 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8405 assert(coeffs_start, "coeffs is null");
8406 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8407 assert(zetas_start, "inverseNtt_zetas is null");
8408 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8409 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8410 stubAddr, stubName, TypePtr::BOTTOM,
8411 coeffs_start, zetas_start);
8412 // return an int
8413 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8414 set_result(retvalue);
8415 return true;
8416 }
8417
8418 //------------------------------inline_dilithiumNttMult
8419 bool LibraryCallKit::inline_dilithiumNttMult() {
8420 address stubAddr;
8421 const char *stubName;
8422 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8423 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8424
8425 stubAddr = StubRoutines::dilithiumNttMult();
8426 stubName = "dilithiumNttMult";
8427 if (!stubAddr) return false;
8428
8429 Node* result = argument(0);
8430 Node* ntta = argument(1);
8431 Node* nttb = argument(2);
8432 Node* zetas = argument(3);
8433
8434 result = must_be_not_null(result, true);
8435 ntta = must_be_not_null(ntta, true);
8436 nttb = must_be_not_null(nttb, true);
8437 zetas = must_be_not_null(zetas, true);
8438
8439 Node* result_start = array_element_address(result, intcon(0), T_INT);
8440 assert(result_start, "result is null");
8441 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8442 assert(ntta_start, "ntta is null");
8443 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8444 assert(nttb_start, "nttb is null");
8445 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8446 OptoRuntime::dilithiumNttMult_Type(),
8447 stubAddr, stubName, TypePtr::BOTTOM,
8448 result_start, ntta_start, nttb_start);
8449
8450 // return an int
8451 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8452 set_result(retvalue);
8453
8454 return true;
8455 }
8456
8457 //------------------------------inline_dilithiumMontMulByConstant
8458 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8459 address stubAddr;
8460 const char *stubName;
8461 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8462 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8463
8464 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8465 stubName = "dilithiumMontMulByConstant";
8466 if (!stubAddr) return false;
8467
8468 Node* coeffs = argument(0);
8469 Node* constant = argument(1);
8470
8471 coeffs = must_be_not_null(coeffs, true);
8472
8473 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8474 assert(coeffs_start, "coeffs is null");
8475 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8476 OptoRuntime::dilithiumMontMulByConstant_Type(),
8477 stubAddr, stubName, TypePtr::BOTTOM,
8478 coeffs_start, constant);
8479
8480 // return an int
8481 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8482 set_result(retvalue);
8483 return true;
8484 }
8485
8486
8487 //------------------------------inline_dilithiumDecomposePoly
8488 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8489 address stubAddr;
8490 const char *stubName;
8491 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8492 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8493
8494 stubAddr = StubRoutines::dilithiumDecomposePoly();
8495 stubName = "dilithiumDecomposePoly";
8496 if (!stubAddr) return false;
8497
8498 Node* input = argument(0);
8499 Node* lowPart = argument(1);
8500 Node* highPart = argument(2);
8501 Node* twoGamma2 = argument(3);
8502 Node* multiplier = argument(4);
8503
8504 input = must_be_not_null(input, true);
8505 lowPart = must_be_not_null(lowPart, true);
8506 highPart = must_be_not_null(highPart, true);
8507
8508 Node* input_start = array_element_address(input, intcon(0), T_INT);
8509 assert(input_start, "input is null");
8510 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8511 assert(lowPart_start, "lowPart is null");
8512 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8513 assert(highPart_start, "highPart is null");
8514
8515 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8516 OptoRuntime::dilithiumDecomposePoly_Type(),
8517 stubAddr, stubName, TypePtr::BOTTOM,
8518 input_start, lowPart_start, highPart_start,
8519 twoGamma2, multiplier);
8520
8521 // return an int
8522 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8523 set_result(retvalue);
8524 return true;
8525 }
8526
8527 bool LibraryCallKit::inline_base64_encodeBlock() {
8528 address stubAddr;
8529 const char *stubName;
8530 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8531 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8532 stubAddr = StubRoutines::base64_encodeBlock();
8533 stubName = "encodeBlock";
8534
8535 if (!stubAddr) return false;
8536 Node* base64obj = argument(0);
8537 Node* src = argument(1);
8538 Node* offset = argument(2);
8539 Node* len = argument(3);
8540 Node* dest = argument(4);
8541 Node* dp = argument(5);
8542 Node* isURL = argument(6);
8543
8544 src = must_be_not_null(src, true);
8545 dest = must_be_not_null(dest, true);
8546
8547 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8548 assert(src_start, "source array is null");
