1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompilerThread::current()->compiler();
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
124 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
125 if (C->log()) {
126 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
127 vmIntrinsics::name_at(intrinsic_id()),
128 (is_virtual() ? " virtual='1'" : ""),
129 C->unique() - nodes);
130 }
131 // Push the result from the inlined method onto the stack.
132 kit.push_result();
133 return kit.transfer_exceptions_into_jvms();
134 }
135
136 // The intrinsic bailed out
137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
138 assert(jvms->map() == kit.map(), "Out of sync JVM state");
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::_dsinh:
249 case vmIntrinsics::_dtanh:
250 case vmIntrinsics::_dcbrt:
251 case vmIntrinsics::_dabs:
252 case vmIntrinsics::_fabs:
253 case vmIntrinsics::_iabs:
254 case vmIntrinsics::_labs:
255 case vmIntrinsics::_datan2:
256 case vmIntrinsics::_dsqrt:
257 case vmIntrinsics::_dsqrt_strict:
258 case vmIntrinsics::_dexp:
259 case vmIntrinsics::_dlog:
260 case vmIntrinsics::_dlog10:
261 case vmIntrinsics::_dpow:
262 case vmIntrinsics::_dcopySign:
263 case vmIntrinsics::_fcopySign:
264 case vmIntrinsics::_dsignum:
265 case vmIntrinsics::_roundF:
266 case vmIntrinsics::_roundD:
267 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
268
269 case vmIntrinsics::_notify:
270 case vmIntrinsics::_notifyAll:
271 return inline_notify(intrinsic_id());
272
273 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
274 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
275 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
276 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
277 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
278 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
279 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
280 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
281 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
282 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
283 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
284 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
285 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
286 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
287
288 case vmIntrinsics::_arraycopy: return inline_arraycopy();
289
290 case vmIntrinsics::_arraySort: return inline_array_sort();
291 case vmIntrinsics::_arrayPartition: return inline_array_partition();
292
293 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
294 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
295 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
296 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
297
298 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
305 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
306
307 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
308
309 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
310
311 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
312 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
313 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
314 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
315
316 case vmIntrinsics::_compressStringC:
317 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
318 case vmIntrinsics::_inflateStringC:
319 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
320
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
331 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
332 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
333 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
334 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
335 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
336 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
337 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
338 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
339 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
340
341 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
342 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
343 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
344 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
345 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
346 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
347 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
348 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
349 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
350
351 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
352 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
353 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
354 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
355 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
356 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
357 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
358 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
359 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
360
361 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
362 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
363 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
364 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
365
366 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
367 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
368 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
369 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
370
371 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
372 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
373 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
374 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
375 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
376 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
377 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
378 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
379 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
380
381 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
382 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
383 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
384 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
385 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
386 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
387 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
388 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
389 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
390
391 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
392 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
393 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
394 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
395 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
396 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
397 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
398 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
399 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
400
401 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
402 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
403 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
404 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
405 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
406 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
407 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
408 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
409 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
410
411 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
416
417 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
418 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
419 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
420 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
421 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
422 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
423 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
424 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
425 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
426 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
427 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
428 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
429 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
437
438 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
439 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
440 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
441 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
442 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
443 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
444 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
453
454 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
455 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
458
459 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
464
465 case vmIntrinsics::_loadFence:
466 case vmIntrinsics::_storeFence:
467 case vmIntrinsics::_storeStoreFence:
468 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
469
470 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
471
472 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
473 case vmIntrinsics::_currentThread: return inline_native_currentThread();
474 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
475
476 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
477 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
478
479 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
480 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
481
482 #if INCLUDE_JVMTI
483 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
484 "notifyJvmtiStart", true, false);
485 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
486 "notifyJvmtiEnd", false, true);
487 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
488 "notifyJvmtiMount", false, false);
489 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
490 "notifyJvmtiUnmount", false, false);
491 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
492 #endif
493
494 #ifdef JFR_HAVE_INTRINSICS
495 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
496 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
497 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
498 #endif
499 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
500 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
501 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
502 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
503 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
504 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
505 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
506 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
507 case vmIntrinsics::_getLength: return inline_native_getLength();
508 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
509 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
510 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
511 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
512 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
513 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
514 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
515
516 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
517 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
518
519 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
520
521 case vmIntrinsics::_isInstance:
522 case vmIntrinsics::_isHidden:
523 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
524
525 case vmIntrinsics::_floatToRawIntBits:
526 case vmIntrinsics::_floatToIntBits:
527 case vmIntrinsics::_intBitsToFloat:
528 case vmIntrinsics::_doubleToRawLongBits:
529 case vmIntrinsics::_doubleToLongBits:
530 case vmIntrinsics::_longBitsToDouble:
531 case vmIntrinsics::_floatToFloat16:
532 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
533 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
534 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
535 case vmIntrinsics::_floatIsFinite:
536 case vmIntrinsics::_floatIsInfinite:
537 case vmIntrinsics::_doubleIsFinite:
538 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
539
540 case vmIntrinsics::_numberOfLeadingZeros_i:
541 case vmIntrinsics::_numberOfLeadingZeros_l:
542 case vmIntrinsics::_numberOfTrailingZeros_i:
543 case vmIntrinsics::_numberOfTrailingZeros_l:
544 case vmIntrinsics::_bitCount_i:
545 case vmIntrinsics::_bitCount_l:
546 case vmIntrinsics::_reverse_i:
547 case vmIntrinsics::_reverse_l:
548 case vmIntrinsics::_reverseBytes_i:
549 case vmIntrinsics::_reverseBytes_l:
550 case vmIntrinsics::_reverseBytes_s:
551 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
552
553 case vmIntrinsics::_compress_i:
554 case vmIntrinsics::_compress_l:
555 case vmIntrinsics::_expand_i:
556 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
557
558 case vmIntrinsics::_compareUnsigned_i:
559 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
560
561 case vmIntrinsics::_divideUnsigned_i:
562 case vmIntrinsics::_divideUnsigned_l:
563 case vmIntrinsics::_remainderUnsigned_i:
564 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
565
566 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
567
568 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
569 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
570 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
571 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
572 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
573
574 case vmIntrinsics::_Class_cast: return inline_Class_cast();
575
576 case vmIntrinsics::_aescrypt_encryptBlock:
577 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
578
579 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
580 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
581 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
582
583 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
584 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
585 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
586
587 case vmIntrinsics::_counterMode_AESCrypt:
588 return inline_counterMode_AESCrypt(intrinsic_id());
589
590 case vmIntrinsics::_galoisCounterMode_AESCrypt:
591 return inline_galoisCounterMode_AESCrypt();
592
593 case vmIntrinsics::_md5_implCompress:
594 case vmIntrinsics::_sha_implCompress:
595 case vmIntrinsics::_sha2_implCompress:
596 case vmIntrinsics::_sha5_implCompress:
597 case vmIntrinsics::_sha3_implCompress:
598 return inline_digestBase_implCompress(intrinsic_id());
599 case vmIntrinsics::_double_keccak:
600 return inline_double_keccak();
601
602 case vmIntrinsics::_digestBase_implCompressMB:
603 return inline_digestBase_implCompressMB(predicate);
604
605 case vmIntrinsics::_multiplyToLen:
606 return inline_multiplyToLen();
607
608 case vmIntrinsics::_squareToLen:
609 return inline_squareToLen();
610
611 case vmIntrinsics::_mulAdd:
612 return inline_mulAdd();
613
614 case vmIntrinsics::_montgomeryMultiply:
615 return inline_montgomeryMultiply();
616 case vmIntrinsics::_montgomerySquare:
617 return inline_montgomerySquare();
618
619 case vmIntrinsics::_bigIntegerRightShiftWorker:
620 return inline_bigIntegerShift(true);
621 case vmIntrinsics::_bigIntegerLeftShiftWorker:
622 return inline_bigIntegerShift(false);
623
624 case vmIntrinsics::_vectorizedMismatch:
625 return inline_vectorizedMismatch();
626
627 case vmIntrinsics::_ghash_processBlocks:
628 return inline_ghash_processBlocks();
629 case vmIntrinsics::_chacha20Block:
630 return inline_chacha20Block();
631 case vmIntrinsics::_kyberNtt:
632 return inline_kyberNtt();
633 case vmIntrinsics::_kyberInverseNtt:
634 return inline_kyberInverseNtt();
635 case vmIntrinsics::_kyberNttMult:
636 return inline_kyberNttMult();
637 case vmIntrinsics::_kyberAddPoly_2:
638 return inline_kyberAddPoly_2();
639 case vmIntrinsics::_kyberAddPoly_3:
640 return inline_kyberAddPoly_3();
641 case vmIntrinsics::_kyber12To16:
642 return inline_kyber12To16();
643 case vmIntrinsics::_kyberBarrettReduce:
644 return inline_kyberBarrettReduce();
645 case vmIntrinsics::_dilithiumAlmostNtt:
646 return inline_dilithiumAlmostNtt();
647 case vmIntrinsics::_dilithiumAlmostInverseNtt:
648 return inline_dilithiumAlmostInverseNtt();
649 case vmIntrinsics::_dilithiumNttMult:
650 return inline_dilithiumNttMult();
651 case vmIntrinsics::_dilithiumMontMulByConstant:
652 return inline_dilithiumMontMulByConstant();
653 case vmIntrinsics::_dilithiumDecomposePoly:
654 return inline_dilithiumDecomposePoly();
655 case vmIntrinsics::_base64_encodeBlock:
656 return inline_base64_encodeBlock();
657 case vmIntrinsics::_base64_decodeBlock:
658 return inline_base64_decodeBlock();
659 case vmIntrinsics::_poly1305_processBlocks:
660 return inline_poly1305_processBlocks();
661 case vmIntrinsics::_intpoly_montgomeryMult_P256:
662 return inline_intpoly_montgomeryMult_P256();
663 case vmIntrinsics::_intpoly_assign:
664 return inline_intpoly_assign();
665 case vmIntrinsics::_encodeISOArray:
666 case vmIntrinsics::_encodeByteISOArray:
667 return inline_encodeISOArray(false);
668 case vmIntrinsics::_encodeAsciiArray:
669 return inline_encodeISOArray(true);
670
671 case vmIntrinsics::_updateCRC32:
672 return inline_updateCRC32();
673 case vmIntrinsics::_updateBytesCRC32:
674 return inline_updateBytesCRC32();
675 case vmIntrinsics::_updateByteBufferCRC32:
676 return inline_updateByteBufferCRC32();
677
678 case vmIntrinsics::_updateBytesCRC32C:
679 return inline_updateBytesCRC32C();
680 case vmIntrinsics::_updateDirectByteBufferCRC32C:
681 return inline_updateDirectByteBufferCRC32C();
682
683 case vmIntrinsics::_updateBytesAdler32:
684 return inline_updateBytesAdler32();
685 case vmIntrinsics::_updateByteBufferAdler32:
686 return inline_updateByteBufferAdler32();
687
688 case vmIntrinsics::_profileBoolean:
689 return inline_profileBoolean();
690 case vmIntrinsics::_isCompileConstant:
691 return inline_isCompileConstant();
692
693 case vmIntrinsics::_countPositives:
694 return inline_countPositives();
695
696 case vmIntrinsics::_fmaD:
697 case vmIntrinsics::_fmaF:
698 return inline_fma(intrinsic_id());
699
700 case vmIntrinsics::_isDigit:
701 case vmIntrinsics::_isLowerCase:
702 case vmIntrinsics::_isUpperCase:
703 case vmIntrinsics::_isWhitespace:
704 return inline_character_compare(intrinsic_id());
705
706 case vmIntrinsics::_min:
707 case vmIntrinsics::_max:
708 case vmIntrinsics::_min_strict:
709 case vmIntrinsics::_max_strict:
710 case vmIntrinsics::_minL:
711 case vmIntrinsics::_maxL:
712 case vmIntrinsics::_minF:
713 case vmIntrinsics::_maxF:
714 case vmIntrinsics::_minD:
715 case vmIntrinsics::_maxD:
716 case vmIntrinsics::_minF_strict:
717 case vmIntrinsics::_maxF_strict:
718 case vmIntrinsics::_minD_strict:
719 case vmIntrinsics::_maxD_strict:
720 return inline_min_max(intrinsic_id());
721
722 case vmIntrinsics::_VectorUnaryOp:
723 return inline_vector_nary_operation(1);
724 case vmIntrinsics::_VectorBinaryOp:
725 return inline_vector_nary_operation(2);
726 case vmIntrinsics::_VectorUnaryLibOp:
727 return inline_vector_call(1);
728 case vmIntrinsics::_VectorBinaryLibOp:
729 return inline_vector_call(2);
730 case vmIntrinsics::_VectorTernaryOp:
731 return inline_vector_nary_operation(3);
732 case vmIntrinsics::_VectorFromBitsCoerced:
733 return inline_vector_frombits_coerced();
734 case vmIntrinsics::_VectorMaskOp:
735 return inline_vector_mask_operation();
736 case vmIntrinsics::_VectorLoadOp:
737 return inline_vector_mem_operation(/*is_store=*/false);
738 case vmIntrinsics::_VectorLoadMaskedOp:
739 return inline_vector_mem_masked_operation(/*is_store*/false);
740 case vmIntrinsics::_VectorStoreOp:
741 return inline_vector_mem_operation(/*is_store=*/true);
742 case vmIntrinsics::_VectorStoreMaskedOp:
743 return inline_vector_mem_masked_operation(/*is_store=*/true);
744 case vmIntrinsics::_VectorGatherOp:
745 return inline_vector_gather_scatter(/*is_scatter*/ false);
746 case vmIntrinsics::_VectorScatterOp:
747 return inline_vector_gather_scatter(/*is_scatter*/ true);
748 case vmIntrinsics::_VectorReductionCoerced:
749 return inline_vector_reduction();
750 case vmIntrinsics::_VectorTest:
751 return inline_vector_test();
752 case vmIntrinsics::_VectorBlend:
753 return inline_vector_blend();
754 case vmIntrinsics::_VectorRearrange:
755 return inline_vector_rearrange();
756 case vmIntrinsics::_VectorSelectFrom:
757 return inline_vector_select_from();
758 case vmIntrinsics::_VectorCompare:
759 return inline_vector_compare();
760 case vmIntrinsics::_VectorBroadcastInt:
761 return inline_vector_broadcast_int();
762 case vmIntrinsics::_VectorConvert:
763 return inline_vector_convert();
764 case vmIntrinsics::_VectorInsert:
765 return inline_vector_insert();
766 case vmIntrinsics::_VectorExtract:
767 return inline_vector_extract();
768 case vmIntrinsics::_VectorCompressExpand:
769 return inline_vector_compress_expand();
770 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
771 return inline_vector_select_from_two_vectors();
772 case vmIntrinsics::_IndexVector:
773 return inline_index_vector();
774 case vmIntrinsics::_IndexPartiallyInUpperRange:
775 return inline_index_partially_in_upper_range();
776
777 case vmIntrinsics::_getObjectSize:
778 return inline_getObjectSize();
779
780 case vmIntrinsics::_blackhole:
781 return inline_blackhole();
782
783 default:
784 // If you get here, it may be that someone has added a new intrinsic
785 // to the list in vmIntrinsics.hpp without implementing it here.
786 #ifndef PRODUCT
787 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
788 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
789 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
790 }
791 #endif
792 return false;
793 }
794 }
795
796 Node* LibraryCallKit::try_to_predicate(int predicate) {
797 if (!jvms()->has_method()) {
798 // Root JVMState has a null method.
799 assert(map()->memory()->Opcode() == Op_Parm, "");
800 // Insert the memory aliasing node
801 set_all_memory(reset_memory());
802 }
803 assert(merged_memory(), "");
804
805 switch (intrinsic_id()) {
806 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
807 return inline_cipherBlockChaining_AESCrypt_predicate(false);
808 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
809 return inline_cipherBlockChaining_AESCrypt_predicate(true);
810 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
811 return inline_electronicCodeBook_AESCrypt_predicate(false);
812 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
813 return inline_electronicCodeBook_AESCrypt_predicate(true);
814 case vmIntrinsics::_counterMode_AESCrypt:
815 return inline_counterMode_AESCrypt_predicate();
816 case vmIntrinsics::_digestBase_implCompressMB:
817 return inline_digestBase_implCompressMB_predicate(predicate);
818 case vmIntrinsics::_galoisCounterMode_AESCrypt:
819 return inline_galoisCounterMode_AESCrypt_predicate();
820
821 default:
822 // If you get here, it may be that someone has added a new intrinsic
823 // to the list in vmIntrinsics.hpp without implementing it here.
824 #ifndef PRODUCT
825 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
826 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
827 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
828 }
829 #endif
830 Node* slow_ctl = control();
831 set_control(top()); // No fast path intrinsic
832 return slow_ctl;
833 }
834 }
835
836 //------------------------------set_result-------------------------------
837 // Helper function for finishing intrinsics.
838 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
839 record_for_igvn(region);
840 set_control(_gvn.transform(region));
841 set_result( _gvn.transform(value));
842 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
843 }
844
845 //------------------------------generate_guard---------------------------
846 // Helper function for generating guarded fast-slow graph structures.
847 // The given 'test', if true, guards a slow path. If the test fails
848 // then a fast path can be taken. (We generally hope it fails.)
849 // In all cases, GraphKit::control() is updated to the fast path.
850 // The returned value represents the control for the slow path.
851 // The return value is never 'top'; it is either a valid control
852 // or null if it is obvious that the slow path can never be taken.
853 // Also, if region and the slow control are not null, the slow edge
854 // is appended to the region.
855 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
856 if (stopped()) {
857 // Already short circuited.
858 return nullptr;
859 }
860
861 // Build an if node and its projections.
862 // If test is true we take the slow path, which we assume is uncommon.
863 if (_gvn.type(test) == TypeInt::ZERO) {
864 // The slow branch is never taken. No need to build this guard.
865 return nullptr;
866 }
867
868 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
869
870 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
871 if (if_slow == top()) {
872 // The slow branch is never taken. No need to build this guard.
873 return nullptr;
874 }
875
876 if (region != nullptr)
877 region->add_req(if_slow);
878
879 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
880 set_control(if_fast);
881
882 return if_slow;
883 }
884
885 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
886 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
887 }
888 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
889 return generate_guard(test, region, PROB_FAIR);
890 }
891
892 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
893 Node* *pos_index) {
894 if (stopped())
895 return nullptr; // already stopped
896 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
897 return nullptr; // index is already adequately typed
898 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
899 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
900 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
901 if (is_neg != nullptr && pos_index != nullptr) {
902 // Emulate effect of Parse::adjust_map_after_if.
903 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
904 (*pos_index) = _gvn.transform(ccast);
905 }
906 return is_neg;
907 }
908
909 // Make sure that 'position' is a valid limit index, in [0..length].
910 // There are two equivalent plans for checking this:
911 // A. (offset + copyLength) unsigned<= arrayLength
912 // B. offset <= (arrayLength - copyLength)
913 // We require that all of the values above, except for the sum and
914 // difference, are already known to be non-negative.
915 // Plan A is robust in the face of overflow, if offset and copyLength
916 // are both hugely positive.
917 //
918 // Plan B is less direct and intuitive, but it does not overflow at
919 // all, since the difference of two non-negatives is always
920 // representable. Whenever Java methods must perform the equivalent
921 // check they generally use Plan B instead of Plan A.
922 // For the moment we use Plan A.
923 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
924 Node* subseq_length,
925 Node* array_length,
926 RegionNode* region) {
927 if (stopped())
928 return nullptr; // already stopped
929 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
930 if (zero_offset && subseq_length->eqv_uncast(array_length))
931 return nullptr; // common case of whole-array copy
932 Node* last = subseq_length;
933 if (!zero_offset) // last += offset
934 last = _gvn.transform(new AddINode(last, offset));
935 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
936 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
937 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
938 return is_over;
939 }
940
941 // Emit range checks for the given String.value byte array
942 void LibraryCallKit::generate_string_range_check(Node* array,
943 Node* offset,
944 Node* count,
945 bool char_count,
946 bool halt_on_oob) {
947 if (stopped()) {
948 return; // already stopped
949 }
950 RegionNode* bailout = new RegionNode(1);
951 record_for_igvn(bailout);
952 if (char_count) {
953 // Convert char count to byte count
954 count = _gvn.transform(new LShiftINode(count, intcon(1)));
955 }
956
957 // Offset and count must not be negative
958 generate_negative_guard(offset, bailout);
959 generate_negative_guard(count, bailout);
960 // Offset + count must not exceed length of array
961 generate_limit_guard(offset, count, load_array_length(array), bailout);
962
963 if (bailout->req() > 1) {
964 if (halt_on_oob) {
965 bailout = _gvn.transform(bailout)->as_Region();
966 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
967 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
968 C->root()->add_req(halt);
969 } else {
970 PreserveJVMState pjvms(this);
971 set_control(_gvn.transform(bailout));
972 uncommon_trap(Deoptimization::Reason_intrinsic,
973 Deoptimization::Action_maybe_recompile);
974 }
975 }
976 }
977
978 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
979 bool is_immutable) {
980 ciKlass* thread_klass = env()->Thread_klass();
981 const Type* thread_type
982 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
983
984 Node* thread = _gvn.transform(new ThreadLocalNode());
985 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
986 tls_output = thread;
987
988 Node* thread_obj_handle
989 = (is_immutable
990 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
991 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
992 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
993 thread_obj_handle = _gvn.transform(thread_obj_handle);
994
995 DecoratorSet decorators = IN_NATIVE;
996 if (is_immutable) {
997 decorators |= C2_IMMUTABLE_MEMORY;
998 }
999 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1000 }
1001
1002 //--------------------------generate_current_thread--------------------
1003 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1004 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1005 /*is_immutable*/false);
1006 }
1007
1008 //--------------------------generate_virtual_thread--------------------
1009 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1010 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1011 !C->method()->changes_current_thread());
1012 }
1013
1014 //------------------------------make_string_method_node------------------------
1015 // Helper method for String intrinsic functions. This version is called with
1016 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1017 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1018 // containing the lengths of str1 and str2.
1019 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1020 Node* result = nullptr;
1021 switch (opcode) {
1022 case Op_StrIndexOf:
1023 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1024 str1_start, cnt1, str2_start, cnt2, ae);
1025 break;
1026 case Op_StrComp:
1027 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1028 str1_start, cnt1, str2_start, cnt2, ae);
1029 break;
1030 case Op_StrEquals:
1031 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1032 // Use the constant length if there is one because optimized match rule may exist.
1033 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1034 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1035 break;
1036 default:
1037 ShouldNotReachHere();
1038 return nullptr;
1039 }
1040
1041 // All these intrinsics have checks.
1042 C->set_has_split_ifs(true); // Has chance for split-if optimization
1043 clear_upper_avx();
1044
1045 return _gvn.transform(result);
1046 }
1047
1048 //------------------------------inline_string_compareTo------------------------
1049 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1050 Node* arg1 = argument(0);
1051 Node* arg2 = argument(1);
1052
1053 arg1 = must_be_not_null(arg1, true);
1054 arg2 = must_be_not_null(arg2, true);
1055
1056 // Get start addr and length of first argument
1057 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1058 Node* arg1_cnt = load_array_length(arg1);
1059
1060 // Get start addr and length of second argument
1061 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1062 Node* arg2_cnt = load_array_length(arg2);
1063
1064 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1065 set_result(result);
1066 return true;
1067 }
1068
1069 //------------------------------inline_string_equals------------------------
1070 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1071 Node* arg1 = argument(0);
1072 Node* arg2 = argument(1);
1073
1074 // paths (plus control) merge
1075 RegionNode* region = new RegionNode(3);
1076 Node* phi = new PhiNode(region, TypeInt::BOOL);
1077
1078 if (!stopped()) {
1079
1080 arg1 = must_be_not_null(arg1, true);
1081 arg2 = must_be_not_null(arg2, true);
1082
1083 // Get start addr and length of first argument
1084 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1085 Node* arg1_cnt = load_array_length(arg1);
1086
1087 // Get start addr and length of second argument
1088 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1089 Node* arg2_cnt = load_array_length(arg2);
1090
1091 // Check for arg1_cnt != arg2_cnt
1092 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1093 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1094 Node* if_ne = generate_slow_guard(bol, nullptr);
1095 if (if_ne != nullptr) {
1096 phi->init_req(2, intcon(0));
1097 region->init_req(2, if_ne);
1098 }
1099
1100 // Check for count == 0 is done by assembler code for StrEquals.
1101
1102 if (!stopped()) {
1103 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1104 phi->init_req(1, equals);
1105 region->init_req(1, control());
1106 }
1107 }
1108
1109 // post merge
1110 set_control(_gvn.transform(region));
1111 record_for_igvn(region);
1112
1113 set_result(_gvn.transform(phi));
1114 return true;
1115 }
1116
1117 //------------------------------inline_array_equals----------------------------
1118 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1119 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1120 Node* arg1 = argument(0);
1121 Node* arg2 = argument(1);
1122
1123 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1124 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1125 clear_upper_avx();
1126
1127 return true;
1128 }
1129
1130
1131 //------------------------------inline_countPositives------------------------------
1132 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1133 bool LibraryCallKit::inline_countPositives() {
1134 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1135 return false;
1136 }
1137
1138 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1139 // no receiver since it is static method
1140 Node* ba = argument(0);
1141 Node* offset = argument(1);
1142 Node* len = argument(2);
1143
1144 if (VerifyIntrinsicChecks) {
1145 ba = must_be_not_null(ba, true);
1146 generate_string_range_check(ba, offset, len, false, true);
1147 if (stopped()) {
1148 return true;
1149 }
1150 }
1151
1152 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1153 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1154 set_result(_gvn.transform(result));
1155 clear_upper_avx();
1156 return true;
1157 }
1158
1159 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1160 Node* index = argument(0);
1161 Node* length = bt == T_INT ? argument(1) : argument(2);
1162 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1163 return false;
1164 }
1165
1166 // check that length is positive
1167 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1168 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1169
1170 {
1171 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1172 uncommon_trap(Deoptimization::Reason_intrinsic,
1173 Deoptimization::Action_make_not_entrant);
1174 }
1175
1176 if (stopped()) {
1177 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1178 return true;
1179 }
1180
1181 // length is now known positive, add a cast node to make this explicit
1182 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1183 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1184 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1185 ConstraintCastNode::RegularDependency, bt);
1186 casted_length = _gvn.transform(casted_length);
1187 replace_in_map(length, casted_length);
1188 length = casted_length;
1189
1190 // Use an unsigned comparison for the range check itself
1191 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1192 BoolTest::mask btest = BoolTest::lt;
1193 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1194 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1195 _gvn.set_type(rc, rc->Value(&_gvn));
1196 if (!rc_bool->is_Con()) {
1197 record_for_igvn(rc);
1198 }
1199 set_control(_gvn.transform(new IfTrueNode(rc)));
1200 {
1201 PreserveJVMState pjvms(this);
1202 set_control(_gvn.transform(new IfFalseNode(rc)));
1203 uncommon_trap(Deoptimization::Reason_range_check,
1204 Deoptimization::Action_make_not_entrant);
1205 }
1206
1207 if (stopped()) {
1208 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1209 return true;
1210 }
1211
1212 // index is now known to be >= 0 and < length, cast it
1213 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1214 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1215 ConstraintCastNode::RegularDependency, bt);
1216 result = _gvn.transform(result);
1217 set_result(result);
1218 replace_in_map(index, result);
1219 return true;
1220 }
1221
1222 //------------------------------inline_string_indexOf------------------------
1223 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1224 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1225 return false;
1226 }
1227 Node* src = argument(0);
1228 Node* tgt = argument(1);
1229
1230 // Make the merge point
1231 RegionNode* result_rgn = new RegionNode(4);
1232 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1233
1234 src = must_be_not_null(src, true);
1235 tgt = must_be_not_null(tgt, true);
1236
1237 // Get start addr and length of source string
1238 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1239 Node* src_count = load_array_length(src);
1240
1241 // Get start addr and length of substring
1242 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1243 Node* tgt_count = load_array_length(tgt);
1244
1245 Node* result = nullptr;
1246 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1247
1248 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1249 // Divide src size by 2 if String is UTF16 encoded
1250 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1251 }
1252 if (ae == StrIntrinsicNode::UU) {
1253 // Divide substring size by 2 if String is UTF16 encoded
1254 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1255 }
1256
1257 if (call_opt_stub) {
1258 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1259 StubRoutines::_string_indexof_array[ae],
1260 "stringIndexOf", TypePtr::BOTTOM, src_start,
1261 src_count, tgt_start, tgt_count);
1262 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1263 } else {
1264 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1265 result_rgn, result_phi, ae);
1266 }
1267 if (result != nullptr) {
1268 result_phi->init_req(3, result);
1269 result_rgn->init_req(3, control());
1270 }
1271 set_control(_gvn.transform(result_rgn));
1272 record_for_igvn(result_rgn);
1273 set_result(_gvn.transform(result_phi));
1274
1275 return true;
1276 }
1277
1278 //-----------------------------inline_string_indexOfI-----------------------
1279 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1280 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1281 return false;
1282 }
1283 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1284 return false;
1285 }
1286
1287 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1288 Node* src = argument(0); // byte[]
1289 Node* src_count = argument(1); // char count
1290 Node* tgt = argument(2); // byte[]
1291 Node* tgt_count = argument(3); // char count
1292 Node* from_index = argument(4); // char index
1293
1294 src = must_be_not_null(src, true);
1295 tgt = must_be_not_null(tgt, true);
1296
1297 // Multiply byte array index by 2 if String is UTF16 encoded
1298 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1299 src_count = _gvn.transform(new SubINode(src_count, from_index));
1300 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1301 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1302
1303 // Range checks
1304 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1305 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1306 if (stopped()) {
1307 return true;
1308 }
1309
1310 RegionNode* region = new RegionNode(5);
1311 Node* phi = new PhiNode(region, TypeInt::INT);
1312 Node* result = nullptr;
1313
1314 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1315
1316 if (call_opt_stub) {
1317 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1318 StubRoutines::_string_indexof_array[ae],
1319 "stringIndexOf", TypePtr::BOTTOM, src_start,
1320 src_count, tgt_start, tgt_count);
1321 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1322 } else {
1323 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1324 region, phi, ae);
1325 }
1326 if (result != nullptr) {
1327 // The result is index relative to from_index if substring was found, -1 otherwise.
1328 // Generate code which will fold into cmove.
1329 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1330 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1331
1332 Node* if_lt = generate_slow_guard(bol, nullptr);
1333 if (if_lt != nullptr) {
1334 // result == -1
1335 phi->init_req(3, result);
1336 region->init_req(3, if_lt);
1337 }
1338 if (!stopped()) {
1339 result = _gvn.transform(new AddINode(result, from_index));
1340 phi->init_req(4, result);
1341 region->init_req(4, control());
1342 }
1343 }
1344
1345 set_control(_gvn.transform(region));
1346 record_for_igvn(region);
1347 set_result(_gvn.transform(phi));
1348 clear_upper_avx();
1349
1350 return true;
1351 }
1352
1353 // Create StrIndexOfNode with fast path checks
1354 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1355 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1356 // Check for substr count > string count
1357 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1358 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1359 Node* if_gt = generate_slow_guard(bol, nullptr);
1360 if (if_gt != nullptr) {
1361 phi->init_req(1, intcon(-1));
1362 region->init_req(1, if_gt);
1363 }
1364 if (!stopped()) {
1365 // Check for substr count == 0
1366 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1367 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1368 Node* if_zero = generate_slow_guard(bol, nullptr);
1369 if (if_zero != nullptr) {
1370 phi->init_req(2, intcon(0));
1371 region->init_req(2, if_zero);
1372 }
1373 }
1374 if (!stopped()) {
1375 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1376 }
1377 return nullptr;
1378 }
1379
1380 //-----------------------------inline_string_indexOfChar-----------------------
1381 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1382 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1383 return false;
1384 }
1385 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1386 return false;
1387 }
1388 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1389 Node* src = argument(0); // byte[]
1390 Node* int_ch = argument(1);
1391 Node* from_index = argument(2);
1392 Node* max = argument(3);
1393
1394 src = must_be_not_null(src, true);
1395
1396 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1397 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1398 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1399
1400 // Range checks
1401 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1402
1403 // Check for int_ch >= 0
1404 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1405 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1406 {
1407 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1408 uncommon_trap(Deoptimization::Reason_intrinsic,
1409 Deoptimization::Action_maybe_recompile);
1410 }
1411 if (stopped()) {
1412 return true;
1413 }
1414
1415 RegionNode* region = new RegionNode(3);
1416 Node* phi = new PhiNode(region, TypeInt::INT);
1417
1418 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1419 C->set_has_split_ifs(true); // Has chance for split-if optimization
1420 _gvn.transform(result);
1421
1422 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1423 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1424
1425 Node* if_lt = generate_slow_guard(bol, nullptr);
1426 if (if_lt != nullptr) {
1427 // result == -1
1428 phi->init_req(2, result);
1429 region->init_req(2, if_lt);
1430 }
1431 if (!stopped()) {
1432 result = _gvn.transform(new AddINode(result, from_index));
1433 phi->init_req(1, result);
1434 region->init_req(1, control());
1435 }
1436 set_control(_gvn.transform(region));
1437 record_for_igvn(region);
1438 set_result(_gvn.transform(phi));
1439 clear_upper_avx();
1440
1441 return true;
1442 }
1443 //---------------------------inline_string_copy---------------------
1444 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1445 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1446 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1447 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1448 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1449 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1450 bool LibraryCallKit::inline_string_copy(bool compress) {
1451 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1452 return false;
1453 }
1454 int nargs = 5; // 2 oops, 3 ints
1455 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1456
1457 Node* src = argument(0);
1458 Node* src_offset = argument(1);
1459 Node* dst = argument(2);
1460 Node* dst_offset = argument(3);
1461 Node* length = argument(4);
1462
1463 // Check for allocation before we add nodes that would confuse
1464 // tightly_coupled_allocation()
1465 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1466
1467 // Figure out the size and type of the elements we will be copying.