8549 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8550 assert(dest_start, "destination array is null");
8551
8552 Node* base64 = make_runtime_call(RC_LEAF,
8553 OptoRuntime::base64_encodeBlock_Type(),
8554 stubAddr, stubName, TypePtr::BOTTOM,
8555 src_start, offset, len, dest_start, dp, isURL);
8556 return true;
8557 }
8558
8559 bool LibraryCallKit::inline_base64_decodeBlock() {
8560 address stubAddr;
8561 const char *stubName;
8562 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8563 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8564 stubAddr = StubRoutines::base64_decodeBlock();
8565 stubName = "decodeBlock";
8566
8567 if (!stubAddr) return false;
8568 Node* base64obj = argument(0);
8569 Node* src = argument(1);
8570 Node* src_offset = argument(2);
8571 Node* len = argument(3);
8572 Node* dest = argument(4);
8573 Node* dest_offset = argument(5);
8574 Node* isURL = argument(6);
8575 Node* isMIME = argument(7);
8576
8577 src = must_be_not_null(src, true);
8578 dest = must_be_not_null(dest, true);
8579
8580 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8581 assert(src_start, "source array is null");
8582 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8583 assert(dest_start, "destination array is null");
8584
8585 Node* call = make_runtime_call(RC_LEAF,
8586 OptoRuntime::base64_decodeBlock_Type(),
8587 stubAddr, stubName, TypePtr::BOTTOM,
8588 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8589 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8590 set_result(result);
8591 return true;
8592 }
8593
8594 bool LibraryCallKit::inline_poly1305_processBlocks() {
8595 address stubAddr;
8596 const char *stubName;
8597 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8598 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8599 stubAddr = StubRoutines::poly1305_processBlocks();
8600 stubName = "poly1305_processBlocks";
8601
8602 if (!stubAddr) return false;
8603 null_check_receiver(); // null-check receiver
8604 if (stopped()) return true;
8605
8606 Node* input = argument(1);
8607 Node* input_offset = argument(2);
8608 Node* len = argument(3);
8609 Node* alimbs = argument(4);
8610 Node* rlimbs = argument(5);
8611
8612 input = must_be_not_null(input, true);
8613 alimbs = must_be_not_null(alimbs, true);
8614 rlimbs = must_be_not_null(rlimbs, true);
8615
8616 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8617 assert(input_start, "input array is null");
8618 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8619 assert(acc_start, "acc array is null");
8620 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8621 assert(r_start, "r array is null");
8622
8623 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8624 OptoRuntime::poly1305_processBlocks_Type(),
8625 stubAddr, stubName, TypePtr::BOTTOM,
8626 input_start, len, acc_start, r_start);
8627 return true;
8628 }
8629
8630 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8631 address stubAddr;
8632 const char *stubName;
8633 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8634 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8635 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8636 stubName = "intpoly_montgomeryMult_P256";
8637
8638 if (!stubAddr) return false;
8639 null_check_receiver(); // null-check receiver
8640 if (stopped()) return true;
8641
8642 Node* a = argument(1);
8643 Node* b = argument(2);
8644 Node* r = argument(3);
8645
8646 a = must_be_not_null(a, true);
8647 b = must_be_not_null(b, true);
8648 r = must_be_not_null(r, true);
8649
8650 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8651 assert(a_start, "a array is null");
8652 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8653 assert(b_start, "b array is null");
8654 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8655 assert(r_start, "r array is null");
8656
8657 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8658 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8659 stubAddr, stubName, TypePtr::BOTTOM,
8660 a_start, b_start, r_start);
8661 return true;
8662 }
8663
8664 bool LibraryCallKit::inline_intpoly_assign() {
8665 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8666 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8667 const char *stubName = "intpoly_assign";
8668 address stubAddr = StubRoutines::intpoly_assign();
8669 if (!stubAddr) return false;
8670
8671 Node* set = argument(0);
8672 Node* a = argument(1);
8673 Node* b = argument(2);
8674 Node* arr_length = load_array_length(a);
8675
8676 a = must_be_not_null(a, true);
8677 b = must_be_not_null(b, true);
8678
8679 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8680 assert(a_start, "a array is null");
8681 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8682 assert(b_start, "b array is null");
8683
8684 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8685 OptoRuntime::intpoly_assign_Type(),
8686 stubAddr, stubName, TypePtr::BOTTOM,
8687 set, a_start, b_start, arr_length);
8688 return true;
8689 }
8690
8691 //------------------------------inline_digestBase_implCompress-----------------------
8692 //
8693 // Calculate MD5 for single-block byte[] array.
8694 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8695 //
8696 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8697 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8698 //
8699 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8700 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8701 //
8702 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8703 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8704 //
8705 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8706 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8707 //
8708 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8709 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8710
8711 Node* digestBase_obj = argument(0);
8712 Node* src = argument(1); // type oop
8713 Node* ofs = argument(2); // type int
8714
8715 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8716 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8717 // failed array check
8718 return false;
8719 }
8720 // Figure out the size and type of the elements we will be copying.
8721 BasicType src_elem = src_type->elem()->array_element_basic_type();
8722 if (src_elem != T_BYTE) {
8723 return false;
8724 }
8725 // 'src_start' points to src array + offset
8726 src = must_be_not_null(src, true);
8727 Node* src_start = array_element_address(src, ofs, src_elem);
8728 Node* state = nullptr;
8729 Node* block_size = nullptr;
8730 address stubAddr;
8731 const char *stubName;
8732
8733 switch(id) {
8734 case vmIntrinsics::_md5_implCompress:
8735 assert(UseMD5Intrinsics, "need MD5 instruction support");
8736 state = get_state_from_digest_object(digestBase_obj, T_INT);
8737 stubAddr = StubRoutines::md5_implCompress();
8738 stubName = "md5_implCompress";
8739 break;
8740 case vmIntrinsics::_sha_implCompress:
8741 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8742 state = get_state_from_digest_object(digestBase_obj, T_INT);
8743 stubAddr = StubRoutines::sha1_implCompress();
8744 stubName = "sha1_implCompress";
8745 break;
8746 case vmIntrinsics::_sha2_implCompress:
8747 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8748 state = get_state_from_digest_object(digestBase_obj, T_INT);
8749 stubAddr = StubRoutines::sha256_implCompress();
8750 stubName = "sha256_implCompress";
8751 break;
8752 case vmIntrinsics::_sha5_implCompress:
8753 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8754 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8755 stubAddr = StubRoutines::sha512_implCompress();
8756 stubName = "sha512_implCompress";
8757 break;
8758 case vmIntrinsics::_sha3_implCompress:
8759 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8760 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8761 stubAddr = StubRoutines::sha3_implCompress();
8762 stubName = "sha3_implCompress";
8763 block_size = get_block_size_from_digest_object(digestBase_obj);
8764 if (block_size == nullptr) return false;
8765 break;
8766 default:
8767 fatal_unexpected_iid(id);
8768 return false;
8769 }
8770 if (state == nullptr) return false;
8771
8772 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8773 if (stubAddr == nullptr) return false;
8774
8775 // Call the stub.
8776 Node* call;
8777 if (block_size == nullptr) {
8778 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8779 stubAddr, stubName, TypePtr::BOTTOM,
8780 src_start, state);
8781 } else {
8782 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8783 stubAddr, stubName, TypePtr::BOTTOM,
8784 src_start, state, block_size);
8785 }
8786
8787 return true;
8788 }
8789
8790 //------------------------------inline_double_keccak
8791 bool LibraryCallKit::inline_double_keccak() {
8792 address stubAddr;
8793 const char *stubName;
8794 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8795 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8796
8797 stubAddr = StubRoutines::double_keccak();
8798 stubName = "double_keccak";
8799 if (!stubAddr) return false;
8800
8801 Node* status0 = argument(0);
8802 Node* status1 = argument(1);
8803
8804 status0 = must_be_not_null(status0, true);
8805 status1 = must_be_not_null(status1, true);
8806
8807 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8808 assert(status0_start, "status0 is null");
8809 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8810 assert(status1_start, "status1 is null");
8811 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8812 OptoRuntime::double_keccak_Type(),
8813 stubAddr, stubName, TypePtr::BOTTOM,
8814 status0_start, status1_start);
8815 // return an int
8816 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8817 set_result(retvalue);
8818 return true;
8819 }
8820
8821
8822 //------------------------------inline_digestBase_implCompressMB-----------------------
8823 //
8824 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8825 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8826 //
8827 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8828 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8829 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8830 assert((uint)predicate < 5, "sanity");
8831 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8832
8833 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8834 Node* src = argument(1); // byte[] array
8835 Node* ofs = argument(2); // type int
8836 Node* limit = argument(3); // type int
8837
8838 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8839 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8840 // failed array check
8841 return false;
8842 }
8843 // Figure out the size and type of the elements we will be copying.