1468 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1469 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1470 if (src_type == nullptr || dst_type == nullptr) {
1471 return false;
1472 }
1473 BasicType src_elem = src_type->elem()->array_element_basic_type();
1474 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1475 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1476 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1477 "Unsupported array types for inline_string_copy");
1478
1479 src = must_be_not_null(src, true);
1480 dst = must_be_not_null(dst, true);
1481
1482 // Convert char[] offsets to byte[] offsets
1483 bool convert_src = (compress && src_elem == T_BYTE);
1484 bool convert_dst = (!compress && dst_elem == T_BYTE);
1485 if (convert_src) {
1486 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1487 } else if (convert_dst) {
1488 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1489 }
1490
1491 // Range checks
1492 generate_string_range_check(src, src_offset, length, convert_src);
1493 generate_string_range_check(dst, dst_offset, length, convert_dst);
1494 if (stopped()) {
1495 return true;
1496 }
1497
1498 Node* src_start = array_element_address(src, src_offset, src_elem);
1499 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1500 // 'src_start' points to src array + scaled offset
1501 // 'dst_start' points to dst array + scaled offset
1502 Node* count = nullptr;
1503 if (compress) {
1504 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1505 } else {
1506 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1507 }
1508
1509 if (alloc != nullptr) {
1510 if (alloc->maybe_set_complete(&_gvn)) {
1511 // "You break it, you buy it."
1512 InitializeNode* init = alloc->initialization();
1513 assert(init->is_complete(), "we just did this");
1514 init->set_complete_with_arraycopy();
1515 assert(dst->is_CheckCastPP(), "sanity");
1516 assert(dst->in(0)->in(0) == init, "dest pinned");
1517 }
1518 // Do not let stores that initialize this object be reordered with
1519 // a subsequent store that would make this object accessible by
1520 // other threads.
1521 // Record what AllocateNode this StoreStore protects so that
1522 // escape analysis can go from the MemBarStoreStoreNode to the
1523 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1524 // based on the escape status of the AllocateNode.
1525 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1526 }
1527 if (compress) {
1528 set_result(_gvn.transform(count));
1529 }
1530 clear_upper_avx();
1531
1532 return true;
1533 }
1534
1535 #ifdef _LP64
1536 #define XTOP ,top() /*additional argument*/
1537 #else //_LP64
1538 #define XTOP /*no additional argument*/
1539 #endif //_LP64
1540
1541 //------------------------inline_string_toBytesU--------------------------
1542 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1543 bool LibraryCallKit::inline_string_toBytesU() {
1544 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1545 return false;
1546 }
1547 // Get the arguments.
1548 Node* value = argument(0);
1549 Node* offset = argument(1);
1550 Node* length = argument(2);
1551
1552 Node* newcopy = nullptr;
1553
1554 // Set the original stack and the reexecute bit for the interpreter to reexecute
1555 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1556 { PreserveReexecuteState preexecs(this);
1557 jvms()->set_should_reexecute(true);
1558
1559 // Check if a null path was taken unconditionally.
1560 value = null_check(value);
1561
1562 RegionNode* bailout = new RegionNode(1);
1563 record_for_igvn(bailout);
1564
1565 // Range checks
1566 generate_negative_guard(offset, bailout);
1567 generate_negative_guard(length, bailout);
1568 generate_limit_guard(offset, length, load_array_length(value), bailout);
1569 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1570 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1571
1572 if (bailout->req() > 1) {
1573 PreserveJVMState pjvms(this);
1574 set_control(_gvn.transform(bailout));
1575 uncommon_trap(Deoptimization::Reason_intrinsic,
1576 Deoptimization::Action_maybe_recompile);
1577 }
1578 if (stopped()) {
1579 return true;
1580 }
1581
1582 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1583 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1584 newcopy = new_array(klass_node, size, 0); // no arguments to push
1585 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1586 guarantee(alloc != nullptr, "created above");
1587
1588 // Calculate starting addresses.
1589 Node* src_start = array_element_address(value, offset, T_CHAR);
1590 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1591
1592 // Check if dst array address is aligned to HeapWordSize
1593 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1594 // If true, then check if src array address is aligned to HeapWordSize
1595 if (aligned) {
1596 const TypeInt* toffset = gvn().type(offset)->is_int();
1597 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1598 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1599 }
1600
1601 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1602 const char* copyfunc_name = "arraycopy";
1603 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1604 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1605 OptoRuntime::fast_arraycopy_Type(),
1606 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1607 src_start, dst_start, ConvI2X(length) XTOP);
1608 // Do not let reads from the cloned object float above the arraycopy.
1609 if (alloc->maybe_set_complete(&_gvn)) {
1610 // "You break it, you buy it."
1611 InitializeNode* init = alloc->initialization();
1612 assert(init->is_complete(), "we just did this");
1613 init->set_complete_with_arraycopy();
1614 assert(newcopy->is_CheckCastPP(), "sanity");
1615 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1616 }
1617 // Do not let stores that initialize this object be reordered with
1618 // a subsequent store that would make this object accessible by
1619 // other threads.
1620 // Record what AllocateNode this StoreStore protects so that
1621 // escape analysis can go from the MemBarStoreStoreNode to the
1622 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1623 // based on the escape status of the AllocateNode.
1624 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1625 } // original reexecute is set back here
1626
1627 C->set_has_split_ifs(true); // Has chance for split-if optimization
1628 if (!stopped()) {
1629 set_result(newcopy);
1630 }
1631 clear_upper_avx();
1632
1633 return true;
1634 }
1635
1636 //------------------------inline_string_getCharsU--------------------------
1637 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1638 bool LibraryCallKit::inline_string_getCharsU() {
1639 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1640 return false;
1641 }
1642
1643 // Get the arguments.
1644 Node* src = argument(0);
1645 Node* src_begin = argument(1);
1646 Node* src_end = argument(2); // exclusive offset (i < src_end)
1647 Node* dst = argument(3);
1648 Node* dst_begin = argument(4);
1649
1650 // Check for allocation before we add nodes that would confuse
1651 // tightly_coupled_allocation()
1652 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1653
1654 // Check if a null path was taken unconditionally.
1655 src = null_check(src);
1656 dst = null_check(dst);
1657 if (stopped()) {
1658 return true;
1659 }
1660
1661 // Get length and convert char[] offset to byte[] offset
1662 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1663 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1664
1665 // Range checks
1666 generate_string_range_check(src, src_begin, length, true);
1667 generate_string_range_check(dst, dst_begin, length, false);
1668 if (stopped()) {
1669 return true;
1670 }
1671
1672 if (!stopped()) {
1673 // Calculate starting addresses.
1674 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1675 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1676
1677 // Check if array addresses are aligned to HeapWordSize
1678 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1679 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1680 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1681 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1682
1683 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1684 const char* copyfunc_name = "arraycopy";
1685 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1686 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1687 OptoRuntime::fast_arraycopy_Type(),
1688 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1689 src_start, dst_start, ConvI2X(length) XTOP);
1690 // Do not let reads from the cloned object float above the arraycopy.
1691 if (alloc != nullptr) {
1692 if (alloc->maybe_set_complete(&_gvn)) {
1693 // "You break it, you buy it."
1694 InitializeNode* init = alloc->initialization();
1695 assert(init->is_complete(), "we just did this");
1696 init->set_complete_with_arraycopy();
1697 assert(dst->is_CheckCastPP(), "sanity");
1698 assert(dst->in(0)->in(0) == init, "dest pinned");
1699 }
1700 // Do not let stores that initialize this object be reordered with
1701 // a subsequent store that would make this object accessible by
1702 // other threads.
1703 // Record what AllocateNode this StoreStore protects so that
1704 // escape analysis can go from the MemBarStoreStoreNode to the
1705 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1706 // based on the escape status of the AllocateNode.
1707 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1708 } else {
1709 insert_mem_bar(Op_MemBarCPUOrder);
1710 }
1711 }
1712
1713 C->set_has_split_ifs(true); // Has chance for split-if optimization
1714 return true;
1715 }
1716
1717 //----------------------inline_string_char_access----------------------------
1718 // Store/Load char to/from byte[] array.
1719 // static void StringUTF16.putChar(byte[] val, int index, int c)
1720 // static char StringUTF16.getChar(byte[] val, int index)
1721 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1722 Node* value = argument(0);
1723 Node* index = argument(1);
1724 Node* ch = is_store ? argument(2) : nullptr;
1725
1726 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1727 // correctly requires matched array shapes.
1728 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1729 "sanity: byte[] and char[] bases agree");
1730 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1731 "sanity: byte[] and char[] scales agree");
1732
1733 // Bail when getChar over constants is requested: constant folding would
1734 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1735 // Java method would constant fold nicely instead.
1736 if (!is_store && value->is_Con() && index->is_Con()) {
1737 return false;
1738 }
1739
1740 // Save state and restore on bailout
1741 SavedState old_state(this);
1742
1743 value = must_be_not_null(value, true);
1744
1745 Node* adr = array_element_address(value, index, T_CHAR);
1746 if (adr->is_top()) {
1747 return false;
1748 }
1749 old_state.discard();
1750 if (is_store) {
1751 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1752 } else {
1753 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);
1754 set_result(ch);
1755 }
1756 return true;
1757 }
1758
1759
1760 //------------------------------inline_math-----------------------------------
1761 // public static double Math.abs(double)
1762 // public static double Math.sqrt(double)
1763 // public static double Math.log(double)
1764 // public static double Math.log10(double)
1765 // public static double Math.round(double)
1766 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1767 Node* arg = argument(0);
1768 Node* n = nullptr;
1769 switch (id) {
1770 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1771 case vmIntrinsics::_dsqrt:
1772 case vmIntrinsics::_dsqrt_strict:
1773 n = new SqrtDNode(C, control(), arg); break;
1774 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1775 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1776 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1777 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1778 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1779 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1780 default: fatal_unexpected_iid(id); break;
1781 }
1782 set_result(_gvn.transform(n));
1783 return true;
1784 }
1785
1786 //------------------------------inline_math-----------------------------------
1787 // public static float Math.abs(float)
1788 // public static int Math.abs(int)
1789 // public static long Math.abs(long)
1790 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1791 Node* arg = argument(0);
1792 Node* n = nullptr;
1793 switch (id) {
1794 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1795 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1796 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1797 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1798 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1799 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1800 default: fatal_unexpected_iid(id); break;
1801 }
1802 set_result(_gvn.transform(n));
1803 return true;
1804 }
1805
1806 //------------------------------runtime_math-----------------------------
1807 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1808 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1809 "must be (DD)D or (D)D type");
1810
1811 // Inputs
1812 Node* a = argument(0);
1813 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1814
1815 const TypePtr* no_memory_effects = nullptr;
1816 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1817 no_memory_effects,
1818 a, top(), b, b ? top() : nullptr);
1819 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1820 #ifdef ASSERT
1821 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1822 assert(value_top == top(), "second value must be top");
1823 #endif
1824
1825 set_result(value);
1826 return true;
1827 }
1828
1829 //------------------------------inline_math_pow-----------------------------
1830 bool LibraryCallKit::inline_math_pow() {
1831 Node* exp = argument(2);
1832 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1833 if (d != nullptr) {
1834 if (d->getd() == 2.0) {
1835 // Special case: pow(x, 2.0) => x * x
1836 Node* base = argument(0);
1837 set_result(_gvn.transform(new MulDNode(base, base)));
1838 return true;
1839 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1840 // Special case: pow(x, 0.5) => sqrt(x)
1841 Node* base = argument(0);
1842 Node* zero = _gvn.zerocon(T_DOUBLE);
1843
1844 RegionNode* region = new RegionNode(3);
1845 Node* phi = new PhiNode(region, Type::DOUBLE);
1846
1847 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1848 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1849 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1850 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1851 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1852
1853 Node* if_pow = generate_slow_guard(test, nullptr);
1854 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1855 phi->init_req(1, value_sqrt);
1856 region->init_req(1, control());
1857
1858 if (if_pow != nullptr) {
1859 set_control(if_pow);
1860 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1861 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1862 const TypePtr* no_memory_effects = nullptr;
1863 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1864 no_memory_effects, base, top(), exp, top());
1865 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1866 #ifdef ASSERT
1867 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1868 assert(value_top == top(), "second value must be top");
1869 #endif
1870 phi->init_req(2, value_pow);
1871 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1872 }
1873
1874 C->set_has_split_ifs(true); // Has chance for split-if optimization
1875 set_control(_gvn.transform(region));
1876 record_for_igvn(region);
1877 set_result(_gvn.transform(phi));
1878
1879 return true;
1880 }
1881 }
1882
1883 return StubRoutines::dpow() != nullptr ?
1884 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1885 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1886 }
1887
1888 //------------------------------inline_math_native-----------------------------
1889 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1890 switch (id) {
1891 case vmIntrinsics::_dsin:
1892 return StubRoutines::dsin() != nullptr ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1894 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1895 case vmIntrinsics::_dcos:
1896 return StubRoutines::dcos() != nullptr ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1898 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1899 case vmIntrinsics::_dtan:
1900 return StubRoutines::dtan() != nullptr ?
1901 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1902 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1903 case vmIntrinsics::_dsinh:
1904 return StubRoutines::dsinh() != nullptr ?
1905 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1906 case vmIntrinsics::_dtanh:
1907 return StubRoutines::dtanh() != nullptr ?
1908 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1909 case vmIntrinsics::_dcbrt:
1910 return StubRoutines::dcbrt() != nullptr ?
1911 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1912 case vmIntrinsics::_dexp:
1913 return StubRoutines::dexp() != nullptr ?
1914 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1915 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1916 case vmIntrinsics::_dlog:
1917 return StubRoutines::dlog() != nullptr ?
1918 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1919 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1920 case vmIntrinsics::_dlog10:
1921 return StubRoutines::dlog10() != nullptr ?
1922 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1923 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1924
1925 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1926 case vmIntrinsics::_ceil:
1927 case vmIntrinsics::_floor:
1928 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1929
1930 case vmIntrinsics::_dsqrt:
1931 case vmIntrinsics::_dsqrt_strict:
1932 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1933 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1934 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1935 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1936 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1937
1938 case vmIntrinsics::_dpow: return inline_math_pow();
1939 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1940 case vmIntrinsics::_fcopySign: return inline_math(id);
1941 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1942 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1943 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1944
1945 // These intrinsics are not yet correctly implemented
1946 case vmIntrinsics::_datan2:
1947 return false;
1948
1949 default:
1950 fatal_unexpected_iid(id);
1951 return false;
1952 }
1953 }
1954
1955 //----------------------------inline_notify-----------------------------------*
1956 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1957 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1958 address func;
1959 if (id == vmIntrinsics::_notify) {
1960 func = OptoRuntime::monitor_notify_Java();
1961 } else {
1962 func = OptoRuntime::monitor_notifyAll_Java();
1963 }
1964 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1965 make_slow_call_ex(call, env()->Throwable_klass(), false);
1966 return true;
1967 }
1968
1969
1970 //----------------------------inline_min_max-----------------------------------
1971 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1972 Node* a = nullptr;
1973 Node* b = nullptr;
1974 Node* n = nullptr;
1975 switch (id) {
1976 case vmIntrinsics::_min:
1977 case vmIntrinsics::_max:
1978 case vmIntrinsics::_minF:
1979 case vmIntrinsics::_maxF:
1980 case vmIntrinsics::_minF_strict:
1981 case vmIntrinsics::_maxF_strict:
1982 case vmIntrinsics::_min_strict:
1983 case vmIntrinsics::_max_strict:
1984 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1985 a = argument(0);
1986 b = argument(1);
1987 break;
1988 case vmIntrinsics::_minD:
1989 case vmIntrinsics::_maxD:
1990 case vmIntrinsics::_minD_strict:
1991 case vmIntrinsics::_maxD_strict:
1992 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1993 a = argument(0);
1994 b = argument(2);
1995 break;
1996 case vmIntrinsics::_minL:
1997 case vmIntrinsics::_maxL:
1998 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1999 a = argument(0);
2000 b = argument(2);
2001 break;
2002 default:
2003 fatal_unexpected_iid(id);
2004 break;
2005 }
2006
2007 switch (id) {
2008 case vmIntrinsics::_min:
2009 case vmIntrinsics::_min_strict:
2010 n = new MinINode(a, b);
2011 break;
2012 case vmIntrinsics::_max:
2013 case vmIntrinsics::_max_strict:
2014 n = new MaxINode(a, b);
2015 break;
2016 case vmIntrinsics::_minF:
2017 case vmIntrinsics::_minF_strict:
2018 n = new MinFNode(a, b);
2019 break;
2020 case vmIntrinsics::_maxF:
2021 case vmIntrinsics::_maxF_strict:
2022 n = new MaxFNode(a, b);
2023 break;
2024 case vmIntrinsics::_minD:
2025 case vmIntrinsics::_minD_strict:
2026 n = new MinDNode(a, b);
2027 break;
2028 case vmIntrinsics::_maxD:
2029 case vmIntrinsics::_maxD_strict:
2030 n = new MaxDNode(a, b);
2031 break;
2032 case vmIntrinsics::_minL:
2033 n = new MinLNode(_gvn.C, a, b);
2034 break;
2035 case vmIntrinsics::_maxL:
2036 n = new MaxLNode(_gvn.C, a, b);
2037 break;
2038 default:
2039 fatal_unexpected_iid(id);
2040 break;
2041 }
2042
2043 set_result(_gvn.transform(n));
2044 return true;
2045 }
2046
2047 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2048 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2049 env()->ArithmeticException_instance())) {
2050 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2051 // so let's bail out intrinsic rather than risking deopting again.
2052 return false;
2053 }
2054
2055 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2056 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2057 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2058 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2059
2060 {
2061 PreserveJVMState pjvms(this);
2062 PreserveReexecuteState preexecs(this);
2063 jvms()->set_should_reexecute(true);
2064
2065 set_control(slow_path);
2066 set_i_o(i_o());
2067
2068 builtin_throw(Deoptimization::Reason_intrinsic,
2069 env()->ArithmeticException_instance(),
2070 /*allow_too_many_traps*/ false);
2071 }
2072
2073 set_control(fast_path);
2074 set_result(math);
2075 return true;
2076 }
2077
2078 template <typename OverflowOp>
2079 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2080 typedef typename OverflowOp::MathOp MathOp;
2081
2082 MathOp* mathOp = new MathOp(arg1, arg2);
2083 Node* operation = _gvn.transform( mathOp );
2084 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2085 return inline_math_mathExact(operation, ofcheck);
2086 }
2087
2088 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2089 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2090 }
2091
2092 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2093 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2094 }
2095
2096 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2097 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2098 }
2099
2100 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2101 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2102 }
2103
2104 bool LibraryCallKit::inline_math_negateExactI() {
2105 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2106 }
2107
2108 bool LibraryCallKit::inline_math_negateExactL() {
2109 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2110 }
2111
2112 bool LibraryCallKit::inline_math_multiplyExactI() {
2113 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2114 }
2115
2116 bool LibraryCallKit::inline_math_multiplyExactL() {
2117 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2118 }
2119
2120 bool LibraryCallKit::inline_math_multiplyHigh() {
2121 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2122 return true;
2123 }
2124
2125 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2126 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2127 return true;
2128 }
2129
2130 inline int
2131 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2132 const TypePtr* base_type = TypePtr::NULL_PTR;
2133 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2134 if (base_type == nullptr) {
2135 // Unknown type.
2136 return Type::AnyPtr;
2137 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2138 // Since this is a null+long form, we have to switch to a rawptr.
2139 base = _gvn.transform(new CastX2PNode(offset));
2140 offset = MakeConX(0);
2141 return Type::RawPtr;
2142 } else if (base_type->base() == Type::RawPtr) {
2143 return Type::RawPtr;
2144 } else if (base_type->isa_oopptr()) {
2145 // Base is never null => always a heap address.
2146 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2147 return Type::OopPtr;
2148 }
2149 // Offset is small => always a heap address.
2150 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2151 if (offset_type != nullptr &&
2152 base_type->offset() == 0 && // (should always be?)
2153 offset_type->_lo >= 0 &&
2154 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2155 return Type::OopPtr;
2156 } else if (type == T_OBJECT) {
2157 // off heap access to an oop doesn't make any sense. Has to be on
2158 // heap.
2159 return Type::OopPtr;
2160 }
2161 // Otherwise, it might either be oop+off or null+addr.
2162 return Type::AnyPtr;
2163 } else {
2164 // No information:
2165 return Type::AnyPtr;
2166 }
2167 }
2168
2169 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2170 Node* uncasted_base = base;
2171 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2172 if (kind == Type::RawPtr) {
2173 return basic_plus_adr(top(), uncasted_base, offset);
2174 } else if (kind == Type::AnyPtr) {
2175 assert(base == uncasted_base, "unexpected base change");
2176 if (can_cast) {
2177 if (!_gvn.type(base)->speculative_maybe_null() &&
2178 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2179 // According to profiling, this access is always on
2180 // heap. Casting the base to not null and thus avoiding membars
2181 // around the access should allow better optimizations
2182 Node* null_ctl = top();
2183 base = null_check_oop(base, &null_ctl, true, true, true);
2184 assert(null_ctl->is_top(), "no null control here");
2185 return basic_plus_adr(base, offset);
2186 } else if (_gvn.type(base)->speculative_always_null() &&
2187 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2188 // According to profiling, this access is always off
2189 // heap.
2190 base = null_assert(base);
2191 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2192 offset = MakeConX(0);
2193 return basic_plus_adr(top(), raw_base, offset);
2194 }
2195 }
2196 // We don't know if it's an on heap or off heap access. Fall back
2197 // to raw memory access.
2198 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2199 return basic_plus_adr(top(), raw, offset);
2200 } else {
2201 assert(base == uncasted_base, "unexpected base change");
2202 // We know it's an on heap access so base can't be null
2203 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2204 base = must_be_not_null(base, true);
2205 }
2206 return basic_plus_adr(base, offset);
2207 }
2208 }
2209
2210 //--------------------------inline_number_methods-----------------------------
2211 // inline int Integer.numberOfLeadingZeros(int)
2212 // inline int Long.numberOfLeadingZeros(long)
2213 //
2214 // inline int Integer.numberOfTrailingZeros(int)
2215 // inline int Long.numberOfTrailingZeros(long)
2216 //
2217 // inline int Integer.bitCount(int)
2218 // inline int Long.bitCount(long)
2219 //
2220 // inline char Character.reverseBytes(char)
2221 // inline short Short.reverseBytes(short)
2222 // inline int Integer.reverseBytes(int)
2223 // inline long Long.reverseBytes(long)
2224 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2225 Node* arg = argument(0);
2226 Node* n = nullptr;
2227 switch (id) {
2228 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2229 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2230 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2231 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2232 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2233 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2234 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2235 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2236 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2237 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2238 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2239 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2240 default: fatal_unexpected_iid(id); break;
2241 }
2242 set_result(_gvn.transform(n));
2243 return true;
2244 }
2245
2246 //--------------------------inline_bitshuffle_methods-----------------------------
2247 // inline int Integer.compress(int, int)
2248 // inline int Integer.expand(int, int)
2249 // inline long Long.compress(long, long)
2250 // inline long Long.expand(long, long)
2251 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2252 Node* n = nullptr;
2253 switch (id) {
2254 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2255 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2256 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2257 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2258 default: fatal_unexpected_iid(id); break;
2259 }
2260 set_result(_gvn.transform(n));
2261 return true;
2262 }
2263
2264 //--------------------------inline_number_methods-----------------------------
2265 // inline int Integer.compareUnsigned(int, int)
2266 // inline int Long.compareUnsigned(long, long)
2267 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2268 Node* arg1 = argument(0);
2269 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2270 Node* n = nullptr;
2271 switch (id) {
2272 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2273 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2274 default: fatal_unexpected_iid(id); break;
2275 }
2276 set_result(_gvn.transform(n));
2277 return true;
2278 }
2279
2280 //--------------------------inline_unsigned_divmod_methods-----------------------------
2281 // inline int Integer.divideUnsigned(int, int)
2282 // inline int Integer.remainderUnsigned(int, int)
2283 // inline long Long.divideUnsigned(long, long)
2284 // inline long Long.remainderUnsigned(long, long)
2285 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2286 Node* n = nullptr;
2287 switch (id) {
2288 case vmIntrinsics::_divideUnsigned_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 UDivINode(control(), argument(0), argument(1));
2295 break;
2296 }
2297 case vmIntrinsics::_divideUnsigned_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 UDivLNode(control(), argument(0), argument(2));
2304 break;
2305 }
2306 case vmIntrinsics::_remainderUnsigned_i: {
2307 zero_check_int(argument(1));
2308 // Compile-time detect of null-exception
2309 if (stopped()) {
2310 return true; // keep the graph constructed so far
2311 }
2312 n = new UModINode(control(), argument(0), argument(1));
2313 break;
2314 }
2315 case vmIntrinsics::_remainderUnsigned_l: {
2316 zero_check_long(argument(2));
2317 // Compile-time detect of null-exception
2318 if (stopped()) {
2319 return true; // keep the graph constructed so far
2320 }
2321 n = new UModLNode(control(), argument(0), argument(2));
2322 break;
2323 }
2324 default: fatal_unexpected_iid(id); break;
2325 }
2326 set_result(_gvn.transform(n));
2327 return true;
2328 }
2329
2330 //----------------------------inline_unsafe_access----------------------------
2331
2332 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2333 // Attempt to infer a sharper value type from the offset and base type.
2334 ciKlass* sharpened_klass = nullptr;
2335
2336 // See if it is an instance field, with an object type.
2337 if (alias_type->field() != nullptr) {
2338 if (alias_type->field()->type()->is_klass()) {
2339 sharpened_klass = alias_type->field()->type()->as_klass();
2340 }
2341 }
2342
2343 const TypeOopPtr* result = nullptr;
2344 // See if it is a narrow oop array.
2345 if (adr_type->isa_aryptr()) {
2346 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2347 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2348 if (elem_type != nullptr && elem_type->is_loaded()) {
2349 // Sharpen the value type.
2350 result = elem_type;
2351 }
2352 }
2353 }
2354
2355 // The sharpened class might be unloaded if there is no class loader
2356 // contraint in place.
2357 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2358 // Sharpen the value type.
2359 result = TypeOopPtr::make_from_klass(sharpened_klass);
2360 }
2361 if (result != nullptr) {
2362 #ifndef PRODUCT
2363 if (C->print_intrinsics() || C->print_inlining()) {
2364 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2365 tty->print(" sharpened value: "); result->dump(); tty->cr();
2366 }
2367 #endif
2368 }
2369 return result;
2370 }
2371
2372 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2373 switch (kind) {
2374 case Relaxed:
2375 return MO_UNORDERED;
2376 case Opaque:
2377 return MO_RELAXED;
2378 case Acquire:
2379 return MO_ACQUIRE;
2380 case Release:
2381 return MO_RELEASE;
2382 case Volatile:
2383 return MO_SEQ_CST;
2384 default:
2385 ShouldNotReachHere();
2386 return 0;
2387 }
2388 }
2389
2390 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2391 _kit(kit),
2392 _sp(kit->sp()),
2393 _jvms(kit->jvms()),
2394 _map(kit->clone_map()),
2395 _discarded(false)
2396 {
2397 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2398 Node* out = kit->control()->fast_out(i);
2399 if (out->is_CFG()) {
2400 _ctrl_succ.push(out);
2401 }
2402 }
2403 }
2404
2405 LibraryCallKit::SavedState::~SavedState() {
2406 if (_discarded) {
2407 _kit->destruct_map_clone(_map);
2408 return;
2409 }
2410 _kit->jvms()->set_map(_map);
2411 _kit->jvms()->set_sp(_sp);
2412 _map->set_jvms(_kit->jvms());
2413 _kit->set_map(_map);
2414 _kit->set_sp(_sp);
2415 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2416 Node* out = _kit->control()->fast_out(i);
2417 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2418 _kit->_gvn.hash_delete(out);
2419 out->set_req(0, _kit->C->top());
2420 _kit->C->record_for_igvn(out);
2421 --i; --imax;
2422 _kit->_gvn.hash_find_insert(out);
2423 }
2424 }
2425 }
2426
2427 void LibraryCallKit::SavedState::discard() {
2428 _discarded = true;
2429 }
2430
2431 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2432 if (callee()->is_static()) return false; // caller must have the capability!
2433 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2434 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2435 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2436 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2437
2438 if (is_reference_type(type)) {
2439 decorators |= ON_UNKNOWN_OOP_REF;
2440 }
2441
2442 if (unaligned) {
2443 decorators |= C2_UNALIGNED;
2444 }
2445
2446 #ifndef PRODUCT
2447 {
2448 ResourceMark rm;
2449 // Check the signatures.
2450 ciSignature* sig = callee()->signature();
2451 #ifdef ASSERT
2452 if (!is_store) {
2453 // Object getReference(Object base, int/long offset), etc.
2454 BasicType rtype = sig->return_type()->basic_type();
2455 assert(rtype == type, "getter must return the expected value");
2456 assert(sig->count() == 2, "oop getter has 2 arguments");
2457 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2458 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2459 } else {
2460 // void putReference(Object base, int/long offset, Object x), etc.
2461 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2462 assert(sig->count() == 3, "oop putter has 3 arguments");
2463 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2464 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2465 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2466 assert(vtype == type, "putter must accept the expected value");
2467 }
2468 #endif // ASSERT
2469 }
2470 #endif //PRODUCT
2471
2472 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2473
2474 Node* receiver = argument(0); // type: oop
2475
2476 // Build address expression.
2477 Node* heap_base_oop = top();
2478
2479 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2480 Node* base = argument(1); // type: oop
2481 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2482 Node* offset = argument(2); // type: long
2483 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2484 // to be plain byte offsets, which are also the same as those accepted
2485 // by oopDesc::field_addr.
2486 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2487 "fieldOffset must be byte-scaled");
2488 // 32-bit machines ignore the high half!
2489 offset = ConvL2X(offset);
2490
2491 // Save state and restore on bailout
2492 SavedState old_state(this);
2493
2494 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2495 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2496
2497 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2498 if (type != T_OBJECT) {
2499 decorators |= IN_NATIVE; // off-heap primitive access
2500 } else {
2501 return false; // off-heap oop accesses are not supported
2502 }
2503 } else {
2504 heap_base_oop = base; // on-heap or mixed access
2505 }
2506
2507 // Can base be null? Otherwise, always on-heap access.
2508 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2509
2510 if (!can_access_non_heap) {
2511 decorators |= IN_HEAP;
2512 }
2513
2514 Node* val = is_store ? argument(4) : nullptr;
2515
2516 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2517 if (adr_type == TypePtr::NULL_PTR) {
2518 return false; // off-heap access with zero address
2519 }
2520
2521 // Try to categorize the address.
2522 Compile::AliasType* alias_type = C->alias_type(adr_type);
2523 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2524
2525 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2526 alias_type->adr_type() == TypeAryPtr::RANGE) {
2527 return false; // not supported
2528 }
2529
2530 bool mismatched = false;
2531 BasicType bt = alias_type->basic_type();
2532 if (bt != T_ILLEGAL) {
2533 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2534 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2535 // Alias type doesn't differentiate between byte[] and boolean[]).