8844 BasicType src_elem = src_type->elem()->array_element_basic_type();
8845 if (src_elem != T_BYTE) {
8846 return false;
8847 }
8848 // 'src_start' points to src array + offset
8849 src = must_be_not_null(src, false);
8850 Node* src_start = array_element_address(src, ofs, src_elem);
8851
8852 const char* klass_digestBase_name = nullptr;
8853 const char* stub_name = nullptr;
8854 address stub_addr = nullptr;
8855 BasicType elem_type = T_INT;
8856
8857 switch (predicate) {
8858 case 0:
8859 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8860 klass_digestBase_name = "sun/security/provider/MD5";
8861 stub_name = "md5_implCompressMB";
8862 stub_addr = StubRoutines::md5_implCompressMB();
8863 }
8864 break;
8865 case 1:
8866 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8867 klass_digestBase_name = "sun/security/provider/SHA";
8868 stub_name = "sha1_implCompressMB";
8869 stub_addr = StubRoutines::sha1_implCompressMB();
8870 }
8871 break;
8872 case 2:
8873 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8874 klass_digestBase_name = "sun/security/provider/SHA2";
8875 stub_name = "sha256_implCompressMB";
8876 stub_addr = StubRoutines::sha256_implCompressMB();
8877 }
8878 break;
8879 case 3:
8880 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8881 klass_digestBase_name = "sun/security/provider/SHA5";
8882 stub_name = "sha512_implCompressMB";
8883 stub_addr = StubRoutines::sha512_implCompressMB();
8884 elem_type = T_LONG;
8885 }
8886 break;
8887 case 4:
8888 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8889 klass_digestBase_name = "sun/security/provider/SHA3";
8890 stub_name = "sha3_implCompressMB";
8891 stub_addr = StubRoutines::sha3_implCompressMB();
8892 elem_type = T_LONG;
8893 }
8894 break;
8895 default:
8896 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8897 }
8898 if (klass_digestBase_name != nullptr) {
8899 assert(stub_addr != nullptr, "Stub is generated");
8900 if (stub_addr == nullptr) return false;
8901
8902 // get DigestBase klass to lookup for SHA klass
8903 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8904 assert(tinst != nullptr, "digestBase_obj is not instance???");
8905 assert(tinst->is_loaded(), "DigestBase is not loaded");
8906
8907 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8908 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8909 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8910 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8911 }
8912 return false;
8913 }
8914
8915 //------------------------------inline_digestBase_implCompressMB-----------------------
8916 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8917 BasicType elem_type, address stubAddr, const char *stubName,
8918 Node* src_start, Node* ofs, Node* limit) {
8919 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8920 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8921 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8922 digest_obj = _gvn.transform(digest_obj);
8923
8924 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8925 if (state == nullptr) return false;
8926
8927 Node* block_size = nullptr;
8928 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8929 block_size = get_block_size_from_digest_object(digest_obj);
8930 if (block_size == nullptr) return false;
8931 }
8932
8933 // Call the stub.
8934 Node* call;
8935 if (block_size == nullptr) {
8936 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8937 OptoRuntime::digestBase_implCompressMB_Type(false),
8938 stubAddr, stubName, TypePtr::BOTTOM,
8939 src_start, state, ofs, limit);
8940 } else {
8941 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8942 OptoRuntime::digestBase_implCompressMB_Type(true),
8943 stubAddr, stubName, TypePtr::BOTTOM,
8944 src_start, state, block_size, ofs, limit);
8945 }
8946
8947 // return ofs (int)
8948 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8949 set_result(result);
8950
8951 return true;
8952 }
8953
8954 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8955 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8956 assert(UseAES, "need AES instruction support");
8957 address stubAddr = nullptr;
8958 const char *stubName = nullptr;
8959 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8960 stubName = "galoisCounterMode_AESCrypt";
8961
8962 if (stubAddr == nullptr) return false;
8963
8964 Node* in = argument(0);
8965 Node* inOfs = argument(1);
8966 Node* len = argument(2);
8967 Node* ct = argument(3);
8968 Node* ctOfs = argument(4);
8969 Node* out = argument(5);
8970 Node* outOfs = argument(6);
8971 Node* gctr_object = argument(7);
8972 Node* ghash_object = argument(8);
8973
8974 // (1) in, ct and out are arrays.
8975 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8976 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8977 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8978 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8979 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8980 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8981
8982 // checks are the responsibility of the caller
8983 Node* in_start = in;
8984 Node* ct_start = ct;
8985 Node* out_start = out;
8986 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8987 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8988 in_start = array_element_address(in, inOfs, T_BYTE);
8989 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8990 out_start = array_element_address(out, outOfs, T_BYTE);
8991 }
8992
8993 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8994 // (because of the predicated logic executed earlier).
8995 // so we cast it here safely.
8996 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8997 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8998 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8999 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9000 Node* state = load_field_from_object(ghash_object, "state", "[J");
9001
9002 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9003 return false;
9004 }
9005 // cast it to what we know it will be at runtime
9006 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9007 assert(tinst != nullptr, "GCTR obj is null");
9008 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9009 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9010 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9011 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9012 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9013 const TypeOopPtr* xtype = aklass->as_instance_type();
9014 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9015 aescrypt_object = _gvn.transform(aescrypt_object);
9016 // we need to get the start of the aescrypt_object's expanded key array
9017 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
9018 if (k_start == nullptr) return false;
9019 // similarly, get the start address of the r vector
9020 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9021 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9022 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9023
9024
9025 // Call the stub, passing params
9026 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9027 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9028 stubAddr, stubName, TypePtr::BOTTOM,
9029 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9030
9031 // return cipher length (int)
9032 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9033 set_result(retvalue);
9034
9035 return true;
9036 }
9037
9038 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9039 // Return node representing slow path of predicate check.
9040 // the pseudo code we want to emulate with this predicate is:
9041 // for encryption:
9042 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9043 // for decryption:
9044 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9045 // note cipher==plain is more conservative than the original java code but that's OK
9046 //
9047
9048 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9049 // The receiver was checked for null already.
9050 Node* objGCTR = argument(7);
9051 // Load embeddedCipher field of GCTR object.
9052 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9053 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9054
9055 // get AESCrypt klass for instanceOf check
9056 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9057 // will have same classloader as CipherBlockChaining object
9058 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9059 assert(tinst != nullptr, "GCTR obj is null");
9060 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9061
9062 // we want to do an instanceof comparison against the AESCrypt class
9063 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9064 if (!klass_AESCrypt->is_loaded()) {
9065 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9066 Node* ctrl = control();
9067 set_control(top()); // no regular fast path
9068 return ctrl;
9069 }
9070
9071 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9072 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9073 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9074 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9075 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9076
9077 return instof_false; // even if it is null
9078 }
9079
9080 //------------------------------get_state_from_digest_object-----------------------
9081 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9082 const char* state_type;
9083 switch (elem_type) {
9084 case T_BYTE: state_type = "[B"; break;
9085 case T_INT: state_type = "[I"; break;
9086 case T_LONG: state_type = "[J"; break;
9087 default: ShouldNotReachHere();
9088 }
9089 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9090 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9091 if (digest_state == nullptr) return (Node *) nullptr;
9092
9093 // now have the array, need to get the start address of the state array
9094 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9095 return state;
9096 }
9097
9098 //------------------------------get_block_size_from_sha3_object----------------------------------
9099 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9100 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9101 assert (block_size != nullptr, "sanity");
9102 return block_size;
9103 }
9104
9105 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9106 // Return node representing slow path of predicate check.
9107 // the pseudo code we want to emulate with this predicate is:
9108 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9109 //
9110 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9111 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9112 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9113 assert((uint)predicate < 5, "sanity");
9114
9115 // The receiver was checked for null already.