2536 // Use address type to get the element type.
2537 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2538 }
2539 if (is_reference_type(bt, true)) {
2540 // accessing an array field with getReference is not a mismatch
2541 bt = T_OBJECT;
2542 }
2543 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2544 // Don't intrinsify mismatched object accesses
2545 return false;
2546 }
2547 mismatched = (bt != type);
2548 } else if (alias_type->adr_type()->isa_oopptr()) {
2549 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2550 }
2551
2552 old_state.discard();
2553 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2554
2555 if (mismatched) {
2556 decorators |= C2_MISMATCHED;
2557 }
2558
2559 // First guess at the value type.
2560 const Type *value_type = Type::get_const_basic_type(type);
2561
2562 // Figure out the memory ordering.
2563 decorators |= mo_decorator_for_access_kind(kind);
2564
2565 if (!is_store && type == T_OBJECT) {
2566 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2567 if (tjp != nullptr) {
2568 value_type = tjp;
2569 }
2570 }
2571
2572 receiver = null_check(receiver);
2573 if (stopped()) {
2574 return true;
2575 }
2576 // Heap pointers get a null-check from the interpreter,
2577 // as a courtesy. However, this is not guaranteed by Unsafe,
2578 // and it is not possible to fully distinguish unintended nulls
2579 // from intended ones in this API.
2580
2581 if (!is_store) {
2582 Node* p = nullptr;
2583 // Try to constant fold a load from a constant field
2584 ciField* field = alias_type->field();
2585 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2586 // final or stable field
2587 p = make_constant_from_field(field, heap_base_oop);
2588 }
2589
2590 if (p == nullptr) { // Could not constant fold the load
2591 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2592 // Normalize the value returned by getBoolean in the following cases
2593 if (type == T_BOOLEAN &&
2594 (mismatched ||
2595 heap_base_oop == top() || // - heap_base_oop is null or
2596 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2597 // and the unsafe access is made to large offset
2598 // (i.e., larger than the maximum offset necessary for any
2599 // field access)
2600 ) {
2601 IdealKit ideal = IdealKit(this);
2602 #define __ ideal.
2603 IdealVariable normalized_result(ideal);
2604 __ declarations_done();
2605 __ set(normalized_result, p);
2606 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2607 __ set(normalized_result, ideal.ConI(1));
2608 ideal.end_if();
2609 final_sync(ideal);
2610 p = __ value(normalized_result);
2611 #undef __
2612 }
2613 }
2614 if (type == T_ADDRESS) {
2615 p = gvn().transform(new CastP2XNode(nullptr, p));
2616 p = ConvX2UL(p);
2617 }
2618 // The load node has the control of the preceding MemBarCPUOrder. All
2619 // following nodes will have the control of the MemBarCPUOrder inserted at
2620 // the end of this method. So, pushing the load onto the stack at a later
2621 // point is fine.
2622 set_result(p);
2623 } else {
2624 if (bt == T_ADDRESS) {
2625 // Repackage the long as a pointer.
2626 val = ConvL2X(val);
2627 val = gvn().transform(new CastX2PNode(val));
2628 }
2629 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2630 }
2631
2632 return true;
2633 }
2634
2635 //----------------------------inline_unsafe_load_store----------------------------
2636 // This method serves a couple of different customers (depending on LoadStoreKind):
2637 //
2638 // LS_cmp_swap:
2639 //
2640 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2641 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2642 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2643 //
2644 // LS_cmp_swap_weak:
2645 //
2646 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2647 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2648 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2649 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2650 //
2651 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2652 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2653 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2654 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2655 //
2656 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2657 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2658 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2659 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2660 //
2661 // LS_cmp_exchange:
2662 //
2663 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2664 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2665 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2666 //
2667 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2668 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2669 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2670 //
2671 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2672 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2673 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2674 //
2675 // LS_get_add:
2676 //
2677 // int getAndAddInt( Object o, long offset, int delta)
2678 // long getAndAddLong(Object o, long offset, long delta)
2679 //
2680 // LS_get_set:
2681 //
2682 // int getAndSet(Object o, long offset, int newValue)
2683 // long getAndSet(Object o, long offset, long newValue)
2684 // Object getAndSet(Object o, long offset, Object newValue)
2685 //
2686 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2687 // This basic scheme here is the same as inline_unsafe_access, but
2688 // differs in enough details that combining them would make the code
2689 // overly confusing. (This is a true fact! I originally combined
2690 // them, but even I was confused by it!) As much code/comments as
2691 // possible are retained from inline_unsafe_access though to make
2692 // the correspondences clearer. - dl
2693
2694 if (callee()->is_static()) return false; // caller must have the capability!
2695
2696 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2697 decorators |= mo_decorator_for_access_kind(access_kind);
2698
2699 #ifndef PRODUCT
2700 BasicType rtype;
2701 {
2702 ResourceMark rm;
2703 // Check the signatures.
2704 ciSignature* sig = callee()->signature();
2705 rtype = sig->return_type()->basic_type();
2706 switch(kind) {
2707 case LS_get_add:
2708 case LS_get_set: {
2709 // Check the signatures.
2710 #ifdef ASSERT
2711 assert(rtype == type, "get and set must return the expected type");
2712 assert(sig->count() == 3, "get and set has 3 arguments");
2713 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2714 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2715 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2716 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2717 #endif // ASSERT
2718 break;
2719 }
2720 case LS_cmp_swap:
2721 case LS_cmp_swap_weak: {
2722 // Check the signatures.
2723 #ifdef ASSERT
2724 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2725 assert(sig->count() == 4, "CAS has 4 arguments");
2726 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2727 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2728 #endif // ASSERT
2729 break;
2730 }
2731 case LS_cmp_exchange: {
2732 // Check the signatures.
2733 #ifdef ASSERT
2734 assert(rtype == type, "CAS must return the expected type");
2735 assert(sig->count() == 4, "CAS has 4 arguments");
2736 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2737 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2738 #endif // ASSERT
2739 break;
2740 }
2741 default:
2742 ShouldNotReachHere();
2743 }
2744 }
2745 #endif //PRODUCT
2746
2747 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2748
2749 // Get arguments:
2750 Node* receiver = nullptr;
2751 Node* base = nullptr;
2752 Node* offset = nullptr;
2753 Node* oldval = nullptr;
2754 Node* newval = nullptr;
2755 switch(kind) {
2756 case LS_cmp_swap:
2757 case LS_cmp_swap_weak:
2758 case LS_cmp_exchange: {
2759 const bool two_slot_type = type2size[type] == 2;
2760 receiver = argument(0); // type: oop
2761 base = argument(1); // type: oop
2762 offset = argument(2); // type: long
2763 oldval = argument(4); // type: oop, int, or long
2764 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2765 break;
2766 }
2767 case LS_get_add:
2768 case LS_get_set: {
2769 receiver = argument(0); // type: oop
2770 base = argument(1); // type: oop
2771 offset = argument(2); // type: long
2772 oldval = nullptr;
2773 newval = argument(4); // type: oop, int, or long
2774 break;
2775 }
2776 default:
2777 ShouldNotReachHere();
2778 }
2779
2780 // Build field offset expression.
2781 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2782 // to be plain byte offsets, which are also the same as those accepted
2783 // by oopDesc::field_addr.
2784 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2785 // 32-bit machines ignore the high half of long offsets
2786 offset = ConvL2X(offset);
2787 // Save state and restore on bailout
2788 SavedState old_state(this);
2789 Node* adr = make_unsafe_address(base, offset,type, false);
2790 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2791
2792 Compile::AliasType* alias_type = C->alias_type(adr_type);
2793 BasicType bt = alias_type->basic_type();
2794 if (bt != T_ILLEGAL &&
2795 (is_reference_type(bt) != (type == T_OBJECT))) {
2796 // Don't intrinsify mismatched object accesses.
2797 return false;
2798 }
2799
2800 old_state.discard();
2801
2802 // For CAS, unlike inline_unsafe_access, there seems no point in
2803 // trying to refine types. Just use the coarse types here.
2804 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2805 const Type *value_type = Type::get_const_basic_type(type);
2806
2807 switch (kind) {
2808 case LS_get_set:
2809 case LS_cmp_exchange: {
2810 if (type == T_OBJECT) {
2811 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2812 if (tjp != nullptr) {
2813 value_type = tjp;
2814 }
2815 }
2816 break;
2817 }
2818 case LS_cmp_swap:
2819 case LS_cmp_swap_weak:
2820 case LS_get_add:
2821 break;
2822 default:
2823 ShouldNotReachHere();
2824 }
2825
2826 // Null check receiver.
2827 receiver = null_check(receiver);
2828 if (stopped()) {
2829 return true;
2830 }
2831
2832 int alias_idx = C->get_alias_index(adr_type);
2833
2834 if (is_reference_type(type)) {
2835 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2836
2837 // Transformation of a value which could be null pointer (CastPP #null)
2838 // could be delayed during Parse (for example, in adjust_map_after_if()).
2839 // Execute transformation here to avoid barrier generation in such case.
2840 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2841 newval = _gvn.makecon(TypePtr::NULL_PTR);
2842
2843 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2844 // Refine the value to a null constant, when it is known to be null
2845 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2846 }
2847 }
2848
2849 Node* result = nullptr;
2850 switch (kind) {
2851 case LS_cmp_exchange: {
2852 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2853 oldval, newval, value_type, type, decorators);
2854 break;
2855 }
2856 case LS_cmp_swap_weak:
2857 decorators |= C2_WEAK_CMPXCHG;
2858 case LS_cmp_swap: {
2859 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2860 oldval, newval, value_type, type, decorators);
2861 break;
2862 }
2863 case LS_get_set: {
2864 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2865 newval, value_type, type, decorators);
2866 break;
2867 }
2868 case LS_get_add: {
2869 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2870 newval, value_type, type, decorators);
2871 break;
2872 }
2873 default:
2874 ShouldNotReachHere();
2875 }
2876
2877 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2878 set_result(result);
2879 return true;
2880 }
2881
2882 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2883 // Regardless of form, don't allow previous ld/st to move down,
2884 // then issue acquire, release, or volatile mem_bar.
2885 insert_mem_bar(Op_MemBarCPUOrder);
2886 switch(id) {
2887 case vmIntrinsics::_loadFence:
2888 insert_mem_bar(Op_LoadFence);
2889 return true;
2890 case vmIntrinsics::_storeFence:
2891 insert_mem_bar(Op_StoreFence);
2892 return true;
2893 case vmIntrinsics::_storeStoreFence:
2894 insert_mem_bar(Op_StoreStoreFence);
2895 return true;
2896 case vmIntrinsics::_fullFence:
2897 insert_mem_bar(Op_MemBarVolatile);
2898 return true;
2899 default:
2900 fatal_unexpected_iid(id);
2901 return false;
2902 }
2903 }
2904
2905 bool LibraryCallKit::inline_onspinwait() {
2906 insert_mem_bar(Op_OnSpinWait);
2907 return true;
2908 }
2909
2910 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2911 if (!kls->is_Con()) {
2912 return true;
2913 }
2914 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2915 if (klsptr == nullptr) {
2916 return true;
2917 }
2918 ciInstanceKlass* ik = klsptr->instance_klass();
2919 // don't need a guard for a klass that is already initialized
2920 return !ik->is_initialized();
2921 }
2922
2923 //----------------------------inline_unsafe_writeback0-------------------------
2924 // public native void Unsafe.writeback0(long address)
2925 bool LibraryCallKit::inline_unsafe_writeback0() {
2926 if (!Matcher::has_match_rule(Op_CacheWB)) {
2927 return false;
2928 }
2929 #ifndef PRODUCT
2930 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2931 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2932 ciSignature* sig = callee()->signature();
2933 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2934 #endif
2935 null_check_receiver(); // null-check, then ignore
2936 Node *addr = argument(1);
2937 addr = new CastX2PNode(addr);
2938 addr = _gvn.transform(addr);
2939 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2940 flush = _gvn.transform(flush);
2941 set_memory(flush, TypeRawPtr::BOTTOM);
2942 return true;
2943 }
2944
2945 //----------------------------inline_unsafe_writeback0-------------------------
2946 // public native void Unsafe.writeback0(long address)
2947 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2948 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2949 return false;
2950 }
2951 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2952 return false;
2953 }
2954 #ifndef PRODUCT
2955 assert(Matcher::has_match_rule(Op_CacheWB),
2956 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2957 : "found match rule for CacheWBPostSync but not CacheWB"));
2958
2959 #endif
2960 null_check_receiver(); // null-check, then ignore
2961 Node *sync;
2962 if (is_pre) {
2963 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2964 } else {
2965 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2966 }
2967 sync = _gvn.transform(sync);
2968 set_memory(sync, TypeRawPtr::BOTTOM);
2969 return true;
2970 }
2971
2972 //----------------------------inline_unsafe_allocate---------------------------
2973 // public native Object Unsafe.allocateInstance(Class<?> cls);
2974 bool LibraryCallKit::inline_unsafe_allocate() {
2975
2976 #if INCLUDE_JVMTI
2977 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2978 return false;
2979 }
2980 #endif //INCLUDE_JVMTI
2981
2982 if (callee()->is_static()) return false; // caller must have the capability!
2983
2984 null_check_receiver(); // null-check, then ignore
2985 Node* cls = null_check(argument(1));
2986 if (stopped()) return true;
2987
2988 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2989 kls = null_check(kls);
2990 if (stopped()) return true; // argument was like int.class
2991
2992 #if INCLUDE_JVMTI
2993 // Don't try to access new allocated obj in the intrinsic.
2994 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2995 // Deoptimize and allocate in interpreter instead.
2996 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2997 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2998 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2999 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3000 {
3001 BuildCutout unless(this, tst, PROB_MAX);
3002 uncommon_trap(Deoptimization::Reason_intrinsic,
3003 Deoptimization::Action_make_not_entrant);
3004 }
3005 if (stopped()) {
3006 return true;
3007 }
3008 #endif //INCLUDE_JVMTI
3009
3010 Node* test = nullptr;
3011 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3012 // Note: The argument might still be an illegal value like
3013 // Serializable.class or Object[].class. The runtime will handle it.
3014 // But we must make an explicit check for initialization.
3015 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3016 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3017 // can generate code to load it as unsigned byte.
3018 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3019 Node* bits = intcon(InstanceKlass::fully_initialized);
3020 test = _gvn.transform(new SubINode(inst, bits));
3021 // The 'test' is non-zero if we need to take a slow path.
3022 }
3023
3024 Node* obj = new_instance(kls, test);
3025 set_result(obj);
3026 return true;
3027 }
3028
3029 //------------------------inline_native_time_funcs--------------
3030 // inline code for System.currentTimeMillis() and System.nanoTime()
3031 // these have the same type and signature
3032 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3033 const TypeFunc* tf = OptoRuntime::void_long_Type();
3034 const TypePtr* no_memory_effects = nullptr;
3035 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3036 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3037 #ifdef ASSERT
3038 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3039 assert(value_top == top(), "second value must be top");
3040 #endif
3041 set_result(value);
3042 return true;
3043 }
3044
3045
3046 #if INCLUDE_JVMTI
3047
3048 // When notifications are disabled then just update the VTMS transition bit and return.
3049 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3050 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3051 if (!DoJVMTIVirtualThreadTransitions) {
3052 return true;
3053 }
3054 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3055 IdealKit ideal(this);
3056
3057 Node* ONE = ideal.ConI(1);
3058 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3059 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3060 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3061
3062 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3063 sync_kit(ideal);
3064 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3065 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3066 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3067 ideal.sync_kit(this);
3068 } ideal.else_(); {
3069 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3070 Node* thread = ideal.thread();
3071 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3072 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3073
3074 sync_kit(ideal);
3075 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3076 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3077
3078 ideal.sync_kit(this);
3079 } ideal.end_if();
3080 final_sync(ideal);
3081
3082 return true;
3083 }
3084
3085 // Always update the is_disable_suspend bit.
3086 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3087 if (!DoJVMTIVirtualThreadTransitions) {
3088 return true;
3089 }
3090 IdealKit ideal(this);
3091
3092 {
3093 // unconditionally update the is_disable_suspend bit in current JavaThread
3094 Node* thread = ideal.thread();
3095 Node* arg = _gvn.transform(argument(0)); // argument for notification
3096 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3097 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3098
3099 sync_kit(ideal);
3100 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3101 ideal.sync_kit(this);
3102 }
3103 final_sync(ideal);
3104
3105 return true;
3106 }
3107
3108 #endif // INCLUDE_JVMTI
3109
3110 #ifdef JFR_HAVE_INTRINSICS
3111
3112 /**
3113 * if oop->klass != null
3114 * // normal class
3115 * epoch = _epoch_state ? 2 : 1
3116 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3117 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3118 * }
3119 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3120 * else
3121 * // primitive class
3122 * if oop->array_klass != null
3123 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3124 * else
3125 * id = LAST_TYPE_ID + 1 // void class path
3126 * if (!signaled)
3127 * signaled = true
3128 */
3129 bool LibraryCallKit::inline_native_classID() {
3130 Node* cls = argument(0);
3131
3132 IdealKit ideal(this);
3133 #define __ ideal.
3134 IdealVariable result(ideal); __ declarations_done();
3135 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3136 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3137 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3138
3139
3140 __ if_then(kls, BoolTest::ne, null()); {
3141 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3142 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3143
3144 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3145 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3146 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3147 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3148 mask = _gvn.transform(new OrLNode(mask, epoch));
3149 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3150
3151 float unlikely = PROB_UNLIKELY(0.999);
3152 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3153 sync_kit(ideal);
3154 make_runtime_call(RC_LEAF,
3155 OptoRuntime::class_id_load_barrier_Type(),
3156 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3157 "class id load barrier",
3158 TypePtr::BOTTOM,
3159 kls);
3160 ideal.sync_kit(this);
3161 } __ end_if();
3162
3163 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3164 } __ else_(); {
3165 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3166 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3167 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3168 __ if_then(array_kls, BoolTest::ne, null()); {
3169 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3170 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3171 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3172 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3173 } __ else_(); {
3174 // void class case
3175 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3176 } __ end_if();
3177
3178 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3179 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3180 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3181 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3182 } __ end_if();
3183 } __ end_if();
3184
3185 final_sync(ideal);
3186 set_result(ideal.value(result));
3187 #undef __
3188 return true;
3189 }
3190
3191 //------------------------inline_native_jvm_commit------------------
3192 bool LibraryCallKit::inline_native_jvm_commit() {
3193 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3194
3195 // Save input memory and i_o state.
3196 Node* input_memory_state = reset_memory();
3197 set_all_memory(input_memory_state);
3198 Node* input_io_state = i_o();
3199
3200 // TLS.
3201 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3202 // Jfr java buffer.
3203 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3204 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3205 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3206
3207 // Load the current value of the notified field in the JfrThreadLocal.
3208 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3209 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3210
3211 // Test for notification.
3212 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3213 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3214 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3215
3216 // True branch, is notified.
3217 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3218 set_control(is_notified);
3219
3220 // Reset notified state.
3221 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3222 Node* notified_reset_memory = reset_memory();
3223
3224 // 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.
3225 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3226 // Convert the machine-word to a long.
3227 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3228
3229 // False branch, not notified.
3230 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3231 set_control(not_notified);
3232 set_all_memory(input_memory_state);
3233
3234 // Arg is the next position as a long.
3235 Node* arg = argument(0);
3236 // Convert long to machine-word.
3237 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3238
3239 // Store the next_position to the underlying jfr java buffer.
3240 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3241
3242 Node* commit_memory = reset_memory();
3243 set_all_memory(commit_memory);
3244
3245 // 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.
3246 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3247 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3248 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3249
3250 // And flags with lease constant.
3251 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3252
3253 // Branch on lease to conditionalize returning the leased java buffer.
3254 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3255 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3256 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3257
3258 // False branch, not a lease.
3259 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3260
3261 // True branch, is lease.
3262 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3263 set_control(is_lease);
3264
3265 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3266 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3267 OptoRuntime::void_void_Type(),
3268 SharedRuntime::jfr_return_lease(),
3269 "return_lease", TypePtr::BOTTOM);
3270 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3271
3272 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3273 record_for_igvn(lease_compare_rgn);
3274 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3275 record_for_igvn(lease_compare_mem);
3276 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3277 record_for_igvn(lease_compare_io);
3278 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3279 record_for_igvn(lease_result_value);
3280
3281 // Update control and phi nodes.
3282 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3283 lease_compare_rgn->init_req(_false_path, not_lease);
3284
3285 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3286 lease_compare_mem->init_req(_false_path, commit_memory);
3287
3288 lease_compare_io->init_req(_true_path, i_o());
3289 lease_compare_io->init_req(_false_path, input_io_state);
3290
3291 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3292 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3293
3294 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3295 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3296 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3297 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3298
3299 // Update control and phi nodes.
3300 result_rgn->init_req(_true_path, is_notified);
3301 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3302
3303 result_mem->init_req(_true_path, notified_reset_memory);
3304 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3305
3306 result_io->init_req(_true_path, input_io_state);
3307 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3308
3309 result_value->init_req(_true_path, current_pos);
3310 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3311
3312 // Set output state.
3313 set_control(_gvn.transform(result_rgn));
3314 set_all_memory(_gvn.transform(result_mem));
3315 set_i_o(_gvn.transform(result_io));
3316 set_result(result_rgn, result_value);
3317 return true;
3318 }
3319
3320 /*
3321 * The intrinsic is a model of this pseudo-code:
3322 *
3323 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3324 * jobject h_event_writer = tl->java_event_writer();
3325 * if (h_event_writer == nullptr) {
3326 * return nullptr;
3327 * }
3328 * oop threadObj = Thread::threadObj();
3329 * oop vthread = java_lang_Thread::vthread(threadObj);
3330 * traceid tid;
3331 * bool pinVirtualThread;
3332 * bool excluded;
3333 * if (vthread != threadObj) { // i.e. current thread is virtual
3334 * tid = java_lang_Thread::tid(vthread);
3335 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3336 * pinVirtualThread = VMContinuations;
3337 * excluded = vthread_epoch_raw & excluded_mask;
3338 * if (!excluded) {
3339 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3340 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3341 * if (vthread_epoch != current_epoch) {
3342 * write_checkpoint();
3343 * }
3344 * }
3345 * } else {
3346 * tid = java_lang_Thread::tid(threadObj);
3347 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3348 * pinVirtualThread = false;
3349 * excluded = thread_epoch_raw & excluded_mask;
3350 * }
3351 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3352 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3353 * if (tid_in_event_writer != tid) {
3354 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3355 * setField(event_writer, "excluded", excluded);
3356 * setField(event_writer, "threadID", tid);
3357 * }
3358 * return event_writer
3359 */
3360 bool LibraryCallKit::inline_native_getEventWriter() {
3361 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3362
3363 // Save input memory and i_o state.
3364 Node* input_memory_state = reset_memory();
3365 set_all_memory(input_memory_state);
3366 Node* input_io_state = i_o();
3367
3368 // The most significant bit of the u2 is used to denote thread exclusion
3369 Node* excluded_shift = _gvn.intcon(15);
3370 Node* excluded_mask = _gvn.intcon(1 << 15);
3371 // The epoch generation is the range [1-32767]
3372 Node* epoch_mask = _gvn.intcon(32767);
3373
3374 // TLS
3375 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3376
3377 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3378 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3379
3380 // Load the eventwriter jobject handle.
3381 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3382
3383 // Null check the jobject handle.
3384 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3385 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3386 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3387
3388 // False path, jobj is null.
3389 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3390
3391 // True path, jobj is not null.
3392 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3393
3394 set_control(jobj_is_not_null);
3395
3396 // Load the threadObj for the CarrierThread.
3397 Node* threadObj = generate_current_thread(tls_ptr);
3398
3399 // Load the vthread.
3400 Node* vthread = generate_virtual_thread(tls_ptr);
3401
3402 // If vthread != threadObj, this is a virtual thread.
3403 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3404 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3405 IfNode* iff_vthread_not_equal_threadObj =
3406 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3407
3408 // False branch, fallback to threadObj.
3409 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3410 set_control(vthread_equal_threadObj);
3411
3412 // Load the tid field from the vthread object.
3413 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3414
3415 // Load the raw epoch value from the threadObj.
3416 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3417 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3418 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3419 TypeInt::CHAR, T_CHAR,
3420 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3421
3422 // Mask off the excluded information from the epoch.
3423 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3424
3425 // True branch, this is a virtual thread.
3426 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3427 set_control(vthread_not_equal_threadObj);
3428
3429 // Load the tid field from the vthread object.
3430 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3431
3432 // Continuation support determines if a virtual thread should be pinned.
3433 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3434 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3435
3436 // Load the raw epoch value from the vthread.
3437 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3438 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3439 TypeInt::CHAR, T_CHAR,
3440 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3441
3442 // Mask off the excluded information from the epoch.
3443 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3444
3445 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3446 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3447 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3448 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3449
3450 // False branch, vthread is excluded, no need to write epoch info.
3451 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3452
3453 // True branch, vthread is included, update epoch info.
3454 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3455 set_control(included);
3456
3457 // Get epoch value.
3458 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3459
3460 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3461 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3462 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3463
3464 // Compare the epoch in the vthread to the current epoch generation.
3465 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3466 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3467 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3468
3469 // False path, epoch is equal, checkpoint information is valid.
3470 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3471
3472 // True path, epoch is not equal, write a checkpoint for the vthread.
3473 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3474
3475 set_control(epoch_is_not_equal);
3476
3477 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3478 // The call also updates the native thread local thread id and the vthread with the current epoch.
3479 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3480 OptoRuntime::jfr_write_checkpoint_Type(),
3481 SharedRuntime::jfr_write_checkpoint(),
3482 "write_checkpoint", TypePtr::BOTTOM);
3483 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3484
3485 // vthread epoch != current epoch
3486 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3487 record_for_igvn(epoch_compare_rgn);
3488 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3489 record_for_igvn(epoch_compare_mem);
3490 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3491 record_for_igvn(epoch_compare_io);
3492
3493 // Update control and phi nodes.
3494 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3495 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3496 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3497 epoch_compare_mem->init_req(_false_path, input_memory_state);
3498 epoch_compare_io->init_req(_true_path, i_o());
3499 epoch_compare_io->init_req(_false_path, input_io_state);
3500
3501 // excluded != true
3502 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3503 record_for_igvn(exclude_compare_rgn);
3504 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3505 record_for_igvn(exclude_compare_mem);
3506 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3507 record_for_igvn(exclude_compare_io);
3508
3509 // Update control and phi nodes.
3510 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3511 exclude_compare_rgn->init_req(_false_path, excluded);
3512 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3513 exclude_compare_mem->init_req(_false_path, input_memory_state);
3514 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3515 exclude_compare_io->init_req(_false_path, input_io_state);
3516
3517 // vthread != threadObj
3518 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3519 record_for_igvn(vthread_compare_rgn);
3520 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3521 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3522 record_for_igvn(vthread_compare_io);
3523 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3524 record_for_igvn(tid);
3525 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3526 record_for_igvn(exclusion);
3527 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3528 record_for_igvn(pinVirtualThread);
3529
3530 // Update control and phi nodes.
3531 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3532 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3533 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3534 vthread_compare_mem->init_req(_false_path, input_memory_state);
3535 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3536 vthread_compare_io->init_req(_false_path, input_io_state);
3537 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3538 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3539 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3540 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3541 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3542 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3543
3544 // Update branch state.
3545 set_control(_gvn.transform(vthread_compare_rgn));
3546 set_all_memory(_gvn.transform(vthread_compare_mem));
3547 set_i_o(_gvn.transform(vthread_compare_io));
3548
3549 // Load the event writer oop by dereferencing the jobject handle.
3550 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3551 assert(klass_EventWriter->is_loaded(), "invariant");
3552 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3553 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3554 const TypeOopPtr* const xtype = aklass->as_instance_type();
3555 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3556 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3557
3558 // Load the current thread id from the event writer object.
3559 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3560 // Get the field offset to, conditionally, store an updated tid value later.
3561 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3562 // Get the field offset to, conditionally, store an updated exclusion value later.
3563 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3564 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3565 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3566
3567 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3568 record_for_igvn(event_writer_tid_compare_rgn);
3569 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3570 record_for_igvn(event_writer_tid_compare_mem);
3571 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3572 record_for_igvn(event_writer_tid_compare_io);
3573
3574 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3575 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3576 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3577 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3578
3579 // False path, tids are the same.
3580 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3581
3582 // True path, tid is not equal, need to update the tid in the event writer.
3583 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3584 record_for_igvn(tid_is_not_equal);
3585
3586 // Store the pin state to the event writer.
3587 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3588
3589 // Store the exclusion state to the event writer.
3590 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3591 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3592
3593 // Store the tid to the event writer.
3594 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3595
3596 // Update control and phi nodes.
3597 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3598 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3599 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3600 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3601 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3602 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3603
3604 // Result of top level CFG, Memory, IO and Value.
3605 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3606 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3607 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3608 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3609
3610 // Result control.
3611 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3612 result_rgn->init_req(_false_path, jobj_is_null);
3613
3614 // Result memory.
3615 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3616 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3617
3618 // Result IO.
3619 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3620 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3621
3622 // Result value.
3623 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3624 result_value->init_req(_false_path, null()); // return null
3625
3626 // Set output state.
3627 set_control(_gvn.transform(result_rgn));
3628 set_all_memory(_gvn.transform(result_mem));
3629 set_i_o(_gvn.transform(result_io));
3630 set_result(result_rgn, result_value);
3631 return true;
3632 }
3633
3634 /*
3635 * The intrinsic is a model of this pseudo-code:
3636 *
3637 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3638 * if (carrierThread != thread) { // is virtual thread
3639 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3640 * bool excluded = vthread_epoch_raw & excluded_mask;
3641 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3642 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3643 * if (!excluded) {
3644 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3645 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3646 * }
3647 * AtomicAccess::release_store(&tl->_vthread, true);
3648 * return;
3649 * }
3650 * AtomicAccess::release_store(&tl->_vthread, false);
3651 */
3652 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3653 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3654
3655 Node* input_memory_state = reset_memory();
3656 set_all_memory(input_memory_state);
3657
3658 // The most significant bit of the u2 is used to denote thread exclusion
3659 Node* excluded_mask = _gvn.intcon(1 << 15);
3660 // The epoch generation is the range [1-32767]
3661 Node* epoch_mask = _gvn.intcon(32767);
3662
3663 Node* const carrierThread = generate_current_thread(jt);
3664 // If thread != carrierThread, this is a virtual thread.
3665 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3666 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3667 IfNode* iff_thread_not_equal_carrierThread =
3668 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3669
3670 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3671
3672 // False branch, is carrierThread.
3673 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3674 // Store release
3675 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3676
3677 set_all_memory(input_memory_state);
3678
3679 // True branch, is virtual thread.
3680 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3681 set_control(thread_not_equal_carrierThread);
3682
3683 // Load the raw epoch value from the vthread.
3684 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3685 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3686 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3687
3688 // Mask off the excluded information from the epoch.
3689 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3690
3691 // Load the tid field from the thread.
3692 Node* tid = load_field_from_object(thread, "tid", "J");
3693
3694 // Store the vthread tid to the jfr thread local.
3695 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3696 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3697
3698 // Branch is_excluded to conditionalize updating the epoch .
3699 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3700 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3701 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3702
3703 // True branch, vthread is excluded, no need to write epoch info.
3704 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3705 set_control(excluded);
3706 Node* vthread_is_excluded = _gvn.intcon(1);
3707
3708 // False branch, vthread is included, update epoch info.
3709 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3710 set_control(included);
3711 Node* vthread_is_included = _gvn.intcon(0);
3712
3713 // Get epoch value.
3714 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3715
3716 // Store the vthread epoch to the jfr thread local.
3717 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3718 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3719
3720 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3721 record_for_igvn(excluded_rgn);
3722 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3723 record_for_igvn(excluded_mem);
3724 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3725 record_for_igvn(exclusion);
3726
3727 // Merge the excluded control and memory.
3728 excluded_rgn->init_req(_true_path, excluded);
3729 excluded_rgn->init_req(_false_path, included);
3730 excluded_mem->init_req(_true_path, tid_memory);
3731 excluded_mem->init_req(_false_path, included_memory);
3732 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3733 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3734
3735 // Set intermediate state.
3736 set_control(_gvn.transform(excluded_rgn));
3737 set_all_memory(excluded_mem);
3738
3739 // Store the vthread exclusion state to the jfr thread local.
3740 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3741 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3742
3743 // Store release
3744 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3745
3746 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3747 record_for_igvn(thread_compare_rgn);
3748 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3749 record_for_igvn(thread_compare_mem);
3750 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3751 record_for_igvn(vthread);
3752
3753 // Merge the thread_compare control and memory.
3754 thread_compare_rgn->init_req(_true_path, control());
3755 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3756 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3757 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3758
3759 // Set output state.
3760 set_control(_gvn.transform(thread_compare_rgn));
3761 set_all_memory(_gvn.transform(thread_compare_mem));
3762 }
3763
3764 #endif // JFR_HAVE_INTRINSICS
3765
3766 //------------------------inline_native_currentCarrierThread------------------
3767 bool LibraryCallKit::inline_native_currentCarrierThread() {
3768 Node* junk = nullptr;
3769 set_result(generate_current_thread(junk));
3770 return true;
3771 }
3772
3773 //------------------------inline_native_currentThread------------------
3774 bool LibraryCallKit::inline_native_currentThread() {
3775 Node* junk = nullptr;
3776 set_result(generate_virtual_thread(junk));
3777 return true;
3778 }
3779
3780 //------------------------inline_native_setVthread------------------
3781 bool LibraryCallKit::inline_native_setCurrentThread() {
3782 assert(C->method()->changes_current_thread(),
3783 "method changes current Thread but is not annotated ChangesCurrentThread");
3784 Node* arr = argument(1);
3785 Node* thread = _gvn.transform(new ThreadLocalNode());
3786 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3787 Node* thread_obj_handle
3788 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3789 thread_obj_handle = _gvn.transform(thread_obj_handle);
3790 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3791 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3792
3793 // Change the _monitor_owner_id of the JavaThread
3794 Node* tid = load_field_from_object(arr, "tid", "J");
3795 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3796 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3797
3798 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3799 return true;
3800 }
3801
3802 const Type* LibraryCallKit::scopedValueCache_type() {
3803 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3804 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3805 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3806
3807 // Because we create the scopedValue cache lazily we have to make the
3808 // type of the result BotPTR.
3809 bool xk = etype->klass_is_exact();
3810 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3811 return objects_type;
3812 }
3813
3814 Node* LibraryCallKit::scopedValueCache_helper() {
3815 Node* thread = _gvn.transform(new ThreadLocalNode());
3816 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3817 // We cannot use immutable_memory() because we might flip onto a
3818 // different carrier thread, at which point we'll need to use that
3819 // carrier thread's cache.
3820 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3821 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3822 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3823 }
3824
3825 //------------------------inline_native_scopedValueCache------------------
3826 bool LibraryCallKit::inline_native_scopedValueCache() {
3827 Node* cache_obj_handle = scopedValueCache_helper();
3828 const Type* objects_type = scopedValueCache_type();
3829 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3830
3831 return true;
3832 }
3833
3834 //------------------------inline_native_setScopedValueCache------------------
3835 bool LibraryCallKit::inline_native_setScopedValueCache() {
3836 Node* arr = argument(0);
3837 Node* cache_obj_handle = scopedValueCache_helper();
3838 const Type* objects_type = scopedValueCache_type();
3839
3840 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3841 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3842
3843 return true;
3844 }
3845
3846 //------------------------inline_native_Continuation_pin and unpin-----------
3847
3848 // Shared implementation routine for both pin and unpin.
3849 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3850 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3851
3852 // Save input memory.
3853 Node* input_memory_state = reset_memory();
3854 set_all_memory(input_memory_state);
3855
3856 // TLS
3857 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3858 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3859 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3860
3861 // Null check the last continuation object.
3862 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3863 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3864 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3865
3866 // False path, last continuation is null.
3867 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3868
3869 // True path, last continuation is not null.
3870 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3871
3872 set_control(continuation_is_not_null);
3873
3874 // Load the pin count from the last continuation.
3875 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3876 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3877
3878 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3879 Node* pin_count_rhs;
3880 if (unpin) {
3881 pin_count_rhs = _gvn.intcon(0);
3882 } else {
3883 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3884 }
3885 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3886 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3887 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3888
3889 // True branch, pin count over/underflow.
3890 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3891 {
3892 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3893 // which will throw IllegalStateException for pin count over/underflow.
3894 // No memory changed so far - we can use memory create by reset_memory()
3895 // at the beginning of this intrinsic. No need to call reset_memory() again.
3896 PreserveJVMState pjvms(this);
3897 set_control(pin_count_over_underflow);
3898 uncommon_trap(Deoptimization::Reason_intrinsic,
3899 Deoptimization::Action_none);
3900 assert(stopped(), "invariant");
3901 }
3902
3903 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3904 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3905 set_control(valid_pin_count);
3906
3907 Node* next_pin_count;
3908 if (unpin) {
3909 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3910 } else {
3911 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3912 }
3913
3914 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3915
3916 // Result of top level CFG and Memory.
3917 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3918 record_for_igvn(result_rgn);
3919 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3920 record_for_igvn(result_mem);
3921
3922 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3923 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3924 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3925 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3926
3927 // Set output state.
3928 set_control(_gvn.transform(result_rgn));
3929 set_all_memory(_gvn.transform(result_mem));
3930
3931 return true;
3932 }
3933
3934 //---------------------------load_mirror_from_klass----------------------------
3935 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3936 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3937 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3938 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3939 // mirror = ((OopHandle)mirror)->resolve();
3940 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3941 }
3942
3943 //-----------------------load_klass_from_mirror_common-------------------------
3944 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3945 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3946 // and branch to the given path on the region.
3947 // If never_see_null, take an uncommon trap on null, so we can optimistically
3948 // compile for the non-null case.
3949 // If the region is null, force never_see_null = true.
3950 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3951 bool never_see_null,
3952 RegionNode* region,
3953 int null_path,
3954 int offset) {
3955 if (region == nullptr) never_see_null = true;
3956 Node* p = basic_plus_adr(mirror, offset);
3957 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3958 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3959 Node* null_ctl = top();
3960 kls = null_check_oop(kls, &null_ctl, never_see_null);
3961 if (region != nullptr) {
3962 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3963 region->init_req(null_path, null_ctl);
3964 } else {
3965 assert(null_ctl == top(), "no loose ends");
3966 }
3967 return kls;
3968 }
3969
3970 //--------------------(inline_native_Class_query helpers)---------------------
3971 // Use this for JVM_ACC_INTERFACE.
3972 // Fall through if (mods & mask) == bits, take the guard otherwise.
3973 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3974 ByteSize offset, const Type* type, BasicType bt) {
3975 // Branch around if the given klass has the given modifier bit set.
3976 // Like generate_guard, adds a new path onto the region.
3977 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3978 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3979 Node* mask = intcon(modifier_mask);
3980 Node* bits = intcon(modifier_bits);
3981 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3982 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3983 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3984 return generate_fair_guard(bol, region);
3985 }
3986 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3987 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3988 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3989 }
3990
3991 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3992 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3993 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3994 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3995 }
3996
3997 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3998 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3999 }
4000
4001 //-------------------------inline_native_Class_query-------------------
4002 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4003 const Type* return_type = TypeInt::BOOL;
4004 Node* prim_return_value = top(); // what happens if it's a primitive class?
4005 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4006 bool expect_prim = false; // most of these guys expect to work on refs
4007
4008 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4009
4010 Node* mirror = argument(0);
4011 Node* obj = top();
4012
4013 switch (id) {
4014 case vmIntrinsics::_isInstance:
4015 // nothing is an instance of a primitive type
4016 prim_return_value = intcon(0);
4017 obj = argument(1);
4018 break;
4019 case vmIntrinsics::_isHidden:
4020 prim_return_value = intcon(0);
4021 break;
4022 case vmIntrinsics::_getSuperclass:
4023 prim_return_value = null();
4024 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4025 break;
4026 default:
4027 fatal_unexpected_iid(id);
4028 break;
4029 }
4030
4031 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4032 if (mirror_con == nullptr) return false; // cannot happen?
4033
4034 #ifndef PRODUCT
4035 if (C->print_intrinsics() || C->print_inlining()) {
4036 ciType* k = mirror_con->java_mirror_type();
4037 if (k) {
4038 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4039 k->print_name();
4040 tty->cr();
4041 }
4042 }
4043 #endif
4044
4045 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4046 RegionNode* region = new RegionNode(PATH_LIMIT);
4047 record_for_igvn(region);
4048 PhiNode* phi = new PhiNode(region, return_type);
4049
4050 // The mirror will never be null of Reflection.getClassAccessFlags, however
4051 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4052 // if it is. See bug 4774291.
4053
4054 // For Reflection.getClassAccessFlags(), the null check occurs in
4055 // the wrong place; see inline_unsafe_access(), above, for a similar
4056 // situation.
4057 mirror = null_check(mirror);
4058 // If mirror or obj is dead, only null-path is taken.
4059 if (stopped()) return true;
4060
4061 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4062
4063 // Now load the mirror's klass metaobject, and null-check it.
4064 // Side-effects region with the control path if the klass is null.
4065 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4066 // If kls is null, we have a primitive mirror.
4067 phi->init_req(_prim_path, prim_return_value);
4068 if (stopped()) { set_result(region, phi); return true; }
4069 bool safe_for_replace = (region->in(_prim_path) == top());
4070
4071 Node* p; // handy temp
4072 Node* null_ctl;
4073
4074 // Now that we have the non-null klass, we can perform the real query.
4075 // For constant classes, the query will constant-fold in LoadNode::Value.
4076 Node* query_value = top();
4077 switch (id) {
4078 case vmIntrinsics::_isInstance:
4079 // nothing is an instance of a primitive type
4080 query_value = gen_instanceof(obj, kls, safe_for_replace);
4081 break;
4082
4083 case vmIntrinsics::_isHidden:
4084 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4085 if (generate_hidden_class_guard(kls, region) != nullptr)
4086 // A guard was added. If the guard is taken, it was an hidden class.
4087 phi->add_req(intcon(1));
4088 // If we fall through, it's a plain class.
4089 query_value = intcon(0);
4090 break;
4091
4092
4093 case vmIntrinsics::_getSuperclass:
4094 // The rules here are somewhat unfortunate, but we can still do better
4095 // with random logic than with a JNI call.
4096 // Interfaces store null or Object as _super, but must report null.
4097 // Arrays store an intermediate super as _super, but must report Object.
4098 // Other types can report the actual _super.
4099 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4100 if (generate_interface_guard(kls, region) != nullptr)
4101 // A guard was added. If the guard is taken, it was an interface.
4102 phi->add_req(null());
4103 if (generate_array_guard(kls, region) != nullptr)
4104 // A guard was added. If the guard is taken, it was an array.
4105 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4106 // If we fall through, it's a plain class. Get its _super.
4107 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4108 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4109 null_ctl = top();
4110 kls = null_check_oop(kls, &null_ctl);
4111 if (null_ctl != top()) {
4112 // If the guard is taken, Object.superClass is null (both klass and mirror).
4113 region->add_req(null_ctl);
4114 phi ->add_req(null());
4115 }
4116 if (!stopped()) {
4117 query_value = load_mirror_from_klass(kls);
4118 }
4119 break;
4120
4121 default:
4122 fatal_unexpected_iid(id);
4123 break;
4124 }
4125
4126 // Fall-through is the normal case of a query to a real class.
4127 phi->init_req(1, query_value);
4128 region->init_req(1, control());
4129
4130 C->set_has_split_ifs(true); // Has chance for split-if optimization
4131 set_result(region, phi);
4132 return true;
4133 }
4134
4135 //-------------------------inline_Class_cast-------------------
4136 bool LibraryCallKit::inline_Class_cast() {
4137 Node* mirror = argument(0); // Class
4138 Node* obj = argument(1);
4139 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4140 if (mirror_con == nullptr) {
4141 return false; // dead path (mirror->is_top()).
4142 }
4143 if (obj == nullptr || obj->is_top()) {
4144 return false; // dead path
4145 }
4146 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4147
4148 // First, see if Class.cast() can be folded statically.
4149 // java_mirror_type() returns non-null for compile-time Class constants.
4150 ciType* tm = mirror_con->java_mirror_type();
4151 if (tm != nullptr && tm->is_klass() &&
4152 tp != nullptr) {
4153 if (!tp->is_loaded()) {
4154 // Don't use intrinsic when class is not loaded.
4155 return false;
4156 } else {
4157 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4158 if (static_res == Compile::SSC_always_true) {
4159 // isInstance() is true - fold the code.
4160 set_result(obj);
4161 return true;
4162 } else if (static_res == Compile::SSC_always_false) {
4163 // Don't use intrinsic, have to throw ClassCastException.
4164 // If the reference is null, the non-intrinsic bytecode will
4165 // be optimized appropriately.
4166 return false;
4167 }
4168 }
4169 }
4170
4171 // Bailout intrinsic and do normal inlining if exception path is frequent.
4172 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4173 return false;
4174 }
4175
4176 // Generate dynamic checks.
4177 // Class.cast() is java implementation of _checkcast bytecode.
4178 // Do checkcast (Parse::do_checkcast()) optimizations here.
4179
4180 mirror = null_check(mirror);
4181 // If mirror is dead, only null-path is taken.
4182 if (stopped()) {
4183 return true;
4184 }
4185
4186 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4187 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4188 RegionNode* region = new RegionNode(PATH_LIMIT);
4189 record_for_igvn(region);
4190
4191 // Now load the mirror's klass metaobject, and null-check it.
4192 // If kls is null, we have a primitive mirror and
4193 // nothing is an instance of a primitive type.
4194 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4195
4196 Node* res = top();
4197 if (!stopped()) {
4198 Node* bad_type_ctrl = top();
4199 // Do checkcast optimizations.
4200 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4201 region->init_req(_bad_type_path, bad_type_ctrl);
4202 }
4203 if (region->in(_prim_path) != top() ||
4204 region->in(_bad_type_path) != top()) {
4205 // Let Interpreter throw ClassCastException.
4206 PreserveJVMState pjvms(this);
4207 set_control(_gvn.transform(region));
4208 uncommon_trap(Deoptimization::Reason_intrinsic,
4209 Deoptimization::Action_maybe_recompile);
4210 }
4211 if (!stopped()) {
4212 set_result(res);
4213 }
4214 return true;
4215 }
4216
4217
4218 //--------------------------inline_native_subtype_check------------------------
4219 // This intrinsic takes the JNI calls out of the heart of
4220 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4221 bool LibraryCallKit::inline_native_subtype_check() {
4222 // Pull both arguments off the stack.
4223 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4224 args[0] = argument(0);
4225 args[1] = argument(1);
4226 Node* klasses[2]; // corresponding Klasses: superk, subk
4227 klasses[0] = klasses[1] = top();
4228
4229 enum {
4230 // A full decision tree on {superc is prim, subc is prim}:
4231 _prim_0_path = 1, // {P,N} => false
4232 // {P,P} & superc!=subc => false
4233 _prim_same_path, // {P,P} & superc==subc => true
4234 _prim_1_path, // {N,P} => false
4235 _ref_subtype_path, // {N,N} & subtype check wins => true
4236 _both_ref_path, // {N,N} & subtype check loses => false
4237 PATH_LIMIT
4238 };
4239
4240 RegionNode* region = new RegionNode(PATH_LIMIT);
4241 Node* phi = new PhiNode(region, TypeInt::BOOL);
4242 record_for_igvn(region);
4243
4244 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4245 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4246 int class_klass_offset = java_lang_Class::klass_offset();
4247
4248 // First null-check both mirrors and load each mirror's klass metaobject.
4249 int which_arg;
4250 for (which_arg = 0; which_arg <= 1; which_arg++) {
4251 Node* arg = args[which_arg];
4252 arg = null_check(arg);
4253 if (stopped()) break;
4254 args[which_arg] = arg;
4255
4256 Node* p = basic_plus_adr(arg, class_klass_offset);
4257 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4258 klasses[which_arg] = _gvn.transform(kls);
4259 }
4260
4261 // Having loaded both klasses, test each for null.
4262 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4263 for (which_arg = 0; which_arg <= 1; which_arg++) {
4264 Node* kls = klasses[which_arg];
4265 Node* null_ctl = top();
4266 kls = null_check_oop(kls, &null_ctl, never_see_null);
4267 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4268 region->init_req(prim_path, null_ctl);
4269 if (stopped()) break;
4270 klasses[which_arg] = kls;
4271 }
4272
4273 if (!stopped()) {
4274 // now we have two reference types, in klasses[0..1]
4275 Node* subk = klasses[1]; // the argument to isAssignableFrom
4276 Node* superk = klasses[0]; // the receiver
4277 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4278 // now we have a successful reference subtype check
4279 region->set_req(_ref_subtype_path, control());
4280 }
4281
4282 // If both operands are primitive (both klasses null), then
4283 // we must return true when they are identical primitives.
4284 // It is convenient to test this after the first null klass check.
4285 set_control(region->in(_prim_0_path)); // go back to first null check
4286 if (!stopped()) {
4287 // Since superc is primitive, make a guard for the superc==subc case.
4288 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4289 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4290 generate_guard(bol_eq, region, PROB_FAIR);
4291 if (region->req() == PATH_LIMIT+1) {
4292 // A guard was added. If the added guard is taken, superc==subc.
4293 region->swap_edges(PATH_LIMIT, _prim_same_path);
4294 region->del_req(PATH_LIMIT);
4295 }
4296 region->set_req(_prim_0_path, control()); // Not equal after all.
4297 }
4298
4299 // these are the only paths that produce 'true':
4300 phi->set_req(_prim_same_path, intcon(1));
4301 phi->set_req(_ref_subtype_path, intcon(1));
4302
4303 // pull together the cases:
4304 assert(region->req() == PATH_LIMIT, "sane region");
4305 for (uint i = 1; i < region->req(); i++) {
4306 Node* ctl = region->in(i);
4307 if (ctl == nullptr || ctl == top()) {
4308 region->set_req(i, top());
4309 phi ->set_req(i, top());
4310 } else if (phi->in(i) == nullptr) {
4311 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4312 }
4313 }
4314
4315 set_control(_gvn.transform(region));
4316 set_result(_gvn.transform(phi));
4317 return true;
4318 }
4319
4320 //---------------------generate_array_guard_common------------------------
4321 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4322 bool obj_array, bool not_array, Node** obj) {
4323
4324 if (stopped()) {
4325 return nullptr;
4326 }
4327
4328 // If obj_array/non_array==false/false:
4329 // Branch around if the given klass is in fact an array (either obj or prim).
4330 // If obj_array/non_array==false/true:
4331 // Branch around if the given klass is not an array klass of any kind.
4332 // If obj_array/non_array==true/true:
4333 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4334 // If obj_array/non_array==true/false:
4335 // Branch around if the kls is an oop array (Object[] or subtype)
4336 //
4337 // Like generate_guard, adds a new path onto the region.
4338 jint layout_con = 0;
4339 Node* layout_val = get_layout_helper(kls, layout_con);
4340 if (layout_val == nullptr) {
4341 bool query = (obj_array
4342 ? Klass::layout_helper_is_objArray(layout_con)
4343 : Klass::layout_helper_is_array(layout_con));
4344 if (query == not_array) {
4345 return nullptr; // never a branch
4346 } else { // always a branch
4347 Node* always_branch = control();
4348 if (region != nullptr)
4349 region->add_req(always_branch);
4350 set_control(top());
4351 return always_branch;
4352 }
4353 }
4354 // Now test the correct condition.
4355 jint nval = (obj_array
4356 ? (jint)(Klass::_lh_array_tag_type_value
4357 << Klass::_lh_array_tag_shift)
4358 : Klass::_lh_neutral_value);
4359 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4360 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4361 // invert the test if we are looking for a non-array
4362 if (not_array) btest = BoolTest(btest).negate();
4363 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4364 Node* ctrl = generate_fair_guard(bol, region);
4365 Node* is_array_ctrl = not_array ? control() : ctrl;
4366 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4367 // Keep track of the fact that 'obj' is an array to prevent
4368 // array specific accesses from floating above the guard.
4369 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4370 }
4371 return ctrl;
4372 }
4373
4374
4375 //-----------------------inline_native_newArray--------------------------
4376 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4377 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4378 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4379 Node* mirror;
4380 Node* count_val;
4381 if (uninitialized) {
4382 null_check_receiver();
4383 mirror = argument(1);
4384 count_val = argument(2);
4385 } else {
4386 mirror = argument(0);
4387 count_val = argument(1);
4388 }
4389
4390 mirror = null_check(mirror);
4391 // If mirror or obj is dead, only null-path is taken.
4392 if (stopped()) return true;
4393
4394 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4395 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4396 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4397 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4398 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4399
4400 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4401 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4402 result_reg, _slow_path);
4403 Node* normal_ctl = control();
4404 Node* no_array_ctl = result_reg->in(_slow_path);
4405
4406 // Generate code for the slow case. We make a call to newArray().
4407 set_control(no_array_ctl);
4408 if (!stopped()) {
4409 // Either the input type is void.class, or else the
4410 // array klass has not yet been cached. Either the
4411 // ensuing call will throw an exception, or else it
4412 // will cache the array klass for next time.
4413 PreserveJVMState pjvms(this);
4414 CallJavaNode* slow_call = nullptr;
4415 if (uninitialized) {
4416 // Generate optimized virtual call (holder class 'Unsafe' is final)
4417 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4418 } else {
4419 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4420 }
4421 Node* slow_result = set_results_for_java_call(slow_call);
4422 // this->control() comes from set_results_for_java_call
4423 result_reg->set_req(_slow_path, control());
4424 result_val->set_req(_slow_path, slow_result);
4425 result_io ->set_req(_slow_path, i_o());
4426 result_mem->set_req(_slow_path, reset_memory());
4427 }
4428
4429 set_control(normal_ctl);
4430 if (!stopped()) {
4431 // Normal case: The array type has been cached in the java.lang.Class.
4432 // The following call works fine even if the array type is polymorphic.
4433 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4434 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4435 result_reg->init_req(_normal_path, control());
4436 result_val->init_req(_normal_path, obj);
4437 result_io ->init_req(_normal_path, i_o());
4438 result_mem->init_req(_normal_path, reset_memory());
4439
4440 if (uninitialized) {
4441 // Mark the allocation so that zeroing is skipped
4442 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4443 alloc->maybe_set_complete(&_gvn);
4444 }
4445 }
4446
4447 // Return the combined state.
4448 set_i_o( _gvn.transform(result_io) );
4449 set_all_memory( _gvn.transform(result_mem));
4450
4451 C->set_has_split_ifs(true); // Has chance for split-if optimization
4452 set_result(result_reg, result_val);
4453 return true;
4454 }
4455
4456 //----------------------inline_native_getLength--------------------------
4457 // public static native int java.lang.reflect.Array.getLength(Object array);
4458 bool LibraryCallKit::inline_native_getLength() {
4459 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4460
4461 Node* array = null_check(argument(0));
4462 // If array is dead, only null-path is taken.
4463 if (stopped()) return true;
4464
4465 // Deoptimize if it is a non-array.
4466 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4467
4468 if (non_array != nullptr) {
4469 PreserveJVMState pjvms(this);
4470 set_control(non_array);
4471 uncommon_trap(Deoptimization::Reason_intrinsic,
4472 Deoptimization::Action_maybe_recompile);
4473 }
4474
4475 // If control is dead, only non-array-path is taken.
4476 if (stopped()) return true;
4477
4478 // The works fine even if the array type is polymorphic.
4479 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4480 Node* result = load_array_length(array);
4481
4482 C->set_has_split_ifs(true); // Has chance for split-if optimization
4483 set_result(result);
4484 return true;
4485 }
4486
4487 //------------------------inline_array_copyOf----------------------------
4488 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4489 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4490 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4491 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4492
4493 // Get the arguments.
4494 Node* original = argument(0);
4495 Node* start = is_copyOfRange? argument(1): intcon(0);
4496 Node* end = is_copyOfRange? argument(2): argument(1);
4497 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4498
4499 Node* newcopy = nullptr;
4500
4501 // Set the original stack and the reexecute bit for the interpreter to reexecute
4502 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4503 { PreserveReexecuteState preexecs(this);
4504 jvms()->set_should_reexecute(true);
4505
4506 array_type_mirror = null_check(array_type_mirror);
4507 original = null_check(original);
4508
4509 // Check if a null path was taken unconditionally.
4510 if (stopped()) return true;
4511
4512 Node* orig_length = load_array_length(original);
4513
4514 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4515 klass_node = null_check(klass_node);
4516
4517 RegionNode* bailout = new RegionNode(1);
4518 record_for_igvn(bailout);
4519
4520 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4521 // Bail out if that is so.
4522 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4523 if (not_objArray != nullptr) {
4524 // Improve the klass node's type from the new optimistic assumption:
4525 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4526 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4527 Node* cast = new CastPPNode(control(), klass_node, akls);
4528 klass_node = _gvn.transform(cast);
4529 }
4530
4531 // Bail out if either start or end is negative.
4532 generate_negative_guard(start, bailout, &start);
4533 generate_negative_guard(end, bailout, &end);
4534
4535 Node* length = end;
4536 if (_gvn.type(start) != TypeInt::ZERO) {
4537 length = _gvn.transform(new SubINode(end, start));
4538 }
4539
4540 // Bail out if length is negative (i.e., if start > end).
4541 // Without this the new_array would throw
4542 // NegativeArraySizeException but IllegalArgumentException is what
4543 // should be thrown
4544 generate_negative_guard(length, bailout, &length);
4545
4546 // Bail out if start is larger than the original length
4547 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4548 generate_negative_guard(orig_tail, bailout, &orig_tail);
4549
4550 if (bailout->req() > 1) {
4551 PreserveJVMState pjvms(this);
4552 set_control(_gvn.transform(bailout));
4553 uncommon_trap(Deoptimization::Reason_intrinsic,
4554 Deoptimization::Action_maybe_recompile);
4555 }
4556
4557 if (!stopped()) {
4558 // How many elements will we copy from the original?
4559 // The answer is MinI(orig_tail, length).
4560 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4561
4562 // Generate a direct call to the right arraycopy function(s).
4563 // We know the copy is disjoint but we might not know if the
4564 // oop stores need checking.
4565 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4566 // This will fail a store-check if x contains any non-nulls.
4567
4568 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4569 // loads/stores but it is legal only if we're sure the
4570 // Arrays.copyOf would succeed. So we need all input arguments
4571 // to the copyOf to be validated, including that the copy to the
4572 // new array won't trigger an ArrayStoreException. That subtype
4573 // check can be optimized if we know something on the type of
4574 // the input array from type speculation.
4575 if (_gvn.type(klass_node)->singleton()) {
4576 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4577 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4578
4579 int test = C->static_subtype_check(superk, subk);
4580 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4581 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4582 if (t_original->speculative_type() != nullptr) {
4583 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4584 }
4585 }
4586 }
4587
4588 bool validated = false;
4589 // Reason_class_check rather than Reason_intrinsic because we
4590 // want to intrinsify even if this traps.
4591 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4592 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4593
4594 if (not_subtype_ctrl != top()) {
4595 PreserveJVMState pjvms(this);
4596 set_control(not_subtype_ctrl);
4597 uncommon_trap(Deoptimization::Reason_class_check,
4598 Deoptimization::Action_make_not_entrant);
4599 assert(stopped(), "Should be stopped");
4600 }
4601 validated = true;
4602 }
4603
4604 if (!stopped()) {
4605 newcopy = new_array(klass_node, length, 0); // no arguments to push
4606
4607 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4608 load_object_klass(original), klass_node);
4609 if (!is_copyOfRange) {
4610 ac->set_copyof(validated);
4611 } else {
4612 ac->set_copyofrange(validated);
4613 }
4614 Node* n = _gvn.transform(ac);
4615 if (n == ac) {
4616 ac->connect_outputs(this);
4617 } else {
4618 assert(validated, "shouldn't transform if all arguments not validated");
4619 set_all_memory(n);
4620 }
4621 }
4622 }
4623 } // original reexecute is set back here
4624
4625 C->set_has_split_ifs(true); // Has chance for split-if optimization
4626 if (!stopped()) {
4627 set_result(newcopy);
4628 }
4629 return true;
4630 }
4631
4632
4633 //----------------------generate_virtual_guard---------------------------
4634 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4635 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4636 RegionNode* slow_region) {
4637 ciMethod* method = callee();
4638 int vtable_index = method->vtable_index();
4639 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4640 "bad index %d", vtable_index);
4641 // Get the Method* out of the appropriate vtable entry.
4642 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4643 vtable_index*vtableEntry::size_in_bytes() +
4644 in_bytes(vtableEntry::method_offset());
4645 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4646 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4647
4648 // Compare the target method with the expected method (e.g., Object.hashCode).
4649 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4650
4651 Node* native_call = makecon(native_call_addr);
4652 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4653 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4654
4655 return generate_slow_guard(test_native, slow_region);
4656 }
4657
4658 //-----------------------generate_method_call----------------------------
4659 // Use generate_method_call to make a slow-call to the real
4660 // method if the fast path fails. An alternative would be to
4661 // use a stub like OptoRuntime::slow_arraycopy_Java.
4662 // This only works for expanding the current library call,
4663 // not another intrinsic. (E.g., don't use this for making an
4664 // arraycopy call inside of the copyOf intrinsic.)
4665 CallJavaNode*
4666 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4667 // When compiling the intrinsic method itself, do not use this technique.
4668 guarantee(callee() != C->method(), "cannot make slow-call to self");
4669
4670 ciMethod* method = callee();
4671 // ensure the JVMS we have will be correct for this call
4672 guarantee(method_id == method->intrinsic_id(), "must match");
4673
4674 const TypeFunc* tf = TypeFunc::make(method);
4675 if (res_not_null) {
4676 assert(tf->return_type() == T_OBJECT, "");
4677 const TypeTuple* range = tf->range();
4678 const Type** fields = TypeTuple::fields(range->cnt());
4679 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4680 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4681 tf = TypeFunc::make(tf->domain(), new_range);
4682 }
4683 CallJavaNode* slow_call;
4684 if (is_static) {
4685 assert(!is_virtual, "");
4686 slow_call = new CallStaticJavaNode(C, tf,
4687 SharedRuntime::get_resolve_static_call_stub(), method);
4688 } else if (is_virtual) {
4689 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4690 int vtable_index = Method::invalid_vtable_index;
4691 if (UseInlineCaches) {
4692 // Suppress the vtable call
4693 } else {
4694 // hashCode and clone are not a miranda methods,
4695 // so the vtable index is fixed.
4696 // No need to use the linkResolver to get it.
4697 vtable_index = method->vtable_index();
4698 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4699 "bad index %d", vtable_index);
4700 }
4701 slow_call = new CallDynamicJavaNode(tf,
4702 SharedRuntime::get_resolve_virtual_call_stub(),
4703 method, vtable_index);
4704 } else { // neither virtual nor static: opt_virtual
4705 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4706 slow_call = new CallStaticJavaNode(C, tf,
4707 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4708 slow_call->set_optimized_virtual(true);
4709 }
4710 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4711 // To be able to issue a direct call (optimized virtual or virtual)
4712 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4713 // about the method being invoked should be attached to the call site to
4714 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4715 slow_call->set_override_symbolic_info(true);
4716 }
4717 set_arguments_for_java_call(slow_call);
4718 set_edges_for_java_call(slow_call);
4719 return slow_call;
4720 }
4721
4722
4723 /**
4724 * Build special case code for calls to hashCode on an object. This call may
4725 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4726 * slightly different code.
4727 */
4728 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4729 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4730 assert(!(is_virtual && is_static), "either virtual, special, or static");
4731
4732 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4733
4734 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4735 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4736 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4737 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4738 Node* obj = nullptr;
4739 if (!is_static) {
4740 // Check for hashing null object
4741 obj = null_check_receiver();
4742 if (stopped()) return true; // unconditionally null
4743 result_reg->init_req(_null_path, top());
4744 result_val->init_req(_null_path, top());
4745 } else {
4746 // Do a null check, and return zero if null.
4747 // System.identityHashCode(null) == 0
4748 obj = argument(0);
4749 Node* null_ctl = top();
4750 obj = null_check_oop(obj, &null_ctl);
4751 result_reg->init_req(_null_path, null_ctl);
4752 result_val->init_req(_null_path, _gvn.intcon(0));
4753 }
4754
4755 // Unconditionally null? Then return right away.
4756 if (stopped()) {
4757 set_control( result_reg->in(_null_path));
4758 if (!stopped())
4759 set_result(result_val->in(_null_path));
4760 return true;
4761 }
4762
4763 // We only go to the fast case code if we pass a number of guards. The
4764 // paths which do not pass are accumulated in the slow_region.
4765 RegionNode* slow_region = new RegionNode(1);
4766 record_for_igvn(slow_region);
4767
4768 // If this is a virtual call, we generate a funny guard. We pull out
4769 // the vtable entry corresponding to hashCode() from the target object.