9116 Node* digestBaseObj = argument(0);
9117
9118 // get DigestBase klass for instanceOf check
9119 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9120 assert(tinst != nullptr, "digestBaseObj is null");
9121 assert(tinst->is_loaded(), "DigestBase is not loaded");
9122
9123 const char* klass_name = nullptr;
9124 switch (predicate) {
9125 case 0:
9126 if (UseMD5Intrinsics) {
9127 // we want to do an instanceof comparison against the MD5 class
9128 klass_name = "sun/security/provider/MD5";
9129 }
9130 break;
9131 case 1:
9132 if (UseSHA1Intrinsics) {
9133 // we want to do an instanceof comparison against the SHA class
9134 klass_name = "sun/security/provider/SHA";
9135 }
9136 break;
9137 case 2:
9138 if (UseSHA256Intrinsics) {
9139 // we want to do an instanceof comparison against the SHA2 class
9140 klass_name = "sun/security/provider/SHA2";
9141 }
9142 break;
9143 case 3:
9144 if (UseSHA512Intrinsics) {
9145 // we want to do an instanceof comparison against the SHA5 class
9146 klass_name = "sun/security/provider/SHA5";
9147 }
9148 break;
9149 case 4:
9150 if (UseSHA3Intrinsics) {
9151 // we want to do an instanceof comparison against the SHA3 class
9152 klass_name = "sun/security/provider/SHA3";
9153 }
9154 break;
9155 default:
9156 fatal("unknown SHA intrinsic predicate: %d", predicate);
9157 }
9158
9159 ciKlass* klass = nullptr;
9160 if (klass_name != nullptr) {
9161 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9162 }
9163 if ((klass == nullptr) || !klass->is_loaded()) {
9164 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9165 Node* ctrl = control();
9166 set_control(top()); // no intrinsic path
9167 return ctrl;
9168 }
9169 ciInstanceKlass* instklass = klass->as_instance_klass();
9170
9171 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9172 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9173 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9174 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9175
9176 return instof_false; // even if it is null
9177 }
9178
9179 //-------------inline_fma-----------------------------------
9180 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9181 Node *a = nullptr;
9182 Node *b = nullptr;
9183 Node *c = nullptr;
9184 Node* result = nullptr;
9185 switch (id) {
9186 case vmIntrinsics::_fmaD:
9187 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9188 // no receiver since it is static method
9189 a = argument(0);
9190 b = argument(2);
9191 c = argument(4);
9192 result = _gvn.transform(new FmaDNode(a, b, c));
9193 break;
9194 case vmIntrinsics::_fmaF:
9195 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9196 a = argument(0);
9197 b = argument(1);
9198 c = argument(2);
9199 result = _gvn.transform(new FmaFNode(a, b, c));
9200 break;
9201 default:
9202 fatal_unexpected_iid(id); break;
9203 }
9204 set_result(result);
9205 return true;
9206 }
9207
9208 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9209 // argument(0) is receiver
9210 Node* codePoint = argument(1);
9211 Node* n = nullptr;
9212
9213 switch (id) {
9214 case vmIntrinsics::_isDigit :
9215 n = new DigitNode(control(), codePoint);
9216 break;
9217 case vmIntrinsics::_isLowerCase :
9218 n = new LowerCaseNode(control(), codePoint);
9219 break;
9220 case vmIntrinsics::_isUpperCase :
9221 n = new UpperCaseNode(control(), codePoint);
9222 break;
9223 case vmIntrinsics::_isWhitespace :
9224 n = new WhitespaceNode(control(), codePoint);
9225 break;
9226 default:
9227 fatal_unexpected_iid(id);
9228 }
9229
9230 set_result(_gvn.transform(n));
9231 return true;
9232 }
9233
9234 bool LibraryCallKit::inline_profileBoolean() {
9235 Node* counts = argument(1);
9236 const TypeAryPtr* ary = nullptr;
9237 ciArray* aobj = nullptr;
9238 if (counts->is_Con()
9239 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9240 && (aobj = ary->const_oop()->as_array()) != nullptr
9241 && (aobj->length() == 2)) {
9242 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9243 jint false_cnt = aobj->element_value(0).as_int();
9244 jint true_cnt = aobj->element_value(1).as_int();
9245
9246 if (C->log() != nullptr) {
9247 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9248 false_cnt, true_cnt);
9249 }
9250
9251 if (false_cnt + true_cnt == 0) {
9252 // According to profile, never executed.
9253 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9254 Deoptimization::Action_reinterpret);
9255 return true;
9256 }
9257
9258 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9259 // is a number of each value occurrences.
9260 Node* result = argument(0);
9261 if (false_cnt == 0 || true_cnt == 0) {
9262 // According to profile, one value has been never seen.