4770 // If the target method which we are calling happens to be the native
4771 // Object hashCode() method, we pass the guard. We do not need this
4772 // guard for non-virtual calls -- the caller is known to be the native
4773 // Object hashCode().
4774 if (is_virtual) {
4775 // After null check, get the object's klass.
4776 Node* obj_klass = load_object_klass(obj);
4777 generate_virtual_guard(obj_klass, slow_region);
4778 }
4779
4780 // Get the header out of the object, use LoadMarkNode when available
4781 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4782 // The control of the load must be null. Otherwise, the load can move before
4783 // the null check after castPP removal.
4784 Node* no_ctrl = nullptr;
4785 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4786
4787 if (!UseObjectMonitorTable) {
4788 // Test the header to see if it is safe to read w.r.t. locking.
4789 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4790 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4791 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4792 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4793 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4794
4795 generate_slow_guard(test_monitor, slow_region);
4796 }
4797
4798 // Get the hash value and check to see that it has been properly assigned.
4799 // We depend on hash_mask being at most 32 bits and avoid the use of
4800 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4801 // vm: see markWord.hpp.
4802 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4803 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4804 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4805 // This hack lets the hash bits live anywhere in the mark object now, as long
4806 // as the shift drops the relevant bits into the low 32 bits. Note that
4807 // Java spec says that HashCode is an int so there's no point in capturing
4808 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4809 hshifted_header = ConvX2I(hshifted_header);
4810 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4811
4812 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4813 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4814 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4815
4816 generate_slow_guard(test_assigned, slow_region);
4817
4818 Node* init_mem = reset_memory();
4819 // fill in the rest of the null path:
4820 result_io ->init_req(_null_path, i_o());
4821 result_mem->init_req(_null_path, init_mem);
4822
4823 result_val->init_req(_fast_path, hash_val);
4824 result_reg->init_req(_fast_path, control());
4825 result_io ->init_req(_fast_path, i_o());
4826 result_mem->init_req(_fast_path, init_mem);
4827
4828 // Generate code for the slow case. We make a call to hashCode().
4829 set_control(_gvn.transform(slow_region));
4830 if (!stopped()) {
4831 // No need for PreserveJVMState, because we're using up the present state.
4832 set_all_memory(init_mem);
4833 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4834 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4835 Node* slow_result = set_results_for_java_call(slow_call);
4836 // this->control() comes from set_results_for_java_call
4837 result_reg->init_req(_slow_path, control());
4838 result_val->init_req(_slow_path, slow_result);
4839 result_io ->set_req(_slow_path, i_o());
4840 result_mem ->set_req(_slow_path, reset_memory());
4841 }
4842
4843 // Return the combined state.
4844 set_i_o( _gvn.transform(result_io) );
4845 set_all_memory( _gvn.transform(result_mem));
4846
4847 set_result(result_reg, result_val);
4848 return true;
4849 }
4850
4851 //---------------------------inline_native_getClass----------------------------
4852 // public final native Class<?> java.lang.Object.getClass();
4853 //
4854 // Build special case code for calls to getClass on an object.
4855 bool LibraryCallKit::inline_native_getClass() {
4856 Node* obj = null_check_receiver();
4857 if (stopped()) return true;
4858 set_result(load_mirror_from_klass(load_object_klass(obj)));
4859 return true;
4860 }
4861
4862 //-----------------inline_native_Reflection_getCallerClass---------------------
4863 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4864 //
4865 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4866 //
4867 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4868 // in that it must skip particular security frames and checks for
4869 // caller sensitive methods.
4870 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4871 #ifndef PRODUCT
4872 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4873 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4874 }
4875 #endif
4876
4877 if (!jvms()->has_method()) {
4878 #ifndef PRODUCT
4879 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4880 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4881 }
4882 #endif
4883 return false;
4884 }
4885
4886 // Walk back up the JVM state to find the caller at the required
4887 // depth.
4888 JVMState* caller_jvms = jvms();
4889
4890 // Cf. JVM_GetCallerClass
4891 // NOTE: Start the loop at depth 1 because the current JVM state does
4892 // not include the Reflection.getCallerClass() frame.
4893 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4894 ciMethod* m = caller_jvms->method();
4895 switch (n) {
4896 case 0:
4897 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4898 break;
4899 case 1:
4900 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4901 if (!m->caller_sensitive()) {
4902 #ifndef PRODUCT
4903 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4904 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4905 }
4906 #endif
4907 return false; // bail-out; let JVM_GetCallerClass do the work
4908 }
4909 break;
4910 default:
4911 if (!m->is_ignored_by_security_stack_walk()) {
4912 // We have reached the desired frame; return the holder class.
4913 // Acquire method holder as java.lang.Class and push as constant.
4914 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4915 ciInstance* caller_mirror = caller_klass->java_mirror();
4916 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4917
4918 #ifndef PRODUCT
4919 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4920 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());
4921 tty->print_cr(" JVM state at this point:");
4922 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4923 ciMethod* m = jvms()->of_depth(i)->method();
4924 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4925 }
4926 }
4927 #endif
4928 return true;
4929 }
4930 break;
4931 }
4932 }
4933
4934 #ifndef PRODUCT
4935 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4936 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4937 tty->print_cr(" JVM state at this point:");
4938 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4939 ciMethod* m = jvms()->of_depth(i)->method();
4940 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4941 }
4942 }
4943 #endif
4944
4945 return false; // bail-out; let JVM_GetCallerClass do the work
4946 }
4947
4948 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4949 Node* arg = argument(0);
4950 Node* result = nullptr;
4951
4952 switch (id) {
4953 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4954 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4955 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4956 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4957 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4958 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4959
4960 case vmIntrinsics::_doubleToLongBits: {
4961 // two paths (plus control) merge in a wood
4962 RegionNode *r = new RegionNode(3);
4963 Node *phi = new PhiNode(r, TypeLong::LONG);
4964
4965 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4966 // Build the boolean node
4967 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4968
4969 // Branch either way.
4970 // NaN case is less traveled, which makes all the difference.
4971 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4972 Node *opt_isnan = _gvn.transform(ifisnan);
4973 assert( opt_isnan->is_If(), "Expect an IfNode");
4974 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4975 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4976
4977 set_control(iftrue);
4978
4979 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4980 Node *slow_result = longcon(nan_bits); // return NaN
4981 phi->init_req(1, _gvn.transform( slow_result ));
4982 r->init_req(1, iftrue);
4983
4984 // Else fall through
4985 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4986 set_control(iffalse);
4987
4988 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4989 r->init_req(2, iffalse);
4990
4991 // Post merge
4992 set_control(_gvn.transform(r));
4993 record_for_igvn(r);
4994
4995 C->set_has_split_ifs(true); // Has chance for split-if optimization
4996 result = phi;
4997 assert(result->bottom_type()->isa_long(), "must be");
4998 break;
4999 }
5000
5001 case vmIntrinsics::_floatToIntBits: {
5002 // two paths (plus control) merge in a wood
5003 RegionNode *r = new RegionNode(3);
5004 Node *phi = new PhiNode(r, TypeInt::INT);
5005
5006 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5007 // Build the boolean node
5008 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5009
5010 // Branch either way.
5011 // NaN case is less traveled, which makes all the difference.
5012 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5013 Node *opt_isnan = _gvn.transform(ifisnan);
5014 assert( opt_isnan->is_If(), "Expect an IfNode");
5015 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5016 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5017
5018 set_control(iftrue);
5019
5020 static const jint nan_bits = 0x7fc00000;
5021 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5022 phi->init_req(1, _gvn.transform( slow_result ));
5023 r->init_req(1, iftrue);
5024
5025 // Else fall through
5026 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5027 set_control(iffalse);
5028
5029 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5030 r->init_req(2, iffalse);
5031
5032 // Post merge
5033 set_control(_gvn.transform(r));
5034 record_for_igvn(r);
5035
5036 C->set_has_split_ifs(true); // Has chance for split-if optimization
5037 result = phi;
5038 assert(result->bottom_type()->isa_int(), "must be");
5039 break;
5040 }
5041
5042 default:
5043 fatal_unexpected_iid(id);
5044 break;
5045 }
5046 set_result(_gvn.transform(result));
5047 return true;
5048 }
5049
5050 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5051 Node* arg = argument(0);
5052 Node* result = nullptr;
5053
5054 switch (id) {
5055 case vmIntrinsics::_floatIsInfinite:
5056 result = new IsInfiniteFNode(arg);
5057 break;
5058 case vmIntrinsics::_floatIsFinite:
5059 result = new IsFiniteFNode(arg);
5060 break;
5061 case vmIntrinsics::_doubleIsInfinite:
5062 result = new IsInfiniteDNode(arg);
5063 break;
5064 case vmIntrinsics::_doubleIsFinite:
5065 result = new IsFiniteDNode(arg);
5066 break;
5067 default:
5068 fatal_unexpected_iid(id);
5069 break;
5070 }
5071 set_result(_gvn.transform(result));
5072 return true;
5073 }
5074
5075 //----------------------inline_unsafe_copyMemory-------------------------
5076 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5077
5078 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5079 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5080 const Type* base_t = gvn.type(base);
5081
5082 bool in_native = (base_t == TypePtr::NULL_PTR);
5083 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5084 bool is_mixed = !in_heap && !in_native;
5085
5086 if (is_mixed) {
5087 return true; // mixed accesses can touch both on-heap and off-heap memory
5088 }
5089 if (in_heap) {
5090 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5091 if (!is_prim_array) {
5092 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5093 // there's not enough type information available to determine proper memory slice for it.
5094 return true;
5095 }
5096 }
5097 return false;
5098 }
5099
5100 bool LibraryCallKit::inline_unsafe_copyMemory() {
5101 if (callee()->is_static()) return false; // caller must have the capability!
5102 null_check_receiver(); // null-check receiver
5103 if (stopped()) return true;
5104
5105 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5106
5107 Node* src_base = argument(1); // type: oop
5108 Node* src_off = ConvL2X(argument(2)); // type: long
5109 Node* dst_base = argument(4); // type: oop
5110 Node* dst_off = ConvL2X(argument(5)); // type: long
5111 Node* size = ConvL2X(argument(7)); // type: long
5112
5113 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5114 "fieldOffset must be byte-scaled");
5115
5116 Node* src_addr = make_unsafe_address(src_base, src_off);
5117 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5118
5119 Node* thread = _gvn.transform(new ThreadLocalNode());
5120 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5121 BasicType doing_unsafe_access_bt = T_BYTE;
5122 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5123
5124 // update volatile field
5125 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5126
5127 int flags = RC_LEAF | RC_NO_FP;
5128
5129 const TypePtr* dst_type = TypePtr::BOTTOM;
5130
5131 // Adjust memory effects of the runtime call based on input values.
5132 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5133 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5134 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5135
5136 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5137 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5138 flags |= RC_NARROW_MEM; // narrow in memory
5139 }
5140 }
5141
5142 // Call it. Note that the length argument is not scaled.
5143 make_runtime_call(flags,
5144 OptoRuntime::fast_arraycopy_Type(),
5145 StubRoutines::unsafe_arraycopy(),
5146 "unsafe_arraycopy",
5147 dst_type,
5148 src_addr, dst_addr, size XTOP);
5149
5150 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5151
5152 return true;
5153 }
5154
5155 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5156 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5157 bool LibraryCallKit::inline_unsafe_setMemory() {
5158 if (callee()->is_static()) return false; // caller must have the capability!
5159 null_check_receiver(); // null-check receiver
5160 if (stopped()) return true;
5161
5162 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5163
5164 Node* dst_base = argument(1); // type: oop
5165 Node* dst_off = ConvL2X(argument(2)); // type: long
5166 Node* size = ConvL2X(argument(4)); // type: long
5167 Node* byte = argument(6); // type: byte
5168
5169 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5170 "fieldOffset must be byte-scaled");
5171
5172 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5173
5174 Node* thread = _gvn.transform(new ThreadLocalNode());
5175 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5176 BasicType doing_unsafe_access_bt = T_BYTE;
5177 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5178
5179 // update volatile field
5180 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5181
5182 int flags = RC_LEAF | RC_NO_FP;
5183
5184 const TypePtr* dst_type = TypePtr::BOTTOM;
5185
5186 // Adjust memory effects of the runtime call based on input values.
5187 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5188 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5189
5190 flags |= RC_NARROW_MEM; // narrow in memory
5191 }
5192
5193 // Call it. Note that the length argument is not scaled.
5194 make_runtime_call(flags,
5195 OptoRuntime::unsafe_setmemory_Type(),
5196 StubRoutines::unsafe_setmemory(),
5197 "unsafe_setmemory",
5198 dst_type,
5199 dst_addr, size XTOP, byte);
5200
5201 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5202
5203 return true;
5204 }
5205
5206 #undef XTOP
5207
5208 //------------------------clone_coping-----------------------------------
5209 // Helper function for inline_native_clone.
5210 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5211 assert(obj_size != nullptr, "");
5212 Node* raw_obj = alloc_obj->in(1);
5213 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5214
5215 AllocateNode* alloc = nullptr;
5216 if (ReduceBulkZeroing &&
5217 // If we are implementing an array clone without knowing its source type
5218 // (can happen when compiling the array-guarded branch of a reflective
5219 // Object.clone() invocation), initialize the array within the allocation.
5220 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5221 // to a runtime clone call that assumes fully initialized source arrays.
5222 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5223 // We will be completely responsible for initializing this object -
5224 // mark Initialize node as complete.
5225 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5226 // The object was just allocated - there should be no any stores!
5227 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5228 // Mark as complete_with_arraycopy so that on AllocateNode
5229 // expansion, we know this AllocateNode is initialized by an array
5230 // copy and a StoreStore barrier exists after the array copy.
5231 alloc->initialization()->set_complete_with_arraycopy();
5232 }
5233
5234 Node* size = _gvn.transform(obj_size);
5235 access_clone(obj, alloc_obj, size, is_array);
5236
5237 // Do not let reads from the cloned object float above the arraycopy.
5238 if (alloc != nullptr) {
5239 // Do not let stores that initialize this object be reordered with
5240 // a subsequent store that would make this object accessible by
5241 // other threads.
5242 // Record what AllocateNode this StoreStore protects so that
5243 // escape analysis can go from the MemBarStoreStoreNode to the
5244 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5245 // based on the escape status of the AllocateNode.
5246 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5247 } else {
5248 insert_mem_bar(Op_MemBarCPUOrder);
5249 }
5250 }
5251
5252 //------------------------inline_native_clone----------------------------
5253 // protected native Object java.lang.Object.clone();
5254 //
5255 // Here are the simple edge cases:
5256 // null receiver => normal trap
5257 // virtual and clone was overridden => slow path to out-of-line clone
5258 // not cloneable or finalizer => slow path to out-of-line Object.clone
5259 //
5260 // The general case has two steps, allocation and copying.
5261 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5262 //
5263 // Copying also has two cases, oop arrays and everything else.
5264 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5265 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5266 //
5267 // These steps fold up nicely if and when the cloned object's klass
5268 // can be sharply typed as an object array, a type array, or an instance.
5269 //
5270 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5271 PhiNode* result_val;
5272
5273 // Set the reexecute bit for the interpreter to reexecute
5274 // the bytecode that invokes Object.clone if deoptimization happens.
5275 { PreserveReexecuteState preexecs(this);
5276 jvms()->set_should_reexecute(true);
5277
5278 Node* obj = null_check_receiver();
5279 if (stopped()) return true;
5280
5281 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5282
5283 // If we are going to clone an instance, we need its exact type to
5284 // know the number and types of fields to convert the clone to
5285 // loads/stores. Maybe a speculative type can help us.
5286 if (!obj_type->klass_is_exact() &&
5287 obj_type->speculative_type() != nullptr &&
5288 obj_type->speculative_type()->is_instance_klass()) {
5289 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5290 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5291 !spec_ik->has_injected_fields()) {
5292 if (!obj_type->isa_instptr() ||
5293 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5294 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5295 }
5296 }
5297 }
5298
5299 // Conservatively insert a memory barrier on all memory slices.
5300 // Do not let writes into the original float below the clone.
5301 insert_mem_bar(Op_MemBarCPUOrder);
5302
5303 // paths into result_reg:
5304 enum {
5305 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5306 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5307 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5308 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5309 PATH_LIMIT
5310 };
5311 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5312 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5313 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5314 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5315 record_for_igvn(result_reg);
5316
5317 Node* obj_klass = load_object_klass(obj);
5318 Node* array_obj = obj;
5319 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5320 if (array_ctl != nullptr) {
5321 // It's an array.
5322 PreserveJVMState pjvms(this);
5323 set_control(array_ctl);
5324 Node* obj_length = load_array_length(array_obj);
5325 Node* array_size = nullptr; // Size of the array without object alignment padding.
5326 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5327
5328 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5329 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5330 // If it is an oop array, it requires very special treatment,
5331 // because gc barriers are required when accessing the array.
5332 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5333 if (is_obja != nullptr) {
5334 PreserveJVMState pjvms2(this);
5335 set_control(is_obja);
5336 // Generate a direct call to the right arraycopy function(s).
5337 // Clones are always tightly coupled.
5338 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5339 ac->set_clone_oop_array();
5340 Node* n = _gvn.transform(ac);
5341 assert(n == ac, "cannot disappear");
5342 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5343
5344 result_reg->init_req(_objArray_path, control());
5345 result_val->init_req(_objArray_path, alloc_obj);
5346 result_i_o ->set_req(_objArray_path, i_o());
5347 result_mem ->set_req(_objArray_path, reset_memory());
5348 }
5349 }
5350 // Otherwise, there are no barriers to worry about.
5351 // (We can dispense with card marks if we know the allocation
5352 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5353 // causes the non-eden paths to take compensating steps to
5354 // simulate a fresh allocation, so that no further
5355 // card marks are required in compiled code to initialize
5356 // the object.)
5357
5358 if (!stopped()) {
5359 copy_to_clone(array_obj, alloc_obj, array_size, true);
5360
5361 // Present the results of the copy.
5362 result_reg->init_req(_array_path, control());
5363 result_val->init_req(_array_path, alloc_obj);
5364 result_i_o ->set_req(_array_path, i_o());
5365 result_mem ->set_req(_array_path, reset_memory());
5366 }
5367 }
5368
5369 // We only go to the instance fast case code if we pass a number of guards.
5370 // The paths which do not pass are accumulated in the slow_region.
5371 RegionNode* slow_region = new RegionNode(1);
5372 record_for_igvn(slow_region);
5373 if (!stopped()) {
5374 // It's an instance (we did array above). Make the slow-path tests.
5375 // If this is a virtual call, we generate a funny guard. We grab
5376 // the vtable entry corresponding to clone() from the target object.
5377 // If the target method which we are calling happens to be the
5378 // Object clone() method, we pass the guard. We do not need this
5379 // guard for non-virtual calls; the caller is known to be the native
5380 // Object clone().
5381 if (is_virtual) {
5382 generate_virtual_guard(obj_klass, slow_region);
5383 }
5384
5385 // The object must be easily cloneable and must not have a finalizer.
5386 // Both of these conditions may be checked in a single test.
5387 // We could optimize the test further, but we don't care.
5388 generate_misc_flags_guard(obj_klass,
5389 // Test both conditions:
5390 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5391 // Must be cloneable but not finalizer:
5392 KlassFlags::_misc_is_cloneable_fast,
5393 slow_region);
5394 }
5395
5396 if (!stopped()) {
5397 // It's an instance, and it passed the slow-path tests.
5398 PreserveJVMState pjvms(this);
5399 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5400 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5401 // is reexecuted if deoptimization occurs and there could be problems when merging
5402 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5403 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5404
5405 copy_to_clone(obj, alloc_obj, obj_size, false);
5406
5407 // Present the results of the slow call.
5408 result_reg->init_req(_instance_path, control());
5409 result_val->init_req(_instance_path, alloc_obj);
5410 result_i_o ->set_req(_instance_path, i_o());
5411 result_mem ->set_req(_instance_path, reset_memory());
5412 }
5413
5414 // Generate code for the slow case. We make a call to clone().
5415 set_control(_gvn.transform(slow_region));
5416 if (!stopped()) {
5417 PreserveJVMState pjvms(this);
5418 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5419 // We need to deoptimize on exception (see comment above)
5420 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5421 // this->control() comes from set_results_for_java_call
5422 result_reg->init_req(_slow_path, control());
5423 result_val->init_req(_slow_path, slow_result);
5424 result_i_o ->set_req(_slow_path, i_o());
5425 result_mem ->set_req(_slow_path, reset_memory());
5426 }
5427
5428 // Return the combined state.
5429 set_control( _gvn.transform(result_reg));
5430 set_i_o( _gvn.transform(result_i_o));
5431 set_all_memory( _gvn.transform(result_mem));
5432 } // original reexecute is set back here
5433
5434 set_result(_gvn.transform(result_val));
5435 return true;
5436 }
5437
5438 // If we have a tightly coupled allocation, the arraycopy may take care
5439 // of the array initialization. If one of the guards we insert between
5440 // the allocation and the arraycopy causes a deoptimization, an
5441 // uninitialized array will escape the compiled method. To prevent that
5442 // we set the JVM state for uncommon traps between the allocation and
5443 // the arraycopy to the state before the allocation so, in case of
5444 // deoptimization, we'll reexecute the allocation and the
5445 // initialization.
5446 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5447 if (alloc != nullptr) {
5448 ciMethod* trap_method = alloc->jvms()->method();
5449 int trap_bci = alloc->jvms()->bci();
5450
5451 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5452 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5453 // Make sure there's no store between the allocation and the
5454 // arraycopy otherwise visible side effects could be rexecuted
5455 // in case of deoptimization and cause incorrect execution.
5456 bool no_interfering_store = true;
5457 Node* mem = alloc->in(TypeFunc::Memory);
5458 if (mem->is_MergeMem()) {
5459 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5460 Node* n = mms.memory();
5461 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5462 assert(n->is_Store(), "what else?");
5463 no_interfering_store = false;
5464 break;
5465 }
5466 }
5467 } else {
5468 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5469 Node* n = mms.memory();
5470 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5471 assert(n->is_Store(), "what else?");
5472 no_interfering_store = false;
5473 break;
5474 }
5475 }
5476 }
5477
5478 if (no_interfering_store) {
5479 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5480
5481 JVMState* saved_jvms = jvms();
5482 saved_reexecute_sp = _reexecute_sp;
5483
5484 set_jvms(sfpt->jvms());
5485 _reexecute_sp = jvms()->sp();
5486
5487 return saved_jvms;
5488 }
5489 }
5490 }
5491 return nullptr;
5492 }
5493
5494 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5495 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5496 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5497 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5498 uint size = alloc->req();
5499 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5500 old_jvms->set_map(sfpt);
5501 for (uint i = 0; i < size; i++) {
5502 sfpt->init_req(i, alloc->in(i));
5503 }
5504 // re-push array length for deoptimization
5505 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5506 old_jvms->set_sp(old_jvms->sp()+1);
5507 old_jvms->set_monoff(old_jvms->monoff()+1);
5508 old_jvms->set_scloff(old_jvms->scloff()+1);
5509 old_jvms->set_endoff(old_jvms->endoff()+1);
5510 old_jvms->set_should_reexecute(true);
5511
5512 sfpt->set_i_o(map()->i_o());
5513 sfpt->set_memory(map()->memory());
5514 sfpt->set_control(map()->control());
5515 return sfpt;
5516 }
5517
5518 // In case of a deoptimization, we restart execution at the
5519 // allocation, allocating a new array. We would leave an uninitialized
5520 // array in the heap that GCs wouldn't expect. Move the allocation
5521 // after the traps so we don't allocate the array if we
5522 // deoptimize. This is possible because tightly_coupled_allocation()
5523 // guarantees there's no observer of the allocated array at this point
5524 // and the control flow is simple enough.
5525 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5526 int saved_reexecute_sp, uint new_idx) {
5527 if (saved_jvms_before_guards != nullptr && !stopped()) {
5528 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5529
5530 assert(alloc != nullptr, "only with a tightly coupled allocation");
5531 // restore JVM state to the state at the arraycopy
5532 saved_jvms_before_guards->map()->set_control(map()->control());
5533 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5534 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5535 // If we've improved the types of some nodes (null check) while
5536 // emitting the guards, propagate them to the current state
5537 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5538 set_jvms(saved_jvms_before_guards);
5539 _reexecute_sp = saved_reexecute_sp;
5540
5541 // Remove the allocation from above the guards
5542 CallProjections callprojs;
5543 alloc->extract_projections(&callprojs, true);
5544 InitializeNode* init = alloc->initialization();
5545 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5546 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5547 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5548
5549 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5550 // the allocation (i.e. is only valid if the allocation succeeds):
5551 // 1) replace CastIINode with AllocateArrayNode's length here
5552 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5553 //
5554 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5555 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5556 Node* init_control = init->proj_out(TypeFunc::Control);
5557 Node* alloc_length = alloc->Ideal_length();
5558 #ifdef ASSERT
5559 Node* prev_cast = nullptr;
5560 #endif
5561 for (uint i = 0; i < init_control->outcnt(); i++) {
5562 Node* init_out = init_control->raw_out(i);
5563 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5564 #ifdef ASSERT
5565 if (prev_cast == nullptr) {
5566 prev_cast = init_out;
5567 } else {
5568 if (prev_cast->cmp(*init_out) == false) {
5569 prev_cast->dump();
5570 init_out->dump();
5571 assert(false, "not equal CastIINode");
5572 }
5573 }
5574 #endif
5575 C->gvn_replace_by(init_out, alloc_length);
5576 }
5577 }
5578 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5579
5580 // move the allocation here (after the guards)
5581 _gvn.hash_delete(alloc);
5582 alloc->set_req(TypeFunc::Control, control());
5583 alloc->set_req(TypeFunc::I_O, i_o());
5584 Node *mem = reset_memory();
5585 set_all_memory(mem);
5586 alloc->set_req(TypeFunc::Memory, mem);
5587 set_control(init->proj_out_or_null(TypeFunc::Control));
5588 set_i_o(callprojs.fallthrough_ioproj);
5589
5590 // Update memory as done in GraphKit::set_output_for_allocation()
5591 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5592 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5593 if (ary_type->isa_aryptr() && length_type != nullptr) {
5594 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5595 }
5596 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5597 int elemidx = C->get_alias_index(telemref);
5598 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5599 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5600
5601 Node* allocx = _gvn.transform(alloc);
5602 assert(allocx == alloc, "where has the allocation gone?");
5603 assert(dest->is_CheckCastPP(), "not an allocation result?");
5604
5605 _gvn.hash_delete(dest);
5606 dest->set_req(0, control());
5607 Node* destx = _gvn.transform(dest);
5608 assert(destx == dest, "where has the allocation result gone?");
5609
5610 array_ideal_length(alloc, ary_type, true);
5611 }
5612 }
5613
5614 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5615 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5616 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5617 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5618 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5619 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5620 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5621 JVMState* saved_jvms_before_guards) {
5622 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5623 // There is at least one unrelated uncommon trap which needs to be replaced.
5624 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5625
5626 JVMState* saved_jvms = jvms();
5627 const int saved_reexecute_sp = _reexecute_sp;
5628 set_jvms(sfpt->jvms());
5629 _reexecute_sp = jvms()->sp();
5630
5631 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5632
5633 // Restore state
5634 set_jvms(saved_jvms);
5635 _reexecute_sp = saved_reexecute_sp;
5636 }
5637 }
5638
5639 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5640 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5641 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5642 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5643 while (if_proj->is_IfProj()) {
5644 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5645 if (uncommon_trap != nullptr) {
5646 create_new_uncommon_trap(uncommon_trap);
5647 }
5648 assert(if_proj->in(0)->is_If(), "must be If");
5649 if_proj = if_proj->in(0)->in(0);
5650 }
5651 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5652 "must have reached control projection of init node");
5653 }
5654
5655 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5656 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5657 assert(trap_request != 0, "no valid UCT trap request");
5658 PreserveJVMState pjvms(this);
5659 set_control(uncommon_trap_call->in(0));
5660 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5661 Deoptimization::trap_request_action(trap_request));
5662 assert(stopped(), "Should be stopped");
5663 _gvn.hash_delete(uncommon_trap_call);
5664 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5665 }
5666
5667 // Common checks for array sorting intrinsics arguments.
5668 // Returns `true` if checks passed.
5669 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5670 // check address of the class
5671 if (elementType == nullptr || elementType->is_top()) {
5672 return false; // dead path
5673 }
5674 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5675 if (elem_klass == nullptr) {
5676 return false; // dead path
5677 }
5678 // java_mirror_type() returns non-null for compile-time Class constants only
5679 ciType* elem_type = elem_klass->java_mirror_type();
5680 if (elem_type == nullptr) {
5681 return false;
5682 }
5683 bt = elem_type->basic_type();
5684 // Disable the intrinsic if the CPU does not support SIMD sort
5685 if (!Matcher::supports_simd_sort(bt)) {
5686 return false;
5687 }
5688 // check address of the array
5689 if (obj == nullptr || obj->is_top()) {
5690 return false; // dead path
5691 }
5692 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5693 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5694 return false; // failed input validation
5695 }
5696 return true;
5697 }
5698
5699 //------------------------------inline_array_partition-----------------------
5700 bool LibraryCallKit::inline_array_partition() {
5701 address stubAddr = StubRoutines::select_array_partition_function();
5702 if (stubAddr == nullptr) {
5703 return false; // Intrinsic's stub is not implemented on this platform
5704 }
5705 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5706
5707 // no receiver because it is a static method
5708 Node* elementType = argument(0);
5709 Node* obj = argument(1);
5710 Node* offset = argument(2); // long
5711 Node* fromIndex = argument(4);
5712 Node* toIndex = argument(5);
5713 Node* indexPivot1 = argument(6);
5714 Node* indexPivot2 = argument(7);
5715 // PartitionOperation: argument(8) is ignored
5716
5717 Node* pivotIndices = nullptr;
5718 BasicType bt = T_ILLEGAL;
5719
5720 if (!check_array_sort_arguments(elementType, obj, bt)) {
5721 return false;
5722 }
5723 null_check(obj);
5724 // If obj is dead, only null-path is taken.
5725 if (stopped()) {
5726 return true;
5727 }
5728 // Set the original stack and the reexecute bit for the interpreter to reexecute
5729 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5730 { PreserveReexecuteState preexecs(this);
5731 jvms()->set_should_reexecute(true);
5732
5733 Node* obj_adr = make_unsafe_address(obj, offset);
5734
5735 // create the pivotIndices array of type int and size = 2
5736 Node* size = intcon(2);
5737 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5738 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5739 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5740 guarantee(alloc != nullptr, "created above");
5741 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5742
5743 // pass the basic type enum to the stub
5744 Node* elemType = intcon(bt);
5745
5746 // Call the stub
5747 const char *stubName = "array_partition_stub";
5748 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5749 stubAddr, stubName, TypePtr::BOTTOM,
5750 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5751 indexPivot1, indexPivot2);
5752
5753 } // original reexecute is set back here
5754
5755 if (!stopped()) {
5756 set_result(pivotIndices);
5757 }
5758
5759 return true;
5760 }
5761
5762
5763 //------------------------------inline_array_sort-----------------------
5764 bool LibraryCallKit::inline_array_sort() {
5765 address stubAddr = StubRoutines::select_arraysort_function();
5766 if (stubAddr == nullptr) {
5767 return false; // Intrinsic's stub is not implemented on this platform
5768 }
5769 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5770
5771 // no receiver because it is a static method
5772 Node* elementType = argument(0);
5773 Node* obj = argument(1);
5774 Node* offset = argument(2); // long
5775 Node* fromIndex = argument(4);
5776 Node* toIndex = argument(5);
5777 // SortOperation: argument(6) is ignored
5778
5779 BasicType bt = T_ILLEGAL;
5780
5781 if (!check_array_sort_arguments(elementType, obj, bt)) {
5782 return false;
5783 }
5784 null_check(obj);
5785 // If obj is dead, only null-path is taken.
5786 if (stopped()) {
5787 return true;
5788 }
5789 Node* obj_adr = make_unsafe_address(obj, offset);
5790
5791 // pass the basic type enum to the stub
5792 Node* elemType = intcon(bt);
5793
5794 // Call the stub.
5795 const char *stubName = "arraysort_stub";
5796 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5797 stubAddr, stubName, TypePtr::BOTTOM,
5798 obj_adr, elemType, fromIndex, toIndex);
5799
5800 return true;
5801 }
5802
5803
5804 //------------------------------inline_arraycopy-----------------------
5805 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5806 // Object dest, int destPos,
5807 // int length);
5808 bool LibraryCallKit::inline_arraycopy() {
5809 // Get the arguments.
5810 Node* src = argument(0); // type: oop
5811 Node* src_offset = argument(1); // type: int
5812 Node* dest = argument(2); // type: oop
5813 Node* dest_offset = argument(3); // type: int
5814 Node* length = argument(4); // type: int
5815
5816 uint new_idx = C->unique();
5817
5818 // Check for allocation before we add nodes that would confuse
5819 // tightly_coupled_allocation()
5820 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5821
5822 int saved_reexecute_sp = -1;
5823 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5824 // See arraycopy_restore_alloc_state() comment
5825 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5826 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5827 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5828 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5829
5830 // The following tests must be performed
5831 // (1) src and dest are arrays.
5832 // (2) src and dest arrays must have elements of the same BasicType
5833 // (3) src and dest must not be null.
5834 // (4) src_offset must not be negative.
5835 // (5) dest_offset must not be negative.
5836 // (6) length must not be negative.
5837 // (7) src_offset + length must not exceed length of src.
5838 // (8) dest_offset + length must not exceed length of dest.
5839 // (9) each element of an oop array must be assignable
5840
5841 // (3) src and dest must not be null.
5842 // always do this here because we need the JVM state for uncommon traps
5843 Node* null_ctl = top();
5844 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5845 assert(null_ctl->is_top(), "no null control here");
5846 dest = null_check(dest, T_ARRAY);
5847
5848 if (!can_emit_guards) {
5849 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5850 // guards but the arraycopy node could still take advantage of a
5851 // tightly allocated allocation. tightly_coupled_allocation() is
5852 // called again to make sure it takes the null check above into
5853 // account: the null check is mandatory and if it caused an
5854 // uncommon trap to be emitted then the allocation can't be
5855 // considered tightly coupled in this context.
5856 alloc = tightly_coupled_allocation(dest);
5857 }
5858
5859 bool validated = false;
5860
5861 const Type* src_type = _gvn.type(src);
5862 const Type* dest_type = _gvn.type(dest);
5863 const TypeAryPtr* top_src = src_type->isa_aryptr();
5864 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5865
5866 // Do we have the type of src?
5867 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5868 // Do we have the type of dest?
5869 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5870 // Is the type for src from speculation?
5871 bool src_spec = false;
5872 // Is the type for dest from speculation?
5873 bool dest_spec = false;
5874
5875 if ((!has_src || !has_dest) && can_emit_guards) {
5876 // We don't have sufficient type information, let's see if
5877 // speculative types can help. We need to have types for both src
5878 // and dest so that it pays off.
5879
5880 // Do we already have or could we have type information for src
5881 bool could_have_src = has_src;
5882 // Do we already have or could we have type information for dest
5883 bool could_have_dest = has_dest;
5884
5885 ciKlass* src_k = nullptr;
5886 if (!has_src) {
5887 src_k = src_type->speculative_type_not_null();
5888 if (src_k != nullptr && src_k->is_array_klass()) {
5889 could_have_src = true;
5890 }
5891 }
5892
5893 ciKlass* dest_k = nullptr;
5894 if (!has_dest) {
5895 dest_k = dest_type->speculative_type_not_null();
5896 if (dest_k != nullptr && dest_k->is_array_klass()) {
5897 could_have_dest = true;
5898 }
5899 }
5900
5901 if (could_have_src && could_have_dest) {
5902 // This is going to pay off so emit the required guards
5903 if (!has_src) {
5904 src = maybe_cast_profiled_obj(src, src_k, true);
5905 src_type = _gvn.type(src);
5906 top_src = src_type->isa_aryptr();
5907 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5908 src_spec = true;
5909 }
5910 if (!has_dest) {
5911 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5912 dest_type = _gvn.type(dest);
5913 top_dest = dest_type->isa_aryptr();
5914 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5915 dest_spec = true;
5916 }
5917 }
5918 }
5919
5920 if (has_src && has_dest && can_emit_guards) {
5921 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5922 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5923 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5924 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5925
5926 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5927 // If both arrays are object arrays then having the exact types
5928 // for both will remove the need for a subtype check at runtime
5929 // before the call and may make it possible to pick a faster copy
5930 // routine (without a subtype check on every element)
5931 // Do we have the exact type of src?
5932 bool could_have_src = src_spec;
5933 // Do we have the exact type of dest?
5934 bool could_have_dest = dest_spec;
5935 ciKlass* src_k = nullptr;
5936 ciKlass* dest_k = nullptr;
5937 if (!src_spec) {
5938 src_k = src_type->speculative_type_not_null();
5939 if (src_k != nullptr && src_k->is_array_klass()) {
5940 could_have_src = true;
5941 }
5942 }
5943 if (!dest_spec) {
5944 dest_k = dest_type->speculative_type_not_null();
5945 if (dest_k != nullptr && dest_k->is_array_klass()) {
5946 could_have_dest = true;
5947 }
5948 }
5949 if (could_have_src && could_have_dest) {
5950 // If we can have both exact types, emit the missing guards
5951 if (could_have_src && !src_spec) {
5952 src = maybe_cast_profiled_obj(src, src_k, true);
5953 }
5954 if (could_have_dest && !dest_spec) {
5955 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5956 }
5957 }
5958 }
5959 }
5960
5961 ciMethod* trap_method = method();
5962 int trap_bci = bci();
5963 if (saved_jvms_before_guards != nullptr) {
5964 trap_method = alloc->jvms()->method();
5965 trap_bci = alloc->jvms()->bci();
5966 }
5967
5968 bool negative_length_guard_generated = false;
5969
5970 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5971 can_emit_guards &&
5972 !src->is_top() && !dest->is_top()) {
5973 // validate arguments: enables transformation the ArrayCopyNode
5974 validated = true;
5975
5976 RegionNode* slow_region = new RegionNode(1);
5977 record_for_igvn(slow_region);
5978
5979 // (1) src and dest are arrays.
5980 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5981 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5982
5983 // (2) src and dest arrays must have elements of the same BasicType
5984 // done at macro expansion or at Ideal transformation time
5985
5986 // (4) src_offset must not be negative.
5987 generate_negative_guard(src_offset, slow_region);
5988
5989 // (5) dest_offset must not be negative.
5990 generate_negative_guard(dest_offset, slow_region);
5991
5992 // (7) src_offset + length must not exceed length of src.
5993 generate_limit_guard(src_offset, length,
5994 load_array_length(src),
5995 slow_region);
5996
5997 // (8) dest_offset + length must not exceed length of dest.
5998 generate_limit_guard(dest_offset, length,
5999 load_array_length(dest),
6000 slow_region);
6001
6002 // (6) length must not be negative.
6003 // This is also checked in generate_arraycopy() during macro expansion, but
6004 // we also have to check it here for the case where the ArrayCopyNode will
6005 // be eliminated by Escape Analysis.
6006 if (EliminateAllocations) {
6007 generate_negative_guard(length, slow_region);
6008 negative_length_guard_generated = true;
6009 }
6010
6011 // (9) each element of an oop array must be assignable
6012 Node* dest_klass = load_object_klass(dest);
6013 if (src != dest) {
6014 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6015
6016 if (not_subtype_ctrl != top()) {
6017 PreserveJVMState pjvms(this);
6018 set_control(not_subtype_ctrl);
6019 uncommon_trap(Deoptimization::Reason_intrinsic,
6020 Deoptimization::Action_make_not_entrant);
6021 assert(stopped(), "Should be stopped");
6022 }
6023 }
6024 {
6025 PreserveJVMState pjvms(this);
6026 set_control(_gvn.transform(slow_region));
6027 uncommon_trap(Deoptimization::Reason_intrinsic,
6028 Deoptimization::Action_make_not_entrant);
6029 assert(stopped(), "Should be stopped");
6030 }
6031
6032 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6033 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6034 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6035 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6036 }
6037
6038 if (stopped()) {
6039 return true;
6040 }
6041
6042 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6043 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6044 // so the compiler has a chance to eliminate them: during macro expansion,
6045 // we have to set their control (CastPP nodes are eliminated).
6046 load_object_klass(src), load_object_klass(dest),
6047 load_array_length(src), load_array_length(dest));
6048
6049 ac->set_arraycopy(validated);
6050
6051 Node* n = _gvn.transform(ac);
6052 if (n == ac) {
6053 ac->connect_outputs(this);
6054 } else {
6055 assert(validated, "shouldn't transform if all arguments not validated");
6056 set_all_memory(n);
6057 }
6058 clear_upper_avx();
6059
6060
6061 return true;
6062 }
6063
6064
6065 // Helper function which determines if an arraycopy immediately follows
6066 // an allocation, with no intervening tests or other escapes for the object.
6067 AllocateArrayNode*
6068 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6069 if (stopped()) return nullptr; // no fast path
6070 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6071
6072 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6073 if (alloc == nullptr) return nullptr;
6074
6075 Node* rawmem = memory(Compile::AliasIdxRaw);
6076 // Is the allocation's memory state untouched?
6077 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6078 // Bail out if there have been raw-memory effects since the allocation.
6079 // (Example: There might have been a call or safepoint.)
6080 return nullptr;
6081 }
6082 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6083 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6084 return nullptr;
6085 }
6086
6087 // There must be no unexpected observers of this allocation.
6088 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6089 Node* obs = ptr->fast_out(i);
6090 if (obs != this->map()) {
6091 return nullptr;
6092 }
6093 }
6094
6095 // This arraycopy must unconditionally follow the allocation of the ptr.
6096 Node* alloc_ctl = ptr->in(0);
6097 Node* ctl = control();
6098 while (ctl != alloc_ctl) {
6099 // There may be guards which feed into the slow_region.
6100 // Any other control flow means that we might not get a chance
6101 // to finish initializing the allocated object.
6102 // Various low-level checks bottom out in uncommon traps. These
6103 // are considered safe since we've already checked above that
6104 // there is no unexpected observer of this allocation.
6105 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6106 assert(ctl->in(0)->is_If(), "must be If");
6107 ctl = ctl->in(0)->in(0);
6108 } else {
6109 return nullptr;
6110 }
6111 }
6112
6113 // If we get this far, we have an allocation which immediately
6114 // precedes the arraycopy, and we can take over zeroing the new object.
6115 // The arraycopy will finish the initialization, and provide
6116 // a new control state to which we will anchor the destination pointer.
6117
6118 return alloc;
6119 }
6120
6121 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6122 if (node->is_IfProj()) {
6123 Node* other_proj = node->as_IfProj()->other_if_proj();
6124 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6125 Node* obs = other_proj->fast_out(j);
6126 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6127 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6128 return obs->as_CallStaticJava();
6129 }
6130 }
6131 }
6132 return nullptr;
6133 }
6134
6135 //-------------inline_encodeISOArray-----------------------------------
6136 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6137 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6138 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6139 // encode char[] to byte[] in ISO_8859_1 or ASCII
6140 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6141 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6142 // no receiver since it is static method
6143 Node *src = argument(0);
6144 Node *src_offset = argument(1);
6145 Node *dst = argument(2);
6146 Node *dst_offset = argument(3);
6147 Node *length = argument(4);
6148
6149 // Cast source & target arrays to not-null
6150 if (VerifyIntrinsicChecks) {
6151 src = must_be_not_null(src, true);
6152 dst = must_be_not_null(dst, true);
6153 if (stopped()) {
6154 return true;
6155 }
6156 }
6157
6158 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6159 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6160 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6161 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6162 // failed array check
6163 return false;
6164 }
6165
6166 // Figure out the size and type of the elements we will be copying.
6167 BasicType src_elem = src_type->elem()->array_element_basic_type();
6168 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6169 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6170 return false;
6171 }
6172
6173 // Check source & target bounds
6174 if (VerifyIntrinsicChecks) {
6175 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6176 generate_string_range_check(dst, dst_offset, length, false, true);
6177 if (stopped()) {
6178 return true;
6179 }
6180 }
6181
6182 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6183 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6184 // 'src_start' points to src array + scaled offset
6185 // 'dst_start' points to dst array + scaled offset
6186
6187 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6188 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6189 enc = _gvn.transform(enc);
6190 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6191 set_memory(res_mem, mtype);
6192 set_result(enc);
6193 clear_upper_avx();
6194
6195 return true;
6196 }
6197
6198 //-------------inline_multiplyToLen-----------------------------------
6199 bool LibraryCallKit::inline_multiplyToLen() {
6200 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6201
6202 address stubAddr = StubRoutines::multiplyToLen();
6203 if (stubAddr == nullptr) {
6204 return false; // Intrinsic's stub is not implemented on this platform
6205 }
6206 const char* stubName = "multiplyToLen";
6207
6208 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6209
6210 // no receiver because it is a static method
6211 Node* x = argument(0);
6212 Node* xlen = argument(1);
6213 Node* y = argument(2);
6214 Node* ylen = argument(3);
6215 Node* z = argument(4);
6216
6217 x = must_be_not_null(x, true);
6218 y = must_be_not_null(y, true);
6219
6220 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6221 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6222 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6223 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6224 // failed array check
6225 return false;
6226 }
6227
6228 BasicType x_elem = x_type->elem()->array_element_basic_type();
6229 BasicType y_elem = y_type->elem()->array_element_basic_type();
6230 if (x_elem != T_INT || y_elem != T_INT) {
6231 return false;
6232 }
6233
6234 Node* x_start = array_element_address(x, intcon(0), x_elem);
6235 Node* y_start = array_element_address(y, intcon(0), y_elem);
6236 // 'x_start' points to x array + scaled xlen
6237 // 'y_start' points to y array + scaled ylen
6238
6239 Node* z_start = array_element_address(z, intcon(0), T_INT);
6240
6241 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6242 OptoRuntime::multiplyToLen_Type(),
6243 stubAddr, stubName, TypePtr::BOTTOM,
6244 x_start, xlen, y_start, ylen, z_start);
6245
6246 C->set_has_split_ifs(true); // Has chance for split-if optimization
6247 set_result(z);
6248 return true;
6249 }
6250
6251 //-------------inline_squareToLen------------------------------------
6252 bool LibraryCallKit::inline_squareToLen() {
6253 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6254
6255 address stubAddr = StubRoutines::squareToLen();
6256 if (stubAddr == nullptr) {
6257 return false; // Intrinsic's stub is not implemented on this platform
6258 }
6259 const char* stubName = "squareToLen";
6260
6261 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6262
6263 Node* x = argument(0);
6264 Node* len = argument(1);
6265 Node* z = argument(2);
6266 Node* zlen = argument(3);
6267
6268 x = must_be_not_null(x, true);
6269 z = must_be_not_null(z, true);
6270
6271 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6272 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6273 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6274 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6275 // failed array check
6276 return false;
6277 }
6278
6279 BasicType x_elem = x_type->elem()->array_element_basic_type();
6280 BasicType z_elem = z_type->elem()->array_element_basic_type();
6281 if (x_elem != T_INT || z_elem != T_INT) {
6282 return false;
6283 }
6284
6285
6286 Node* x_start = array_element_address(x, intcon(0), x_elem);
6287 Node* z_start = array_element_address(z, intcon(0), z_elem);
6288
6289 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6290 OptoRuntime::squareToLen_Type(),
6291 stubAddr, stubName, TypePtr::BOTTOM,
6292 x_start, len, z_start, zlen);
6293
6294 set_result(z);
6295 return true;
6296 }
6297
6298 //-------------inline_mulAdd------------------------------------------
6299 bool LibraryCallKit::inline_mulAdd() {
6300 assert(UseMulAddIntrinsic, "not implemented on this platform");
6301
6302 address stubAddr = StubRoutines::mulAdd();
6303 if (stubAddr == nullptr) {
6304 return false; // Intrinsic's stub is not implemented on this platform
6305 }
6306 const char* stubName = "mulAdd";
6307
6308 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6309
6310 Node* out = argument(0);
6311 Node* in = argument(1);
6312 Node* offset = argument(2);
6313 Node* len = argument(3);
6314 Node* k = argument(4);
6315
6316 in = must_be_not_null(in, true);
6317 out = must_be_not_null(out, true);
6318
6319 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6320 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6321 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6322 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6323 // failed array check
6324 return false;
6325 }
6326
6327 BasicType out_elem = out_type->elem()->array_element_basic_type();
6328 BasicType in_elem = in_type->elem()->array_element_basic_type();
6329 if (out_elem != T_INT || in_elem != T_INT) {
6330 return false;
6331 }
6332
6333 Node* outlen = load_array_length(out);
6334 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6335 Node* out_start = array_element_address(out, intcon(0), out_elem);
6336 Node* in_start = array_element_address(in, intcon(0), in_elem);
6337
6338 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6339 OptoRuntime::mulAdd_Type(),
6340 stubAddr, stubName, TypePtr::BOTTOM,
6341 out_start,in_start, new_offset, len, k);
6342 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6343 set_result(result);
6344 return true;
6345 }
6346
6347 //-------------inline_montgomeryMultiply-----------------------------------
6348 bool LibraryCallKit::inline_montgomeryMultiply() {
6349 address stubAddr = StubRoutines::montgomeryMultiply();
6350 if (stubAddr == nullptr) {
6351 return false; // Intrinsic's stub is not implemented on this platform
6352 }
6353
6354 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6355 const char* stubName = "montgomery_multiply";
6356
6357 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6358
6359 Node* a = argument(0);
6360 Node* b = argument(1);
6361 Node* n = argument(2);
6362 Node* len = argument(3);
6363 Node* inv = argument(4);
6364 Node* m = argument(6);
6365
6366 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6367 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6368 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6369 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6370 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6371 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6372 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6373 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6374 // failed array check
6375 return false;
6376 }
6377
6378 BasicType a_elem = a_type->elem()->array_element_basic_type();
6379 BasicType b_elem = b_type->elem()->array_element_basic_type();
6380 BasicType n_elem = n_type->elem()->array_element_basic_type();
6381 BasicType m_elem = m_type->elem()->array_element_basic_type();
6382 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6383 return false;
6384 }
6385
6386 // Make the call
6387 {
6388 Node* a_start = array_element_address(a, intcon(0), a_elem);
6389 Node* b_start = array_element_address(b, intcon(0), b_elem);
6390 Node* n_start = array_element_address(n, intcon(0), n_elem);
6391 Node* m_start = array_element_address(m, intcon(0), m_elem);
6392
6393 Node* call = make_runtime_call(RC_LEAF,
6394 OptoRuntime::montgomeryMultiply_Type(),
6395 stubAddr, stubName, TypePtr::BOTTOM,
6396 a_start, b_start, n_start, len, inv, top(),
6397 m_start);
6398 set_result(m);
6399 }
6400
6401 return true;
6402 }
6403
6404 bool LibraryCallKit::inline_montgomerySquare() {
6405 address stubAddr = StubRoutines::montgomerySquare();
6406 if (stubAddr == nullptr) {
6407 return false; // Intrinsic's stub is not implemented on this platform
6408 }
6409
6410 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6411 const char* stubName = "montgomery_square";
6412
6413 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6414
6415 Node* a = argument(0);
6416 Node* n = argument(1);
6417 Node* len = argument(2);
6418 Node* inv = argument(3);
6419 Node* m = argument(5);
6420
6421 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6422 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6423 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6424 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6425 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6426 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6427 // failed array check
6428 return false;
6429 }
6430
6431 BasicType a_elem = a_type->elem()->array_element_basic_type();
6432 BasicType n_elem = n_type->elem()->array_element_basic_type();
6433 BasicType m_elem = m_type->elem()->array_element_basic_type();
6434 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6435 return false;
6436 }
6437
6438 // Make the call
6439 {
6440 Node* a_start = array_element_address(a, intcon(0), a_elem);
6441 Node* n_start = array_element_address(n, intcon(0), n_elem);
6442 Node* m_start = array_element_address(m, intcon(0), m_elem);
6443
6444 Node* call = make_runtime_call(RC_LEAF,
6445 OptoRuntime::montgomerySquare_Type(),
6446 stubAddr, stubName, TypePtr::BOTTOM,
6447 a_start, n_start, len, inv, top(),
6448 m_start);
6449 set_result(m);
6450 }
6451
6452 return true;
6453 }
6454
6455 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6456 address stubAddr = nullptr;
6457 const char* stubName = nullptr;
6458
6459 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6460 if (stubAddr == nullptr) {
6461 return false; // Intrinsic's stub is not implemented on this platform
6462 }
6463
6464 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6465
6466 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6467
6468 Node* newArr = argument(0);
6469 Node* oldArr = argument(1);
6470 Node* newIdx = argument(2);
6471 Node* shiftCount = argument(3);
6472 Node* numIter = argument(4);
6473
6474 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6475 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6476 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6477 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6478 return false;
6479 }
6480
6481 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6482 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6483 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6484 return false;
6485 }
6486
6487 // Make the call
6488 {
6489 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6490 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6491
6492 Node* call = make_runtime_call(RC_LEAF,
6493 OptoRuntime::bigIntegerShift_Type(),
6494 stubAddr,
6495 stubName,
6496 TypePtr::BOTTOM,
6497 newArr_start,
6498 oldArr_start,
6499 newIdx,
6500 shiftCount,
6501 numIter);
6502 }
6503
6504 return true;
6505 }
6506
6507 //-------------inline_vectorizedMismatch------------------------------
6508 bool LibraryCallKit::inline_vectorizedMismatch() {
6509 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6510
6511 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6512 Node* obja = argument(0); // Object
6513 Node* aoffset = argument(1); // long
6514 Node* objb = argument(3); // Object
6515 Node* boffset = argument(4); // long
6516 Node* length = argument(6); // int
6517 Node* scale = argument(7); // int
6518
6519 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6520 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6521 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6522 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6523 scale == top()) {
6524 return false; // failed input validation
6525 }
6526
6527 Node* obja_adr = make_unsafe_address(obja, aoffset);
6528 Node* objb_adr = make_unsafe_address(objb, boffset);
6529
6530 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6531 //
6532 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6533 // if (length <= inline_limit) {
6534 // inline_path:
6535 // vmask = VectorMaskGen length
6536 // vload1 = LoadVectorMasked obja, vmask
6537 // vload2 = LoadVectorMasked objb, vmask
6538 // result1 = VectorCmpMasked vload1, vload2, vmask
6539 // } else {
6540 // call_stub_path:
6541 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6542 // }
6543 // exit_block:
6544 // return Phi(result1, result2);
6545 //
6546 enum { inline_path = 1, // input is small enough to process it all at once
6547 stub_path = 2, // input is too large; call into the VM
6548 PATH_LIMIT = 3
6549 };
6550
6551 Node* exit_block = new RegionNode(PATH_LIMIT);
6552 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6553 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6554
6555 Node* call_stub_path = control();
6556
6557 BasicType elem_bt = T_ILLEGAL;
6558
6559 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6560 if (scale_t->is_con()) {
6561 switch (scale_t->get_con()) {
6562 case 0: elem_bt = T_BYTE; break;
6563 case 1: elem_bt = T_SHORT; break;
6564 case 2: elem_bt = T_INT; break;
6565 case 3: elem_bt = T_LONG; break;
6566
6567 default: elem_bt = T_ILLEGAL; break; // not supported
6568 }
6569 }
6570
6571 int inline_limit = 0;
6572 bool do_partial_inline = false;
6573
6574 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6575 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6576 do_partial_inline = inline_limit >= 16;
6577 }
6578
6579 if (do_partial_inline) {
6580 assert(elem_bt != T_ILLEGAL, "sanity");
6581
6582 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6583 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6584 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6585
6586 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6587 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6588 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6589
6590 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6591
6592 if (!stopped()) {
6593 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6594
6595 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6596 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6597 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6598 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6599
6600 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6601 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6602 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6603 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6604
6605 exit_block->init_req(inline_path, control());
6606 memory_phi->init_req(inline_path, map()->memory());
6607 result_phi->init_req(inline_path, result);
6608
6609 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6610 clear_upper_avx();
6611 }
6612 }
6613 }
6614
6615 if (call_stub_path != nullptr) {
6616 set_control(call_stub_path);
6617
6618 Node* call = make_runtime_call(RC_LEAF,
6619 OptoRuntime::vectorizedMismatch_Type(),
6620 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6621 obja_adr, objb_adr, length, scale);
6622
6623 exit_block->init_req(stub_path, control());
6624 memory_phi->init_req(stub_path, map()->memory());
6625 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6626 }
6627
6628 exit_block = _gvn.transform(exit_block);
6629 memory_phi = _gvn.transform(memory_phi);
6630 result_phi = _gvn.transform(result_phi);
6631
6632 record_for_igvn(exit_block);
6633 record_for_igvn(memory_phi);
6634 record_for_igvn(result_phi);
6635
6636 set_control(exit_block);
6637 set_all_memory(memory_phi);
6638 set_result(result_phi);
6639
6640 return true;
6641 }
6642
6643 //------------------------------inline_vectorizedHashcode----------------------------
6644 bool LibraryCallKit::inline_vectorizedHashCode() {
6645 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6646
6647 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6648 Node* array = argument(0);
6649 Node* offset = argument(1);
6650 Node* length = argument(2);
6651 Node* initialValue = argument(3);
6652 Node* basic_type = argument(4);
6653
6654 if (basic_type == top()) {
6655 return false; // failed input validation
6656 }
6657
6658 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6659 if (!basic_type_t->is_con()) {
6660 return false; // Only intrinsify if mode argument is constant
6661 }
6662
6663 array = must_be_not_null(array, true);
6664
6665 BasicType bt = (BasicType)basic_type_t->get_con();
6666
6667 // Resolve address of first element
6668 Node* array_start = array_element_address(array, offset, bt);
6669
6670 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6671 array_start, length, initialValue, basic_type)));
6672 clear_upper_avx();
6673
6674 return true;
6675 }
6676
6677 /**
6678 * Calculate CRC32 for byte.
6679 * int java.util.zip.CRC32.update(int crc, int b)
6680 */
6681 bool LibraryCallKit::inline_updateCRC32() {
6682 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6683 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6684 // no receiver since it is static method
6685 Node* crc = argument(0); // type: int
6686 Node* b = argument(1); // type: int
6687
6688 /*
6689 * int c = ~ crc;
6690 * b = timesXtoThe32[(b ^ c) & 0xFF];
6691 * b = b ^ (c >>> 8);
6692 * crc = ~b;
6693 */
6694
6695 Node* M1 = intcon(-1);
6696 crc = _gvn.transform(new XorINode(crc, M1));
6697 Node* result = _gvn.transform(new XorINode(crc, b));
6698 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6699
6700 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6701 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6702 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6703 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6704
6705 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6706 result = _gvn.transform(new XorINode(crc, result));
6707 result = _gvn.transform(new XorINode(result, M1));
6708 set_result(result);
6709 return true;
6710 }
6711
6712 /**
6713 * Calculate CRC32 for byte[] array.
6714 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6715 */
6716 bool LibraryCallKit::inline_updateBytesCRC32() {
6717 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6718 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6719 // no receiver since it is static method
6720 Node* crc = argument(0); // type: int
6721 Node* src = argument(1); // type: oop
6722 Node* offset = argument(2); // type: int
6723 Node* length = argument(3); // type: int
6724
6725 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6726 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6727 // failed array check
6728 return false;
6729 }
6730
6731 // Figure out the size and type of the elements we will be copying.
6732 BasicType src_elem = src_type->elem()->array_element_basic_type();
6733 if (src_elem != T_BYTE) {
6734 return false;
6735 }
6736
6737 // 'src_start' points to src array + scaled offset
6738 src = must_be_not_null(src, true);
6739 Node* src_start = array_element_address(src, offset, src_elem);
6740
6741 // We assume that range check is done by caller.
6742 // TODO: generate range check (offset+length < src.length) in debug VM.
6743
6744 // Call the stub.
6745 address stubAddr = StubRoutines::updateBytesCRC32();
6746 const char *stubName = "updateBytesCRC32";
6747
6748 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6749 stubAddr, stubName, TypePtr::BOTTOM,
6750 crc, src_start, length);
6751 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6752 set_result(result);
6753 return true;
6754 }
6755
6756 /**
6757 * Calculate CRC32 for ByteBuffer.
6758 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6759 */
6760 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6761 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6762 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6763 // no receiver since it is static method
6764 Node* crc = argument(0); // type: int
6765 Node* src = argument(1); // type: long
6766 Node* offset = argument(3); // type: int
6767 Node* length = argument(4); // type: int
6768
6769 src = ConvL2X(src); // adjust Java long to machine word
6770 Node* base = _gvn.transform(new CastX2PNode(src));
6771 offset = ConvI2X(offset);
6772
6773 // 'src_start' points to src array + scaled offset
6774 Node* src_start = basic_plus_adr(top(), base, offset);
6775
6776 // Call the stub.
6777 address stubAddr = StubRoutines::updateBytesCRC32();
6778 const char *stubName = "updateBytesCRC32";
6779
6780 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6781 stubAddr, stubName, TypePtr::BOTTOM,
6782 crc, src_start, length);
6783 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6784 set_result(result);
6785 return true;
6786 }
6787
6788 //------------------------------get_table_from_crc32c_class-----------------------
6789 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6790 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6791 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6792
6793 return table;
6794 }
6795
6796 //------------------------------inline_updateBytesCRC32C-----------------------
6797 //
6798 // Calculate CRC32C for byte[] array.
6799 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6800 //
6801 bool LibraryCallKit::inline_updateBytesCRC32C() {
6802 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6803 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6804 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6805 // no receiver since it is a static method
6806 Node* crc = argument(0); // type: int
6807 Node* src = argument(1); // type: oop
6808 Node* offset = argument(2); // type: int
6809 Node* end = argument(3); // type: int
6810
6811 Node* length = _gvn.transform(new SubINode(end, offset));
6812
6813 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6814 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6815 // failed array check
6816 return false;
6817 }
6818
6819 // Figure out the size and type of the elements we will be copying.
6820 BasicType src_elem = src_type->elem()->array_element_basic_type();
6821 if (src_elem != T_BYTE) {
6822 return false;
6823 }
6824
6825 // 'src_start' points to src array + scaled offset
6826 src = must_be_not_null(src, true);
6827 Node* src_start = array_element_address(src, offset, src_elem);
6828
6829 // static final int[] byteTable in class CRC32C
6830 Node* table = get_table_from_crc32c_class(callee()->holder());
6831 table = must_be_not_null(table, true);
6832 Node* table_start = array_element_address(table, intcon(0), T_INT);
6833
6834 // We assume that range check is done by caller.
6835 // TODO: generate range check (offset+length < src.length) in debug VM.
6836
6837 // Call the stub.
6838 address stubAddr = StubRoutines::updateBytesCRC32C();
6839 const char *stubName = "updateBytesCRC32C";
6840
6841 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6842 stubAddr, stubName, TypePtr::BOTTOM,
6843 crc, src_start, length, table_start);
6844 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6845 set_result(result);
6846 return true;
6847 }
6848
6849 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6850 //
6851 // Calculate CRC32C for DirectByteBuffer.
6852 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6853 //
6854 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6855 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6856 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6857 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6858 // no receiver since it is a static method
6859 Node* crc = argument(0); // type: int
6860 Node* src = argument(1); // type: long
6861 Node* offset = argument(3); // type: int
6862 Node* end = argument(4); // type: int
6863
6864 Node* length = _gvn.transform(new SubINode(end, offset));
6865
6866 src = ConvL2X(src); // adjust Java long to machine word
6867 Node* base = _gvn.transform(new CastX2PNode(src));
6868 offset = ConvI2X(offset);
6869
6870 // 'src_start' points to src array + scaled offset
6871 Node* src_start = basic_plus_adr(top(), base, offset);
6872
6873 // static final int[] byteTable in class CRC32C
6874 Node* table = get_table_from_crc32c_class(callee()->holder());
6875 table = must_be_not_null(table, true);
6876 Node* table_start = array_element_address(table, intcon(0), T_INT);
6877
6878 // Call the stub.
6879 address stubAddr = StubRoutines::updateBytesCRC32C();
6880 const char *stubName = "updateBytesCRC32C";
6881
6882 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6883 stubAddr, stubName, TypePtr::BOTTOM,
6884 crc, src_start, length, table_start);
6885 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6886 set_result(result);
6887 return true;
6888 }
6889
6890 //------------------------------inline_updateBytesAdler32----------------------
6891 //
6892 // Calculate Adler32 checksum for byte[] array.
6893 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6894 //
6895 bool LibraryCallKit::inline_updateBytesAdler32() {
6896 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6897 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6898 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6899 // no receiver since it is static method
6900 Node* crc = argument(0); // type: int
6901 Node* src = argument(1); // type: oop
6902 Node* offset = argument(2); // type: int
6903 Node* length = argument(3); // type: int
6904
6905 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6906 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6907 // failed array check
6908 return false;
6909 }
6910
6911 // Figure out the size and type of the elements we will be copying.