9263 int expected_val = (false_cnt == 0) ? 1 : 0;
9264
9265 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9266 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9267
9268 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9269 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9270 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9271
9272 { // Slow path: uncommon trap for never seen value and then reexecute
9273 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9274 // the value has been seen at least once.
9275 PreserveJVMState pjvms(this);
9276 PreserveReexecuteState preexecs(this);
9277 jvms()->set_should_reexecute(true);
9278
9279 set_control(slow_path);
9280 set_i_o(i_o());
9281
9282 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9283 Deoptimization::Action_reinterpret);
9284 }
9285 // The guard for never seen value enables sharpening of the result and
9286 // returning a constant. It allows to eliminate branches on the same value
9287 // later on.
9288 set_control(fast_path);
9289 result = intcon(expected_val);
9290 }
9291 // Stop profiling.
9292 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9293 // By replacing method body with profile data (represented as ProfileBooleanNode
9294 // on IR level) we effectively disable profiling.
9295 // It enables full speed execution once optimized code is generated.
9296 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9297 C->record_for_igvn(profile);
9298 set_result(profile);
9299 return true;
9300 } else {
9301 // Continue profiling.
9302 // Profile data isn't available at the moment. So, execute method's bytecode version.
9303 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9304 // is compiled and counters aren't available since corresponding MethodHandle
9305 // isn't a compile-time constant.
9306 return false;
9307 }
9308 }
9309
9310 bool LibraryCallKit::inline_isCompileConstant() {
9311 Node* n = argument(0);
9312 set_result(n->is_Con() ? intcon(1) : intcon(0));
9313 return true;
9314 }
9315
9316 //------------------------------- inline_getObjectSize --------------------------------------
9317 //
9318 // Calculate the runtime size of the object/array.
9319 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9320 //
9321 bool LibraryCallKit::inline_getObjectSize() {
9322 Node* obj = argument(3);
9323 Node* klass_node = load_object_klass(obj);
9324
9325 jint layout_con = Klass::_lh_neutral_value;
9326 Node* layout_val = get_layout_helper(klass_node, layout_con);
9327 int layout_is_con = (layout_val == nullptr);
9328
9329 if (layout_is_con) {
9330 // Layout helper is constant, can figure out things at compile time.
9331
9332 if (Klass::layout_helper_is_instance(layout_con)) {
9333 // Instance case: layout_con contains the size itself.
9334 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9335 set_result(size);
9336 } else {
9337 // Array case: size is round(header + element_size*arraylength).
9338 // Since arraylength is different for every array instance, we have to
9339 // compute the whole thing at runtime.
9340
9341 Node* arr_length = load_array_length(obj);
9342
9343 int round_mask = MinObjAlignmentInBytes - 1;
9344 int hsize = Klass::layout_helper_header_size(layout_con);
9345 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9346
9347 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9348 round_mask = 0; // strength-reduce it if it goes away completely
9349 }
9350 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9351 Node* header_size = intcon(hsize + round_mask);
9352
9353 Node* lengthx = ConvI2X(arr_length);
9354 Node* headerx = ConvI2X(header_size);
9355
9356 Node* abody = lengthx;
9357 if (eshift != 0) {
9358 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9359 }
9360 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9361 if (round_mask != 0) {
9362 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9363 }
9364 size = ConvX2L(size);
9365 set_result(size);
9366 }
9367 } else {
9368 // Layout helper is not constant, need to test for array-ness at runtime.
9369
9370 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9371 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9372 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9373 record_for_igvn(result_reg);
9374
9375 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9376 if (array_ctl != nullptr) {
9377 // Array case: size is round(header + element_size*arraylength).
9378 // Since arraylength is different for every array instance, we have to
9379 // compute the whole thing at runtime.
9380
9381 PreserveJVMState pjvms(this);
9382 set_control(array_ctl);
9383 Node* arr_length = load_array_length(obj);
9384
9385 int round_mask = MinObjAlignmentInBytes - 1;
9386 Node* mask = intcon(round_mask);
9387
9388 Node* hss = intcon(Klass::_lh_header_size_shift);
9389 Node* hsm = intcon(Klass::_lh_header_size_mask);
9390 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9391 header_size = _gvn.transform(new AndINode(header_size, hsm));
9392 header_size = _gvn.transform(new AddINode(header_size, mask));
9393
9394 // There is no need to mask or shift this value.
9395 // The semantics of LShiftINode include an implicit mask to 0x1F.