6912 BasicType src_elem = src_type->elem()->array_element_basic_type();
6913 if (src_elem != T_BYTE) {
6914 return false;
6915 }
6916
6917 // 'src_start' points to src array + scaled offset
6918 Node* src_start = array_element_address(src, offset, src_elem);
6919
6920 // We assume that range check is done by caller.
6921 // TODO: generate range check (offset+length < src.length) in debug VM.
6922
6923 // Call the stub.
6924 address stubAddr = StubRoutines::updateBytesAdler32();
6925 const char *stubName = "updateBytesAdler32";
6926
6927 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6928 stubAddr, stubName, TypePtr::BOTTOM,
6929 crc, src_start, length);
6930 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6931 set_result(result);
6932 return true;
6933 }
6934
6935 //------------------------------inline_updateByteBufferAdler32---------------
6936 //
6937 // Calculate Adler32 checksum for DirectByteBuffer.
6938 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6939 //
6940 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6941 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6942 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6943 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6944 // no receiver since it is static method
6945 Node* crc = argument(0); // type: int
6946 Node* src = argument(1); // type: long
6947 Node* offset = argument(3); // type: int
6948 Node* length = argument(4); // type: int
6949
6950 src = ConvL2X(src); // adjust Java long to machine word
6951 Node* base = _gvn.transform(new CastX2PNode(src));
6952 offset = ConvI2X(offset);
6953
6954 // 'src_start' points to src array + scaled offset
6955 Node* src_start = basic_plus_adr(top(), base, offset);
6956
6957 // Call the stub.
6958 address stubAddr = StubRoutines::updateBytesAdler32();
6959 const char *stubName = "updateBytesAdler32";
6960
6961 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6962 stubAddr, stubName, TypePtr::BOTTOM,
6963 crc, src_start, length);
6964
6965 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6966 set_result(result);
6967 return true;
6968 }
6969
6970 //----------------------------inline_reference_get0----------------------------
6971 // public T java.lang.ref.Reference.get();
6972 bool LibraryCallKit::inline_reference_get0() {
6973 const int referent_offset = java_lang_ref_Reference::referent_offset();
6974
6975 // Get the argument:
6976 Node* reference_obj = null_check_receiver();
6977 if (stopped()) return true;
6978
6979 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6980 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6981 decorators, /*is_static*/ false, nullptr);
6982 if (result == nullptr) return false;
6983
6984 // Add memory barrier to prevent commoning reads from this field
6985 // across safepoint since GC can change its value.
6986 insert_mem_bar(Op_MemBarCPUOrder);
6987
6988 set_result(result);
6989 return true;
6990 }
6991
6992 //----------------------------inline_reference_refersTo0----------------------------
6993 // bool java.lang.ref.Reference.refersTo0();
6994 // bool java.lang.ref.PhantomReference.refersTo0();
6995 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6996 // Get arguments:
6997 Node* reference_obj = null_check_receiver();
6998 Node* other_obj = argument(1);
6999 if (stopped()) return true;
7000
7001 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7002 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7003 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7004 decorators, /*is_static*/ false, nullptr);
7005 if (referent == nullptr) return false;
7006
7007 // Add memory barrier to prevent commoning reads from this field
7008 // across safepoint since GC can change its value.
7009 insert_mem_bar(Op_MemBarCPUOrder);
7010
7011 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7012 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7013 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7014
7015 RegionNode* region = new RegionNode(3);
7016 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7017
7018 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7019 region->init_req(1, if_true);
7020 phi->init_req(1, intcon(1));
7021
7022 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7023 region->init_req(2, if_false);
7024 phi->init_req(2, intcon(0));
7025
7026 set_control(_gvn.transform(region));
7027 record_for_igvn(region);
7028 set_result(_gvn.transform(phi));
7029 return true;
7030 }
7031
7032 //----------------------------inline_reference_clear0----------------------------
7033 // void java.lang.ref.Reference.clear0();
7034 // void java.lang.ref.PhantomReference.clear0();
7035 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7036 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7037
7038 // Get arguments
7039 Node* reference_obj = null_check_receiver();
7040 if (stopped()) return true;
7041
7042 // Common access parameters
7043 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7044 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7045 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7046 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7047 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7048
7049 Node* referent = access_load_at(reference_obj,
7050 referent_field_addr,
7051 referent_field_addr_type,
7052 val_type,
7053 T_OBJECT,
7054 decorators);
7055
7056 IdealKit ideal(this);
7057 #define __ ideal.
7058 __ if_then(referent, BoolTest::ne, null());
7059 sync_kit(ideal);
7060 access_store_at(reference_obj,
7061 referent_field_addr,
7062 referent_field_addr_type,
7063 null(),
7064 val_type,
7065 T_OBJECT,
7066 decorators);
7067 __ sync_kit(this);
7068 __ end_if();
7069 final_sync(ideal);
7070 #undef __
7071
7072 return true;
7073 }
7074
7075 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7076 DecoratorSet decorators, bool is_static,
7077 ciInstanceKlass* fromKls) {
7078 if (fromKls == nullptr) {
7079 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7080 assert(tinst != nullptr, "obj is null");
7081 assert(tinst->is_loaded(), "obj is not loaded");
7082 fromKls = tinst->instance_klass();
7083 } else {
7084 assert(is_static, "only for static field access");
7085 }
7086 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7087 ciSymbol::make(fieldTypeString),
7088 is_static);
7089
7090 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7091 if (field == nullptr) return (Node *) nullptr;
7092
7093 if (is_static) {
7094 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7095 fromObj = makecon(tip);
7096 }
7097
7098 // Next code copied from Parse::do_get_xxx():
7099
7100 // Compute address and memory type.
7101 int offset = field->offset_in_bytes();
7102 bool is_vol = field->is_volatile();
7103 ciType* field_klass = field->type();
7104 assert(field_klass->is_loaded(), "should be loaded");
7105 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7106 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7107 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7108 "slice of address and input slice don't match");
7109 BasicType bt = field->layout_type();
7110
7111 // Build the resultant type of the load
7112 const Type *type;
7113 if (bt == T_OBJECT) {
7114 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7115 } else {
7116 type = Type::get_const_basic_type(bt);
7117 }
7118
7119 if (is_vol) {
7120 decorators |= MO_SEQ_CST;
7121 }
7122
7123 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7124 }
7125
7126 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7127 bool is_exact /* true */, bool is_static /* false */,
7128 ciInstanceKlass * fromKls /* nullptr */) {
7129 if (fromKls == nullptr) {
7130 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7131 assert(tinst != nullptr, "obj is null");
7132 assert(tinst->is_loaded(), "obj is not loaded");
7133 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7134 fromKls = tinst->instance_klass();
7135 }
7136 else {
7137 assert(is_static, "only for static field access");
7138 }
7139 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7140 ciSymbol::make(fieldTypeString),
7141 is_static);
7142
7143 assert(field != nullptr, "undefined field");
7144 assert(!field->is_volatile(), "not defined for volatile fields");
7145
7146 if (is_static) {
7147 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7148 fromObj = makecon(tip);
7149 }
7150
7151 // Next code copied from Parse::do_get_xxx():
7152
7153 // Compute address and memory type.
7154 int offset = field->offset_in_bytes();
7155 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7156
7157 return adr;
7158 }
7159
7160 //------------------------------inline_aescrypt_Block-----------------------
7161 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7162 address stubAddr = nullptr;
7163 const char *stubName;
7164 assert(UseAES, "need AES instruction support");
7165
7166 switch(id) {
7167 case vmIntrinsics::_aescrypt_encryptBlock:
7168 stubAddr = StubRoutines::aescrypt_encryptBlock();
7169 stubName = "aescrypt_encryptBlock";
7170 break;
7171 case vmIntrinsics::_aescrypt_decryptBlock:
7172 stubAddr = StubRoutines::aescrypt_decryptBlock();
7173 stubName = "aescrypt_decryptBlock";
7174 break;
7175 default:
7176 break;
7177 }
7178 if (stubAddr == nullptr) return false;
7179
7180 Node* aescrypt_object = argument(0);
7181 Node* src = argument(1);
7182 Node* src_offset = argument(2);
7183 Node* dest = argument(3);
7184 Node* dest_offset = argument(4);
7185
7186 src = must_be_not_null(src, true);
7187 dest = must_be_not_null(dest, true);
7188
7189 // (1) src and dest are arrays.
7190 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7191 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7192 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7193 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7194
7195 // for the quick and dirty code we will skip all the checks.
7196 // we are just trying to get the call to be generated.
7197 Node* src_start = src;
7198 Node* dest_start = dest;
7199 if (src_offset != nullptr || dest_offset != nullptr) {
7200 assert(src_offset != nullptr && dest_offset != nullptr, "");
7201 src_start = array_element_address(src, src_offset, T_BYTE);
7202 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7203 }
7204
7205 // now need to get the start of its expanded key array
7206 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7207 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7208 if (k_start == nullptr) return false;
7209
7210 // Call the stub.
7211 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7212 stubAddr, stubName, TypePtr::BOTTOM,
7213 src_start, dest_start, k_start);
7214
7215 return true;
7216 }
7217
7218 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7219 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7220 address stubAddr = nullptr;
7221 const char *stubName = nullptr;
7222
7223 assert(UseAES, "need AES instruction support");
7224
7225 switch(id) {
7226 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7227 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7228 stubName = "cipherBlockChaining_encryptAESCrypt";
7229 break;
7230 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7231 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7232 stubName = "cipherBlockChaining_decryptAESCrypt";
7233 break;
7234 default:
7235 break;
7236 }
7237 if (stubAddr == nullptr) return false;
7238
7239 Node* cipherBlockChaining_object = argument(0);
7240 Node* src = argument(1);
7241 Node* src_offset = argument(2);
7242 Node* len = argument(3);
7243 Node* dest = argument(4);
7244 Node* dest_offset = argument(5);
7245
7246 src = must_be_not_null(src, false);
7247 dest = must_be_not_null(dest, false);
7248
7249 // (1) src and dest are arrays.
7250 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7251 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7252 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7253 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7254
7255 // checks are the responsibility of the caller
7256 Node* src_start = src;
7257 Node* dest_start = dest;
7258 if (src_offset != nullptr || dest_offset != nullptr) {
7259 assert(src_offset != nullptr && dest_offset != nullptr, "");
7260 src_start = array_element_address(src, src_offset, T_BYTE);
7261 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7262 }
7263
7264 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7265 // (because of the predicated logic executed earlier).
7266 // so we cast it here safely.
7267 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7268
7269 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7270 if (embeddedCipherObj == nullptr) return false;
7271
7272 // cast it to what we know it will be at runtime
7273 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7274 assert(tinst != nullptr, "CBC obj is null");
7275 assert(tinst->is_loaded(), "CBC obj is not loaded");
7276 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7277 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7278
7279 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7280 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7281 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7282 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7283 aescrypt_object = _gvn.transform(aescrypt_object);
7284
7285 // we need to get the start of the aescrypt_object's expanded key array
7286 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7287 if (k_start == nullptr) return false;
7288
7289 // similarly, get the start address of the r vector
7290 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7291 if (objRvec == nullptr) return false;
7292 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7293
7294 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7295 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7296 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7297 stubAddr, stubName, TypePtr::BOTTOM,
7298 src_start, dest_start, k_start, r_start, len);
7299
7300 // return cipher length (int)
7301 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7302 set_result(retvalue);
7303 return true;
7304 }
7305
7306 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7307 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7308 address stubAddr = nullptr;
7309 const char *stubName = nullptr;
7310
7311 assert(UseAES, "need AES instruction support");
7312
7313 switch (id) {
7314 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7315 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7316 stubName = "electronicCodeBook_encryptAESCrypt";
7317 break;
7318 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7319 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7320 stubName = "electronicCodeBook_decryptAESCrypt";
7321 break;
7322 default:
7323 break;
7324 }
7325
7326 if (stubAddr == nullptr) return false;
7327
7328 Node* electronicCodeBook_object = argument(0);
7329 Node* src = argument(1);
7330 Node* src_offset = argument(2);
7331 Node* len = argument(3);
7332 Node* dest = argument(4);
7333 Node* dest_offset = argument(5);
7334
7335 // (1) src and dest are arrays.
7336 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7337 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7338 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7339 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7340
7341 // checks are the responsibility of the caller
7342 Node* src_start = src;
7343 Node* dest_start = dest;
7344 if (src_offset != nullptr || dest_offset != nullptr) {
7345 assert(src_offset != nullptr && dest_offset != nullptr, "");
7346 src_start = array_element_address(src, src_offset, T_BYTE);
7347 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7348 }
7349
7350 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7351 // (because of the predicated logic executed earlier).
7352 // so we cast it here safely.
7353 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7354
7355 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7356 if (embeddedCipherObj == nullptr) return false;
7357
7358 // cast it to what we know it will be at runtime
7359 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7360 assert(tinst != nullptr, "ECB obj is null");
7361 assert(tinst->is_loaded(), "ECB obj is not loaded");
7362 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7363 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7364
7365 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7366 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7367 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7368 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7369 aescrypt_object = _gvn.transform(aescrypt_object);
7370
7371 // we need to get the start of the aescrypt_object's expanded key array
7372 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7373 if (k_start == nullptr) return false;
7374
7375 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7376 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7377 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7378 stubAddr, stubName, TypePtr::BOTTOM,
7379 src_start, dest_start, k_start, len);
7380
7381 // return cipher length (int)
7382 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7383 set_result(retvalue);
7384 return true;
7385 }
7386
7387 //------------------------------inline_counterMode_AESCrypt-----------------------
7388 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7389 assert(UseAES, "need AES instruction support");
7390 if (!UseAESCTRIntrinsics) return false;
7391
7392 address stubAddr = nullptr;
7393 const char *stubName = nullptr;
7394 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7395 stubAddr = StubRoutines::counterMode_AESCrypt();
7396 stubName = "counterMode_AESCrypt";
7397 }
7398 if (stubAddr == nullptr) return false;
7399
7400 Node* counterMode_object = argument(0);
7401 Node* src = argument(1);
7402 Node* src_offset = argument(2);
7403 Node* len = argument(3);
7404 Node* dest = argument(4);
7405 Node* dest_offset = argument(5);
7406
7407 // (1) src and dest are arrays.
7408 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7409 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7410 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7411 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7412
7413 // checks are the responsibility of the caller
7414 Node* src_start = src;
7415 Node* dest_start = dest;
7416 if (src_offset != nullptr || dest_offset != nullptr) {
7417 assert(src_offset != nullptr && dest_offset != nullptr, "");
7418 src_start = array_element_address(src, src_offset, T_BYTE);
7419 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7420 }
7421
7422 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7423 // (because of the predicated logic executed earlier).
7424 // so we cast it here safely.
7425 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7426 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7427 if (embeddedCipherObj == nullptr) return false;
7428 // cast it to what we know it will be at runtime
7429 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7430 assert(tinst != nullptr, "CTR obj is null");
7431 assert(tinst->is_loaded(), "CTR obj is not loaded");
7432 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7433 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7434 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7435 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7436 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7437 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7438 aescrypt_object = _gvn.transform(aescrypt_object);
7439 // we need to get the start of the aescrypt_object's expanded key array
7440 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7441 if (k_start == nullptr) return false;
7442 // similarly, get the start address of the r vector
7443 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7444 if (obj_counter == nullptr) return false;
7445 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7446
7447 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7448 if (saved_encCounter == nullptr) return false;
7449 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7450 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7451
7452 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7453 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7454 OptoRuntime::counterMode_aescrypt_Type(),
7455 stubAddr, stubName, TypePtr::BOTTOM,
7456 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7457
7458 // return cipher length (int)
7459 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7460 set_result(retvalue);
7461 return true;
7462 }
7463
7464 //------------------------------get_key_start_from_aescrypt_object-----------------------
7465 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7466 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7467 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7468 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7469 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7470 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7471 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7472 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7473 if (objSessionK == nullptr) {
7474 return (Node *) nullptr;
7475 }
7476 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7477 #else
7478 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7479 #endif // PPC64
7480 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7481 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7482
7483 // now have the array, need to get the start address of the K array
7484 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7485 return k_start;
7486 }
7487
7488 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7489 // Return node representing slow path of predicate check.
7490 // the pseudo code we want to emulate with this predicate is:
7491 // for encryption:
7492 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7493 // for decryption:
7494 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7495 // note cipher==plain is more conservative than the original java code but that's OK
7496 //
7497 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7498 // The receiver was checked for null already.
7499 Node* objCBC = argument(0);
7500
7501 Node* src = argument(1);
7502 Node* dest = argument(4);
7503
7504 // Load embeddedCipher field of CipherBlockChaining object.
7505 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7506
7507 // get AESCrypt klass for instanceOf check
7508 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7509 // will have same classloader as CipherBlockChaining object
7510 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7511 assert(tinst != nullptr, "CBCobj is null");
7512 assert(tinst->is_loaded(), "CBCobj is not loaded");
7513
7514 // we want to do an instanceof comparison against the AESCrypt class
7515 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7516 if (!klass_AESCrypt->is_loaded()) {
7517 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7518 Node* ctrl = control();
7519 set_control(top()); // no regular fast path
7520 return ctrl;
7521 }
7522
7523 src = must_be_not_null(src, true);
7524 dest = must_be_not_null(dest, true);
7525
7526 // Resolve oops to stable for CmpP below.
7527 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7528
7529 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7530 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7531 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7532
7533 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7534
7535 // for encryption, we are done
7536 if (!decrypting)
7537 return instof_false; // even if it is null
7538
7539 // for decryption, we need to add a further check to avoid
7540 // taking the intrinsic path when cipher and plain are the same
7541 // see the original java code for why.
7542 RegionNode* region = new RegionNode(3);
7543 region->init_req(1, instof_false);
7544
7545 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7546 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7547 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7548 region->init_req(2, src_dest_conjoint);
7549
7550 record_for_igvn(region);
7551 return _gvn.transform(region);
7552 }
7553
7554 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7555 // Return node representing slow path of predicate check.
7556 // the pseudo code we want to emulate with this predicate is:
7557 // for encryption:
7558 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7559 // for decryption:
7560 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7561 // note cipher==plain is more conservative than the original java code but that's OK
7562 //
7563 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7564 // The receiver was checked for null already.
7565 Node* objECB = argument(0);
7566
7567 // Load embeddedCipher field of ElectronicCodeBook object.
7568 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7569
7570 // get AESCrypt klass for instanceOf check
7571 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7572 // will have same classloader as ElectronicCodeBook object
7573 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7574 assert(tinst != nullptr, "ECBobj is null");
7575 assert(tinst->is_loaded(), "ECBobj is not loaded");
7576
7577 // we want to do an instanceof comparison against the AESCrypt class
7578 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7579 if (!klass_AESCrypt->is_loaded()) {
7580 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7581 Node* ctrl = control();
7582 set_control(top()); // no regular fast path
7583 return ctrl;
7584 }
7585 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7586
7587 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7588 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7589 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7590
7591 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7592
7593 // for encryption, we are done
7594 if (!decrypting)
7595 return instof_false; // even if it is null
7596
7597 // for decryption, we need to add a further check to avoid
7598 // taking the intrinsic path when cipher and plain are the same
7599 // see the original java code for why.
7600 RegionNode* region = new RegionNode(3);
7601 region->init_req(1, instof_false);
7602 Node* src = argument(1);
7603 Node* dest = argument(4);
7604 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7605 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7606 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7607 region->init_req(2, src_dest_conjoint);
7608
7609 record_for_igvn(region);
7610 return _gvn.transform(region);
7611 }
7612
7613 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7614 // Return node representing slow path of predicate check.
7615 // the pseudo code we want to emulate with this predicate is:
7616 // for encryption:
7617 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7618 // for decryption:
7619 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7620 // note cipher==plain is more conservative than the original java code but that's OK
7621 //
7622
7623 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7624 // The receiver was checked for null already.
7625 Node* objCTR = argument(0);
7626
7627 // Load embeddedCipher field of CipherBlockChaining object.
7628 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7629
7630 // get AESCrypt klass for instanceOf check
7631 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7632 // will have same classloader as CipherBlockChaining object
7633 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7634 assert(tinst != nullptr, "CTRobj is null");
7635 assert(tinst->is_loaded(), "CTRobj is not loaded");
7636
7637 // we want to do an instanceof comparison against the AESCrypt class
7638 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7639 if (!klass_AESCrypt->is_loaded()) {
7640 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7641 Node* ctrl = control();
7642 set_control(top()); // no regular fast path
7643 return ctrl;
7644 }
7645
7646 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7647 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7648 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7649 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7650 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7651
7652 return instof_false; // even if it is null
7653 }
7654
7655 //------------------------------inline_ghash_processBlocks
7656 bool LibraryCallKit::inline_ghash_processBlocks() {
7657 address stubAddr;
7658 const char *stubName;
7659 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7660
7661 stubAddr = StubRoutines::ghash_processBlocks();
7662 stubName = "ghash_processBlocks";
7663
7664 Node* data = argument(0);
7665 Node* offset = argument(1);
7666 Node* len = argument(2);
7667 Node* state = argument(3);
7668 Node* subkeyH = argument(4);
7669
7670 state = must_be_not_null(state, true);
7671 subkeyH = must_be_not_null(subkeyH, true);
7672 data = must_be_not_null(data, true);
7673
7674 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7675 assert(state_start, "state is null");
7676 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7677 assert(subkeyH_start, "subkeyH is null");
7678 Node* data_start = array_element_address(data, offset, T_BYTE);
7679 assert(data_start, "data is null");
7680
7681 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7682 OptoRuntime::ghash_processBlocks_Type(),
7683 stubAddr, stubName, TypePtr::BOTTOM,
7684 state_start, subkeyH_start, data_start, len);
7685 return true;
7686 }
7687
7688 //------------------------------inline_chacha20Block
7689 bool LibraryCallKit::inline_chacha20Block() {
7690 address stubAddr;
7691 const char *stubName;
7692 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7693
7694 stubAddr = StubRoutines::chacha20Block();
7695 stubName = "chacha20Block";
7696
7697 Node* state = argument(0);
7698 Node* result = argument(1);
7699
7700 state = must_be_not_null(state, true);
7701 result = must_be_not_null(result, true);
7702
7703 Node* state_start = array_element_address(state, intcon(0), T_INT);
7704 assert(state_start, "state is null");
7705 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7706 assert(result_start, "result is null");
7707
7708 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7709 OptoRuntime::chacha20Block_Type(),
7710 stubAddr, stubName, TypePtr::BOTTOM,
7711 state_start, result_start);
7712 // return key stream length (int)
7713 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7714 set_result(retvalue);
7715 return true;
7716 }
7717
7718 //------------------------------inline_kyberNtt
7719 bool LibraryCallKit::inline_kyberNtt() {
7720 address stubAddr;
7721 const char *stubName;
7722 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7723 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7724
7725 stubAddr = StubRoutines::kyberNtt();
7726 stubName = "kyberNtt";
7727 if (!stubAddr) return false;
7728
7729 Node* coeffs = argument(0);
7730 Node* ntt_zetas = argument(1);
7731
7732 coeffs = must_be_not_null(coeffs, true);
7733 ntt_zetas = must_be_not_null(ntt_zetas, true);
7734
7735 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7736 assert(coeffs_start, "coeffs is null");
7737 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7738 assert(ntt_zetas_start, "ntt_zetas is null");
7739 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7740 OptoRuntime::kyberNtt_Type(),
7741 stubAddr, stubName, TypePtr::BOTTOM,
7742 coeffs_start, ntt_zetas_start);
7743 // return an int
7744 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7745 set_result(retvalue);
7746 return true;
7747 }
7748
7749 //------------------------------inline_kyberInverseNtt
7750 bool LibraryCallKit::inline_kyberInverseNtt() {
7751 address stubAddr;
7752 const char *stubName;
7753 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7754 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7755
7756 stubAddr = StubRoutines::kyberInverseNtt();
7757 stubName = "kyberInverseNtt";
7758 if (!stubAddr) return false;
7759
7760 Node* coeffs = argument(0);
7761 Node* zetas = argument(1);
7762
7763 coeffs = must_be_not_null(coeffs, true);
7764 zetas = must_be_not_null(zetas, true);
7765
7766 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7767 assert(coeffs_start, "coeffs is null");
7768 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7769 assert(zetas_start, "inverseNtt_zetas is null");
7770 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7771 OptoRuntime::kyberInverseNtt_Type(),
7772 stubAddr, stubName, TypePtr::BOTTOM,
7773 coeffs_start, zetas_start);
7774
7775 // return an int
7776 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7777 set_result(retvalue);
7778 return true;
7779 }
7780
7781 //------------------------------inline_kyberNttMult
7782 bool LibraryCallKit::inline_kyberNttMult() {
7783 address stubAddr;
7784 const char *stubName;
7785 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7786 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7787
7788 stubAddr = StubRoutines::kyberNttMult();
7789 stubName = "kyberNttMult";
7790 if (!stubAddr) return false;
7791
7792 Node* result = argument(0);
7793 Node* ntta = argument(1);
7794 Node* nttb = argument(2);
7795 Node* zetas = argument(3);
7796
7797 result = must_be_not_null(result, true);
7798 ntta = must_be_not_null(ntta, true);
7799 nttb = must_be_not_null(nttb, true);
7800 zetas = must_be_not_null(zetas, true);
7801
7802 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7803 assert(result_start, "result is null");
7804 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7805 assert(ntta_start, "ntta is null");
7806 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7807 assert(nttb_start, "nttb is null");
7808 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7809 assert(zetas_start, "nttMult_zetas is null");
7810 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7811 OptoRuntime::kyberNttMult_Type(),
7812 stubAddr, stubName, TypePtr::BOTTOM,
7813 result_start, ntta_start, nttb_start,
7814 zetas_start);
7815
7816 // return an int
7817 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7818 set_result(retvalue);
7819
7820 return true;
7821 }
7822
7823 //------------------------------inline_kyberAddPoly_2
7824 bool LibraryCallKit::inline_kyberAddPoly_2() {
7825 address stubAddr;
7826 const char *stubName;
7827 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7828 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7829
7830 stubAddr = StubRoutines::kyberAddPoly_2();
7831 stubName = "kyberAddPoly_2";
7832 if (!stubAddr) return false;
7833
7834 Node* result = argument(0);
7835 Node* a = argument(1);
7836 Node* b = argument(2);
7837
7838 result = must_be_not_null(result, true);
7839 a = must_be_not_null(a, true);
7840 b = must_be_not_null(b, true);
7841
7842 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7843 assert(result_start, "result is null");
7844 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7845 assert(a_start, "a is null");
7846 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7847 assert(b_start, "b is null");
7848 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7849 OptoRuntime::kyberAddPoly_2_Type(),
7850 stubAddr, stubName, TypePtr::BOTTOM,
7851 result_start, a_start, b_start);
7852 // return an int
7853 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7854 set_result(retvalue);
7855 return true;
7856 }
7857
7858 //------------------------------inline_kyberAddPoly_3
7859 bool LibraryCallKit::inline_kyberAddPoly_3() {
7860 address stubAddr;
7861 const char *stubName;
7862 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7863 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7864
7865 stubAddr = StubRoutines::kyberAddPoly_3();
7866 stubName = "kyberAddPoly_3";
7867 if (!stubAddr) return false;
7868
7869 Node* result = argument(0);
7870 Node* a = argument(1);
7871 Node* b = argument(2);
7872 Node* c = argument(3);
7873
7874 result = must_be_not_null(result, true);
7875 a = must_be_not_null(a, true);
7876 b = must_be_not_null(b, true);
7877 c = must_be_not_null(c, true);
7878
7879 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7880 assert(result_start, "result is null");
7881 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7882 assert(a_start, "a is null");
7883 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7884 assert(b_start, "b is null");
7885 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7886 assert(c_start, "c is null");
7887 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7888 OptoRuntime::kyberAddPoly_3_Type(),
7889 stubAddr, stubName, TypePtr::BOTTOM,
7890 result_start, a_start, b_start, c_start);
7891 // return an int
7892 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7893 set_result(retvalue);
7894 return true;
7895 }
7896
7897 //------------------------------inline_kyber12To16
7898 bool LibraryCallKit::inline_kyber12To16() {
7899 address stubAddr;
7900 const char *stubName;
7901 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7902 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7903
7904 stubAddr = StubRoutines::kyber12To16();
7905 stubName = "kyber12To16";
7906 if (!stubAddr) return false;
7907
7908 Node* condensed = argument(0);
7909 Node* condensedOffs = argument(1);
7910 Node* parsed = argument(2);
7911 Node* parsedLength = argument(3);
7912
7913 condensed = must_be_not_null(condensed, true);
7914 parsed = must_be_not_null(parsed, true);
7915
7916 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7917 assert(condensed_start, "condensed is null");
7918 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7919 assert(parsed_start, "parsed is null");
7920 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7921 OptoRuntime::kyber12To16_Type(),
7922 stubAddr, stubName, TypePtr::BOTTOM,
7923 condensed_start, condensedOffs, parsed_start, parsedLength);
7924 // return an int
7925 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7926 set_result(retvalue);
7927 return true;
7928
7929 }
7930
7931 //------------------------------inline_kyberBarrettReduce
7932 bool LibraryCallKit::inline_kyberBarrettReduce() {
7933 address stubAddr;
7934 const char *stubName;
7935 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7936 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7937
7938 stubAddr = StubRoutines::kyberBarrettReduce();
7939 stubName = "kyberBarrettReduce";
7940 if (!stubAddr) return false;
7941
7942 Node* coeffs = argument(0);
7943
7944 coeffs = must_be_not_null(coeffs, true);
7945
7946 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7947 assert(coeffs_start, "coeffs is null");
7948 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7949 OptoRuntime::kyberBarrettReduce_Type(),
7950 stubAddr, stubName, TypePtr::BOTTOM,
7951 coeffs_start);
7952 // return an int
7953 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7954 set_result(retvalue);
7955 return true;
7956 }
7957
7958 //------------------------------inline_dilithiumAlmostNtt
7959 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7960 address stubAddr;
7961 const char *stubName;
7962 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7963 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7964
7965 stubAddr = StubRoutines::dilithiumAlmostNtt();
7966 stubName = "dilithiumAlmostNtt";
7967 if (!stubAddr) return false;
7968
7969 Node* coeffs = argument(0);
7970 Node* ntt_zetas = argument(1);
7971
7972 coeffs = must_be_not_null(coeffs, true);
7973 ntt_zetas = must_be_not_null(ntt_zetas, true);
7974
7975 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7976 assert(coeffs_start, "coeffs is null");
7977 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7978 assert(ntt_zetas_start, "ntt_zetas is null");
7979 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7980 OptoRuntime::dilithiumAlmostNtt_Type(),
7981 stubAddr, stubName, TypePtr::BOTTOM,
7982 coeffs_start, ntt_zetas_start);
7983 // return an int
7984 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7985 set_result(retvalue);
7986 return true;
7987 }
7988
7989 //------------------------------inline_dilithiumAlmostInverseNtt
7990 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7991 address stubAddr;
7992 const char *stubName;
7993 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7994 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7995
7996 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7997 stubName = "dilithiumAlmostInverseNtt";
7998 if (!stubAddr) return false;
7999
8000 Node* coeffs = argument(0);
8001 Node* zetas = argument(1);
8002
8003 coeffs = must_be_not_null(coeffs, true);
8004 zetas = must_be_not_null(zetas, true);
8005
8006 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8007 assert(coeffs_start, "coeffs is null");
8008 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8009 assert(zetas_start, "inverseNtt_zetas is null");
8010 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8011 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8012 stubAddr, stubName, TypePtr::BOTTOM,
8013 coeffs_start, zetas_start);
8014 // return an int
8015 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8016 set_result(retvalue);
8017 return true;
8018 }
8019
8020 //------------------------------inline_dilithiumNttMult
8021 bool LibraryCallKit::inline_dilithiumNttMult() {
8022 address stubAddr;
8023 const char *stubName;
8024 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8025 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8026
8027 stubAddr = StubRoutines::dilithiumNttMult();
8028 stubName = "dilithiumNttMult";
8029 if (!stubAddr) return false;
8030
8031 Node* result = argument(0);
8032 Node* ntta = argument(1);
8033 Node* nttb = argument(2);
8034 Node* zetas = argument(3);
8035
8036 result = must_be_not_null(result, true);
8037 ntta = must_be_not_null(ntta, true);
8038 nttb = must_be_not_null(nttb, true);
8039 zetas = must_be_not_null(zetas, true);
8040
8041 Node* result_start = array_element_address(result, intcon(0), T_INT);
8042 assert(result_start, "result is null");
8043 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8044 assert(ntta_start, "ntta is null");
8045 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8046 assert(nttb_start, "nttb is null");
8047 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8048 OptoRuntime::dilithiumNttMult_Type(),
8049 stubAddr, stubName, TypePtr::BOTTOM,
8050 result_start, ntta_start, nttb_start);
8051
8052 // return an int
8053 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8054 set_result(retvalue);
8055
8056 return true;
8057 }
8058
8059 //------------------------------inline_dilithiumMontMulByConstant
8060 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8061 address stubAddr;
8062 const char *stubName;
8063 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8064 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8065
8066 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8067 stubName = "dilithiumMontMulByConstant";
8068 if (!stubAddr) return false;
8069
8070 Node* coeffs = argument(0);
8071 Node* constant = argument(1);
8072
8073 coeffs = must_be_not_null(coeffs, true);
8074
8075 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8076 assert(coeffs_start, "coeffs is null");
8077 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8078 OptoRuntime::dilithiumMontMulByConstant_Type(),
8079 stubAddr, stubName, TypePtr::BOTTOM,
8080 coeffs_start, constant);
8081
8082 // return an int
8083 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8084 set_result(retvalue);
8085 return true;
8086 }
8087
8088
8089 //------------------------------inline_dilithiumDecomposePoly
8090 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8091 address stubAddr;
8092 const char *stubName;
8093 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8094 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8095
8096 stubAddr = StubRoutines::dilithiumDecomposePoly();
8097 stubName = "dilithiumDecomposePoly";
8098 if (!stubAddr) return false;
8099
8100 Node* input = argument(0);
8101 Node* lowPart = argument(1);
8102 Node* highPart = argument(2);
8103 Node* twoGamma2 = argument(3);
8104 Node* multiplier = argument(4);
8105
8106 input = must_be_not_null(input, true);
8107 lowPart = must_be_not_null(lowPart, true);
8108 highPart = must_be_not_null(highPart, true);
8109
8110 Node* input_start = array_element_address(input, intcon(0), T_INT);
8111 assert(input_start, "input is null");
8112 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8113 assert(lowPart_start, "lowPart is null");
8114 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8115 assert(highPart_start, "highPart is null");
8116
8117 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8118 OptoRuntime::dilithiumDecomposePoly_Type(),
8119 stubAddr, stubName, TypePtr::BOTTOM,
8120 input_start, lowPart_start, highPart_start,
8121 twoGamma2, multiplier);
8122
8123 // return an int
8124 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8125 set_result(retvalue);
8126 return true;
8127 }
8128
8129 bool LibraryCallKit::inline_base64_encodeBlock() {
8130 address stubAddr;
8131 const char *stubName;
8132 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8133 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8134 stubAddr = StubRoutines::base64_encodeBlock();
8135 stubName = "encodeBlock";
8136
8137 if (!stubAddr) return false;
8138 Node* base64obj = argument(0);
8139 Node* src = argument(1);
8140 Node* offset = argument(2);
8141 Node* len = argument(3);
8142 Node* dest = argument(4);
8143 Node* dp = argument(5);
8144 Node* isURL = argument(6);
8145
8146 src = must_be_not_null(src, true);
8147 dest = must_be_not_null(dest, true);
8148
8149 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8150 assert(src_start, "source array is null");
8151 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8152 assert(dest_start, "destination array is null");
8153
8154 Node* base64 = make_runtime_call(RC_LEAF,
8155 OptoRuntime::base64_encodeBlock_Type(),
8156 stubAddr, stubName, TypePtr::BOTTOM,
8157 src_start, offset, len, dest_start, dp, isURL);
8158 return true;
8159 }
8160
8161 bool LibraryCallKit::inline_base64_decodeBlock() {
8162 address stubAddr;
8163 const char *stubName;
8164 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8165 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8166 stubAddr = StubRoutines::base64_decodeBlock();
8167 stubName = "decodeBlock";
8168
8169 if (!stubAddr) return false;
8170 Node* base64obj = argument(0);
8171 Node* src = argument(1);
8172 Node* src_offset = argument(2);
8173 Node* len = argument(3);
8174 Node* dest = argument(4);
8175 Node* dest_offset = argument(5);
8176 Node* isURL = argument(6);
8177 Node* isMIME = argument(7);
8178
8179 src = must_be_not_null(src, true);
8180 dest = must_be_not_null(dest, true);
8181
8182 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8183 assert(src_start, "source array is null");
8184 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8185 assert(dest_start, "destination array is null");
8186
8187 Node* call = make_runtime_call(RC_LEAF,
8188 OptoRuntime::base64_decodeBlock_Type(),
8189 stubAddr, stubName, TypePtr::BOTTOM,
8190 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8191 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8192 set_result(result);
8193 return true;
8194 }
8195
8196 bool LibraryCallKit::inline_poly1305_processBlocks() {
8197 address stubAddr;
8198 const char *stubName;
8199 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8200 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8201 stubAddr = StubRoutines::poly1305_processBlocks();
8202 stubName = "poly1305_processBlocks";
8203
8204 if (!stubAddr) return false;
8205 null_check_receiver(); // null-check receiver
8206 if (stopped()) return true;
8207
8208 Node* input = argument(1);
8209 Node* input_offset = argument(2);
8210 Node* len = argument(3);
8211 Node* alimbs = argument(4);
8212 Node* rlimbs = argument(5);
8213
8214 input = must_be_not_null(input, true);
8215 alimbs = must_be_not_null(alimbs, true);
8216 rlimbs = must_be_not_null(rlimbs, true);
8217
8218 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8219 assert(input_start, "input array is null");
8220 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8221 assert(acc_start, "acc array is null");
8222 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8223 assert(r_start, "r array is null");
8224
8225 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8226 OptoRuntime::poly1305_processBlocks_Type(),
8227 stubAddr, stubName, TypePtr::BOTTOM,
8228 input_start, len, acc_start, r_start);
8229 return true;
8230 }
8231
8232 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8233 address stubAddr;
8234 const char *stubName;
8235 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8236 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8237 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8238 stubName = "intpoly_montgomeryMult_P256";
8239
8240 if (!stubAddr) return false;
8241 null_check_receiver(); // null-check receiver
8242 if (stopped()) return true;
8243
8244 Node* a = argument(1);
8245 Node* b = argument(2);
8246 Node* r = argument(3);
8247
8248 a = must_be_not_null(a, true);
8249 b = must_be_not_null(b, true);
8250 r = must_be_not_null(r, true);
8251
8252 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8253 assert(a_start, "a array is null");
8254 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8255 assert(b_start, "b array is null");
8256 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8257 assert(r_start, "r array is null");
8258
8259 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8260 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8261 stubAddr, stubName, TypePtr::BOTTOM,
8262 a_start, b_start, r_start);
8263 return true;
8264 }
8265
8266 bool LibraryCallKit::inline_intpoly_assign() {
8267 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8268 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8269 const char *stubName = "intpoly_assign";
8270 address stubAddr = StubRoutines::intpoly_assign();
8271 if (!stubAddr) return false;
8272
8273 Node* set = argument(0);
8274 Node* a = argument(1);
8275 Node* b = argument(2);
8276 Node* arr_length = load_array_length(a);
8277
8278 a = must_be_not_null(a, true);
8279 b = must_be_not_null(b, true);
8280
8281 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8282 assert(a_start, "a array is null");
8283 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8284 assert(b_start, "b array is null");
8285
8286 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8287 OptoRuntime::intpoly_assign_Type(),
8288 stubAddr, stubName, TypePtr::BOTTOM,
8289 set, a_start, b_start, arr_length);
8290 return true;
8291 }
8292
8293 //------------------------------inline_digestBase_implCompress-----------------------
8294 //
8295 // Calculate MD5 for single-block byte[] array.
8296 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8297 //
8298 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8299 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8300 //
8301 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8302 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8303 //
8304 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8305 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8306 //
8307 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8308 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8309 //
8310 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8311 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8312
8313 Node* digestBase_obj = argument(0);
8314 Node* src = argument(1); // type oop
8315 Node* ofs = argument(2); // type int
8316
8317 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8318 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8319 // failed array check
8320 return false;
8321 }
8322 // Figure out the size and type of the elements we will be copying.
8323 BasicType src_elem = src_type->elem()->array_element_basic_type();
8324 if (src_elem != T_BYTE) {
8325 return false;
8326 }
8327 // 'src_start' points to src array + offset
8328 src = must_be_not_null(src, true);
8329 Node* src_start = array_element_address(src, ofs, src_elem);
8330 Node* state = nullptr;
8331 Node* block_size = nullptr;
8332 address stubAddr;
8333 const char *stubName;
8334
8335 switch(id) {
8336 case vmIntrinsics::_md5_implCompress:
8337 assert(UseMD5Intrinsics, "need MD5 instruction support");
8338 state = get_state_from_digest_object(digestBase_obj, T_INT);
8339 stubAddr = StubRoutines::md5_implCompress();
8340 stubName = "md5_implCompress";
8341 break;
8342 case vmIntrinsics::_sha_implCompress:
8343 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8344 state = get_state_from_digest_object(digestBase_obj, T_INT);
8345 stubAddr = StubRoutines::sha1_implCompress();
8346 stubName = "sha1_implCompress";
8347 break;
8348 case vmIntrinsics::_sha2_implCompress:
8349 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8350 state = get_state_from_digest_object(digestBase_obj, T_INT);
8351 stubAddr = StubRoutines::sha256_implCompress();
8352 stubName = "sha256_implCompress";
8353 break;
8354 case vmIntrinsics::_sha5_implCompress:
8355 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8356 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8357 stubAddr = StubRoutines::sha512_implCompress();
8358 stubName = "sha512_implCompress";
8359 break;
8360 case vmIntrinsics::_sha3_implCompress:
8361 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8362 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8363 stubAddr = StubRoutines::sha3_implCompress();
8364 stubName = "sha3_implCompress";
8365 block_size = get_block_size_from_digest_object(digestBase_obj);
8366 if (block_size == nullptr) return false;
8367 break;
8368 default:
8369 fatal_unexpected_iid(id);
8370 return false;
8371 }
8372 if (state == nullptr) return false;
8373
8374 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8375 if (stubAddr == nullptr) return false;
8376
8377 // Call the stub.
8378 Node* call;
8379 if (block_size == nullptr) {
8380 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8381 stubAddr, stubName, TypePtr::BOTTOM,
8382 src_start, state);
8383 } else {
8384 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8385 stubAddr, stubName, TypePtr::BOTTOM,
8386 src_start, state, block_size);
8387 }
8388
8389 return true;
8390 }
8391
8392 //------------------------------inline_double_keccak
8393 bool LibraryCallKit::inline_double_keccak() {
8394 address stubAddr;
8395 const char *stubName;
8396 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8397 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8398
8399 stubAddr = StubRoutines::double_keccak();
8400 stubName = "double_keccak";
8401 if (!stubAddr) return false;
8402
8403 Node* status0 = argument(0);
8404 Node* status1 = argument(1);
8405
8406 status0 = must_be_not_null(status0, true);
8407 status1 = must_be_not_null(status1, true);
8408
8409 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8410 assert(status0_start, "status0 is null");
8411 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8412 assert(status1_start, "status1 is null");
8413 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8414 OptoRuntime::double_keccak_Type(),
8415 stubAddr, stubName, TypePtr::BOTTOM,
8416 status0_start, status1_start);
8417 // return an int
8418 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8419 set_result(retvalue);
8420 return true;
8421 }
8422
8423
8424 //------------------------------inline_digestBase_implCompressMB-----------------------
8425 //
8426 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8427 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8428 //
8429 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8430 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8431 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8432 assert((uint)predicate < 5, "sanity");
8433 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8434
8435 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8436 Node* src = argument(1); // byte[] array
8437 Node* ofs = argument(2); // type int
8438 Node* limit = argument(3); // type int
8439
8440 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8441 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8442 // failed array check
8443 return false;
8444 }
8445 // Figure out the size and type of the elements we will be copying.
8446 BasicType src_elem = src_type->elem()->array_element_basic_type();
8447 if (src_elem != T_BYTE) {
8448 return false;
8449 }
8450 // 'src_start' points to src array + offset
8451 src = must_be_not_null(src, false);
8452 Node* src_start = array_element_address(src, ofs, src_elem);
8453
8454 const char* klass_digestBase_name = nullptr;
8455 const char* stub_name = nullptr;
8456 address stub_addr = nullptr;
8457 BasicType elem_type = T_INT;
8458
8459 switch (predicate) {
8460 case 0:
8461 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8462 klass_digestBase_name = "sun/security/provider/MD5";
8463 stub_name = "md5_implCompressMB";
8464 stub_addr = StubRoutines::md5_implCompressMB();
8465 }
8466 break;
8467 case 1:
8468 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8469 klass_digestBase_name = "sun/security/provider/SHA";
8470 stub_name = "sha1_implCompressMB";
8471 stub_addr = StubRoutines::sha1_implCompressMB();
8472 }
8473 break;
8474 case 2:
8475 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8476 klass_digestBase_name = "sun/security/provider/SHA2";
8477 stub_name = "sha256_implCompressMB";
8478 stub_addr = StubRoutines::sha256_implCompressMB();
8479 }
8480 break;
8481 case 3:
8482 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8483 klass_digestBase_name = "sun/security/provider/SHA5";
8484 stub_name = "sha512_implCompressMB";
8485 stub_addr = StubRoutines::sha512_implCompressMB();
8486 elem_type = T_LONG;
8487 }
8488 break;
8489 case 4:
8490 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8491 klass_digestBase_name = "sun/security/provider/SHA3";
8492 stub_name = "sha3_implCompressMB";
8493 stub_addr = StubRoutines::sha3_implCompressMB();
8494 elem_type = T_LONG;
8495 }
8496 break;
8497 default:
8498 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8499 }
8500 if (klass_digestBase_name != nullptr) {
8501 assert(stub_addr != nullptr, "Stub is generated");
8502 if (stub_addr == nullptr) return false;
8503
8504 // get DigestBase klass to lookup for SHA klass
8505 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8506 assert(tinst != nullptr, "digestBase_obj is not instance???");
8507 assert(tinst->is_loaded(), "DigestBase is not loaded");
8508
8509 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8510 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8511 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8512 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8513 }
8514 return false;
8515 }
8516
8517 //------------------------------inline_digestBase_implCompressMB-----------------------
8518 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8519 BasicType elem_type, address stubAddr, const char *stubName,
8520 Node* src_start, Node* ofs, Node* limit) {
8521 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8522 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8523 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8524 digest_obj = _gvn.transform(digest_obj);
8525
8526 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8527 if (state == nullptr) return false;
8528
8529 Node* block_size = nullptr;
8530 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8531 block_size = get_block_size_from_digest_object(digest_obj);
8532 if (block_size == nullptr) return false;
8533 }
8534
8535 // Call the stub.
8536 Node* call;
8537 if (block_size == nullptr) {
8538 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8539 OptoRuntime::digestBase_implCompressMB_Type(false),
8540 stubAddr, stubName, TypePtr::BOTTOM,
8541 src_start, state, ofs, limit);
8542 } else {
8543 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8544 OptoRuntime::digestBase_implCompressMB_Type(true),
8545 stubAddr, stubName, TypePtr::BOTTOM,
8546 src_start, state, block_size, ofs, limit);
8547 }
8548
8549 // return ofs (int)
8550 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8551 set_result(result);
8552
8553 return true;
8554 }
8555
8556 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8557 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8558 assert(UseAES, "need AES instruction support");
8559 address stubAddr = nullptr;
8560 const char *stubName = nullptr;
8561 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8562 stubName = "galoisCounterMode_AESCrypt";
8563
8564 if (stubAddr == nullptr) return false;
8565
8566 Node* in = argument(0);
8567 Node* inOfs = argument(1);
8568 Node* len = argument(2);
8569 Node* ct = argument(3);
8570 Node* ctOfs = argument(4);
8571 Node* out = argument(5);
8572 Node* outOfs = argument(6);
8573 Node* gctr_object = argument(7);
8574 Node* ghash_object = argument(8);
8575
8576 // (1) in, ct and out are arrays.
8577 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8578 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8579 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8580 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8581 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8582 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8583
8584 // checks are the responsibility of the caller
8585 Node* in_start = in;
8586 Node* ct_start = ct;
8587 Node* out_start = out;
8588 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8589 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8590 in_start = array_element_address(in, inOfs, T_BYTE);
8591 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8592 out_start = array_element_address(out, outOfs, T_BYTE);
8593 }
8594
8595 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8596 // (because of the predicated logic executed earlier).
8597 // so we cast it here safely.
8598 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8599 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8600 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8601 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8602 Node* state = load_field_from_object(ghash_object, "state", "[J");
8603
8604 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8605 return false;
8606 }
8607 // cast it to what we know it will be at runtime
8608 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8609 assert(tinst != nullptr, "GCTR obj is null");
8610 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8611 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8612 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8613 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8614 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8615 const TypeOopPtr* xtype = aklass->as_instance_type();
8616 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8617 aescrypt_object = _gvn.transform(aescrypt_object);
8618 // we need to get the start of the aescrypt_object's expanded key array
8619 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8620 if (k_start == nullptr) return false;
8621 // similarly, get the start address of the r vector
8622 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8623 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8624 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8625
8626
8627 // Call the stub, passing params
8628 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8629 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8630 stubAddr, stubName, TypePtr::BOTTOM,
8631 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8632
8633 // return cipher length (int)
8634 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8635 set_result(retvalue);
8636
8637 return true;
8638 }
8639
8640 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8641 // Return node representing slow path of predicate check.
8642 // the pseudo code we want to emulate with this predicate is:
8643 // for encryption:
8644 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8645 // for decryption:
8646 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8647 // note cipher==plain is more conservative than the original java code but that's OK
8648 //
8649
8650 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8651 // The receiver was checked for null already.
8652 Node* objGCTR = argument(7);
8653 // Load embeddedCipher field of GCTR object.
8654 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8655 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8656
8657 // get AESCrypt klass for instanceOf check
8658 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8659 // will have same classloader as CipherBlockChaining object
8660 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8661 assert(tinst != nullptr, "GCTR obj is null");
8662 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8663
8664 // we want to do an instanceof comparison against the AESCrypt class
8665 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8666 if (!klass_AESCrypt->is_loaded()) {
8667 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8668 Node* ctrl = control();
8669 set_control(top()); // no regular fast path
8670 return ctrl;
8671 }
8672
8673 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8674 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8675 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8676 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8677 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8678
8679 return instof_false; // even if it is null
8680 }
8681
8682 //------------------------------get_state_from_digest_object-----------------------
8683 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8684 const char* state_type;
8685 switch (elem_type) {
8686 case T_BYTE: state_type = "[B"; break;
8687 case T_INT: state_type = "[I"; break;
8688 case T_LONG: state_type = "[J"; break;
8689 default: ShouldNotReachHere();
8690 }
8691 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8692 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8693 if (digest_state == nullptr) return (Node *) nullptr;
8694
8695 // now have the array, need to get the start address of the state array
8696 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8697 return state;
8698 }
8699
8700 //------------------------------get_block_size_from_sha3_object----------------------------------
8701 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8702 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8703 assert (block_size != nullptr, "sanity");
8704 return block_size;
8705 }
8706
8707 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8708 // Return node representing slow path of predicate check.
8709 // the pseudo code we want to emulate with this predicate is:
8710 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8711 //
8712 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8713 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8714 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8715 assert((uint)predicate < 5, "sanity");
8716
8717 // The receiver was checked for null already.
8718 Node* digestBaseObj = argument(0);
8719
8720 // get DigestBase klass for instanceOf check
8721 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8722 assert(tinst != nullptr, "digestBaseObj is null");
8723 assert(tinst->is_loaded(), "DigestBase is not loaded");
8724
8725 const char* klass_name = nullptr;
8726 switch (predicate) {
8727 case 0:
8728 if (UseMD5Intrinsics) {
8729 // we want to do an instanceof comparison against the MD5 class
8730 klass_name = "sun/security/provider/MD5";
8731 }
8732 break;
8733 case 1:
8734 if (UseSHA1Intrinsics) {
8735 // we want to do an instanceof comparison against the SHA class
8736 klass_name = "sun/security/provider/SHA";
8737 }
8738 break;
8739 case 2:
8740 if (UseSHA256Intrinsics) {
8741 // we want to do an instanceof comparison against the SHA2 class
8742 klass_name = "sun/security/provider/SHA2";
8743 }
8744 break;
8745 case 3:
8746 if (UseSHA512Intrinsics) {
8747 // we want to do an instanceof comparison against the SHA5 class
8748 klass_name = "sun/security/provider/SHA5";
8749 }
8750 break;
8751 case 4:
8752 if (UseSHA3Intrinsics) {
8753 // we want to do an instanceof comparison against the SHA3 class
8754 klass_name = "sun/security/provider/SHA3";
8755 }
8756 break;
8757 default:
8758 fatal("unknown SHA intrinsic predicate: %d", predicate);
8759 }
8760
8761 ciKlass* klass = nullptr;
8762 if (klass_name != nullptr) {
8763 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8764 }
8765 if ((klass == nullptr) || !klass->is_loaded()) {
8766 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8767 Node* ctrl = control();
8768 set_control(top()); // no intrinsic path
8769 return ctrl;
8770 }
8771 ciInstanceKlass* instklass = klass->as_instance_klass();
8772
8773 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8774 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8775 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8776 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8777
8778 return instof_false; // even if it is null
8779 }
8780
8781 //-------------inline_fma-----------------------------------
8782 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8783 Node *a = nullptr;
8784 Node *b = nullptr;
8785 Node *c = nullptr;
8786 Node* result = nullptr;
8787 switch (id) {
8788 case vmIntrinsics::_fmaD:
8789 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8790 // no receiver since it is static method
8791 a = argument(0);
8792 b = argument(2);
8793 c = argument(4);
8794 result = _gvn.transform(new FmaDNode(a, b, c));
8795 break;
8796 case vmIntrinsics::_fmaF:
8797 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8798 a = argument(0);
8799 b = argument(1);
8800 c = argument(2);
8801 result = _gvn.transform(new FmaFNode(a, b, c));
8802 break;
8803 default:
8804 fatal_unexpected_iid(id); break;
8805 }
8806 set_result(result);
8807 return true;
8808 }
8809
8810 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8811 // argument(0) is receiver
8812 Node* codePoint = argument(1);
8813 Node* n = nullptr;
8814
8815 switch (id) {
8816 case vmIntrinsics::_isDigit :
8817 n = new DigitNode(control(), codePoint);
8818 break;
8819 case vmIntrinsics::_isLowerCase :
8820 n = new LowerCaseNode(control(), codePoint);
8821 break;
8822 case vmIntrinsics::_isUpperCase :
8823 n = new UpperCaseNode(control(), codePoint);
8824 break;
8825 case vmIntrinsics::_isWhitespace :
8826 n = new WhitespaceNode(control(), codePoint);
8827 break;
8828 default:
8829 fatal_unexpected_iid(id);
8830 }
8831
8832 set_result(_gvn.transform(n));
8833 return true;
8834 }
8835
8836 bool LibraryCallKit::inline_profileBoolean() {
8837 Node* counts = argument(1);
8838 const TypeAryPtr* ary = nullptr;
8839 ciArray* aobj = nullptr;
8840 if (counts->is_Con()
8841 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8842 && (aobj = ary->const_oop()->as_array()) != nullptr
8843 && (aobj->length() == 2)) {
8844 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8845 jint false_cnt = aobj->element_value(0).as_int();
8846 jint true_cnt = aobj->element_value(1).as_int();
8847
8848 if (C->log() != nullptr) {
8849 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8850 false_cnt, true_cnt);
8851 }
8852
8853 if (false_cnt + true_cnt == 0) {
8854 // According to profile, never executed.
8855 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8856 Deoptimization::Action_reinterpret);
8857 return true;
8858 }
8859
8860 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8861 // is a number of each value occurrences.
8862 Node* result = argument(0);
8863 if (false_cnt == 0 || true_cnt == 0) {
8864 // According to profile, one value has been never seen.
8865 int expected_val = (false_cnt == 0) ? 1 : 0;
8866
8867 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8868 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8869
8870 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8871 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8872 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8873
8874 { // Slow path: uncommon trap for never seen value and then reexecute
8875 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8876 // the value has been seen at least once.
8877 PreserveJVMState pjvms(this);
8878 PreserveReexecuteState preexecs(this);
8879 jvms()->set_should_reexecute(true);
8880
8881 set_control(slow_path);
8882 set_i_o(i_o());
8883
8884 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8885 Deoptimization::Action_reinterpret);
8886 }
8887 // The guard for never seen value enables sharpening of the result and
8888 // returning a constant. It allows to eliminate branches on the same value
8889 // later on.
8890 set_control(fast_path);
8891 result = intcon(expected_val);
8892 }
8893 // Stop profiling.
8894 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8895 // By replacing method body with profile data (represented as ProfileBooleanNode
8896 // on IR level) we effectively disable profiling.
8897 // It enables full speed execution once optimized code is generated.
8898 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8899 C->record_for_igvn(profile);
8900 set_result(profile);
8901 return true;
8902 } else {
8903 // Continue profiling.
8904 // Profile data isn't available at the moment. So, execute method's bytecode version.
8905 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8906 // is compiled and counters aren't available since corresponding MethodHandle
8907 // isn't a compile-time constant.
8908 return false;
8909 }
8910 }
8911
8912 bool LibraryCallKit::inline_isCompileConstant() {
8913 Node* n = argument(0);
8914 set_result(n->is_Con() ? intcon(1) : intcon(0));
8915 return true;
8916 }
8917
8918 //------------------------------- inline_getObjectSize --------------------------------------
8919 //
8920 // Calculate the runtime size of the object/array.
8921 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8922 //
8923 bool LibraryCallKit::inline_getObjectSize() {
8924 Node* obj = argument(3);
8925 Node* klass_node = load_object_klass(obj);
8926
8927 jint layout_con = Klass::_lh_neutral_value;
8928 Node* layout_val = get_layout_helper(klass_node, layout_con);
8929 int layout_is_con = (layout_val == nullptr);
8930
8931 if (layout_is_con) {
8932 // Layout helper is constant, can figure out things at compile time.
8933
8934 if (Klass::layout_helper_is_instance(layout_con)) {
8935 // Instance case: layout_con contains the size itself.
8936 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8937 set_result(size);
8938 } else {
8939 // Array case: size is round(header + element_size*arraylength).
8940 // Since arraylength is different for every array instance, we have to
8941 // compute the whole thing at runtime.
8942
8943 Node* arr_length = load_array_length(obj);
8944
8945 int round_mask = MinObjAlignmentInBytes - 1;
8946 int hsize = Klass::layout_helper_header_size(layout_con);
8947 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8948
8949 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8950 round_mask = 0; // strength-reduce it if it goes away completely
8951 }
8952 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8953 Node* header_size = intcon(hsize + round_mask);
8954
8955 Node* lengthx = ConvI2X(arr_length);
8956 Node* headerx = ConvI2X(header_size);
8957
8958 Node* abody = lengthx;
8959 if (eshift != 0) {
8960 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8961 }
8962 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8963 if (round_mask != 0) {
8964 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8965 }
8966 size = ConvX2L(size);
8967 set_result(size);
8968 }
8969 } else {
8970 // Layout helper is not constant, need to test for array-ness at runtime.
8971
8972 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8973 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8974 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8975 record_for_igvn(result_reg);
8976
8977 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8978 if (array_ctl != nullptr) {
8979 // Array case: size is round(header + element_size*arraylength).
8980 // Since arraylength is different for every array instance, we have to
8981 // compute the whole thing at runtime.
8982
8983 PreserveJVMState pjvms(this);
8984 set_control(array_ctl);
8985 Node* arr_length = load_array_length(obj);
8986
8987 int round_mask = MinObjAlignmentInBytes - 1;
8988 Node* mask = intcon(round_mask);
8989
8990 Node* hss = intcon(Klass::_lh_header_size_shift);
8991 Node* hsm = intcon(Klass::_lh_header_size_mask);
8992 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8993 header_size = _gvn.transform(new AndINode(header_size, hsm));
8994 header_size = _gvn.transform(new AddINode(header_size, mask));
8995
8996 // There is no need to mask or shift this value.
8997 // The semantics of LShiftINode include an implicit mask to 0x1F.
8998 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8999 Node* elem_shift = layout_val;
9000
9001 Node* lengthx = ConvI2X(arr_length);
9002 Node* headerx = ConvI2X(header_size);
9003
9004 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9005 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9006 if (round_mask != 0) {
9007 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9008 }
9009 size = ConvX2L(size);
9010
9011 result_reg->init_req(_array_path, control());
9012 result_val->init_req(_array_path, size);
9013 }
9014
9015 if (!stopped()) {
9016 // Instance case: the layout helper gives us instance size almost directly,
9017 // but we need to mask out the _lh_instance_slow_path_bit.
9018 Node* size = ConvI2X(layout_val);
9019 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9020 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9021 size = _gvn.transform(new AndXNode(size, mask));
9022 size = ConvX2L(size);
9023
9024 result_reg->init_req(_instance_path, control());
9025 result_val->init_req(_instance_path, size);
9026 }
9027
9028 set_result(result_reg, result_val);
9029 }
9030
9031 return true;
9032 }
9033
9034 //------------------------------- inline_blackhole --------------------------------------
9035 //
9036 // Make sure all arguments to this node are alive.
9037 // This matches methods that were requested to be blackholed through compile commands.
9038 //
9039 bool LibraryCallKit::inline_blackhole() {
9040 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9041 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9042 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9043
9044 // Blackhole node pinches only the control, not memory. This allows
9045 // the blackhole to be pinned in the loop that computes blackholed
9046 // values, but have no other side effects, like breaking the optimizations
9047 // across the blackhole.
9048
9049 Node* bh = _gvn.transform(new BlackholeNode(control()));
9050 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9051
9052 // Bind call arguments as blackhole arguments to keep them alive
9053 uint nargs = callee()->arg_size();
9054 for (uint i = 0; i < nargs; i++) {
9055 bh->add_req(argument(i));
9056 }
9057
9058 return true;
9059 }
9060
9061 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9062 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9063 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9064 return nullptr; // box klass is not Float16
9065 }
9066
9067 // Null check; get notnull casted pointer
9068 Node* null_ctl = top();
9069 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9070 // If not_null_box is dead, only null-path is taken
9071 if (stopped()) {
9072 set_control(null_ctl);
9073 return nullptr;
9074 }
9075 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9076 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9077 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9078 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9079 }
9080
9081 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9082 PreserveReexecuteState preexecs(this);
9083 jvms()->set_should_reexecute(true);
9084
9085 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9086 Node* klass_node = makecon(klass_type);
9087 Node* box = new_instance(klass_node);
9088
9089 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9090 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9091
9092 Node* field_store = _gvn.transform(access_store_at(box,
9093 value_field,
9094 value_adr_type,
9095 value,
9096 TypeInt::SHORT,
9097 T_SHORT,
9098 IN_HEAP));
9099 set_memory(field_store, value_adr_type);
9100 return box;
9101 }
9102
9103 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9104 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9105 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9106 return false;
9107 }
9108
9109 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9110 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9111 return false;
9112 }
9113
9114 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9115 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9116 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9117 ciSymbols::short_signature(),
9118 false);
9119 assert(field != nullptr, "");
9120
9121 // Transformed nodes
9122 Node* fld1 = nullptr;
9123 Node* fld2 = nullptr;
9124 Node* fld3 = nullptr;
9125 switch(num_args) {
9126 case 3:
9127 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9128 if (fld3 == nullptr) {
9129 return false;
9130 }
9131 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9132 // fall-through
9133 case 2:
9134 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9135 if (fld2 == nullptr) {
9136 return false;
9137 }
9138 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9139 // fall-through
9140 case 1:
9141 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9142 if (fld1 == nullptr) {
9143 return false;
9144 }
9145 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9146 break;
9147 default: fatal("Unsupported number of arguments %d", num_args);
9148 }
9149
9150 Node* result = nullptr;
9151 switch (id) {
9152 // Unary operations
9153 case vmIntrinsics::_sqrt_float16:
9154 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9155 break;
9156 // Ternary operations
9157 case vmIntrinsics::_fma_float16:
9158 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9159 break;
9160 default:
9161 fatal_unexpected_iid(id);
9162 break;
9163 }
9164 result = _gvn.transform(new ReinterpretHF2SNode(result));
9165 set_result(box_fp16_value(float16_box_type, field, result));
9166 return true;
9167 }
9168