9396 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9397 Node* elem_shift = layout_val;
9398
9399 Node* lengthx = ConvI2X(arr_length);
9400 Node* headerx = ConvI2X(header_size);
9401
9402 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9403 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9404 if (round_mask != 0) {
9405 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9406 }
9407 size = ConvX2L(size);
9408
9409 result_reg->init_req(_array_path, control());
9410 result_val->init_req(_array_path, size);
9411 }
9412
9413 if (!stopped()) {
9414 // Instance case: the layout helper gives us instance size almost directly,
9415 // but we need to mask out the _lh_instance_slow_path_bit.
9416 Node* size = ConvI2X(layout_val);
9417 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9418 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9419 size = _gvn.transform(new AndXNode(size, mask));
9420 size = ConvX2L(size);
9421
9422 result_reg->init_req(_instance_path, control());
9423 result_val->init_req(_instance_path, size);
9424 }
9425
9426 set_result(result_reg, result_val);
9427 }
9428
9429 return true;
9430 }
9431
9432 //------------------------------- inline_blackhole --------------------------------------
9433 //
9434 // Make sure all arguments to this node are alive.
9435 // This matches methods that were requested to be blackholed through compile commands.
9436 //
9437 bool LibraryCallKit::inline_blackhole() {
9438 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9439 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9440 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9441
9442 // Blackhole node pinches only the control, not memory. This allows
9443 // the blackhole to be pinned in the loop that computes blackholed
9444 // values, but have no other side effects, like breaking the optimizations
9445 // across the blackhole.
9446
9447 Node* bh = _gvn.transform(new BlackholeNode(control()));
9448 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9449
9450 // Bind call arguments as blackhole arguments to keep them alive
9451 uint nargs = callee()->arg_size();
9452 for (uint i = 0; i < nargs; i++) {
9453 bh->add_req(argument(i));
9454 }
9455
9456 return true;
9457 }
9458
9459 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9460 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9461 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9462 return nullptr; // box klass is not Float16
9463 }
9464
9465 // Null check; get notnull casted pointer
9466 Node* null_ctl = top();
9467 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9468 // If not_null_box is dead, only null-path is taken
9469 if (stopped()) {
9470 set_control(null_ctl);
9471 return nullptr;
9472 }
9473 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9474 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9475 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9476 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9477 }
9478
9479 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9480 PreserveReexecuteState preexecs(this);
9481 jvms()->set_should_reexecute(true);
9482
9483 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9484 Node* klass_node = makecon(klass_type);
9485 Node* box = new_instance(klass_node);
9486
9487 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9488 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9489
9490 Node* field_store = _gvn.transform(access_store_at(box,
9491 value_field,
9492 value_adr_type,
9493 value,
9494 TypeInt::SHORT,
9495 T_SHORT,
9496 IN_HEAP));
9497 set_memory(field_store, value_adr_type);
9498 return box;
9499 }
9500
9501 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9502 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9503 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9504 return false;
9505 }
9506
9507 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9508 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9509 return false;
9510 }
9511
9512 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9513 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9514 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9515 ciSymbols::short_signature(),
9516 false);
9517 assert(field != nullptr, "");
9518
9519 // Transformed nodes
9520 Node* fld1 = nullptr;
9521 Node* fld2 = nullptr;
9522 Node* fld3 = nullptr;
9523 switch(num_args) {
9524 case 3:
9525 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9526 if (fld3 == nullptr) {
9527 return false;
9528 }
9529 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9530 // fall-through
9531 case 2:
9532 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9533 if (fld2 == nullptr) {
9534 return false;
9535 }
9536 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9537 // fall-through
9538 case 1:
9539 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9540 if (fld1 == nullptr) {
9541 return false;
9542 }
9543 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9544 break;
9545 default: fatal("Unsupported number of arguments %d", num_args);
9546 }
9547
9548 Node* result = nullptr;
9549 switch (id) {
9550 // Unary operations
9551 case vmIntrinsics::_sqrt_float16:
9552 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9553 break;
9554 // Ternary operations
9555 case vmIntrinsics::_fma_float16:
9556 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9557 break;
9558 default:
9559 fatal_unexpected_iid(id);
9560 break;
9561 }
9562 result = _gvn.transform(new ReinterpretHF2SNode(result));
9563 set_result(box_fp16_value(float16_box_type, field, result));
9564 return true;
9565 }
9566