1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/mountUnmountDisabler.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "utilities/macros.hpp"
63 #include "utilities/powerOfTwo.hpp"
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return nullptr;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
85 if (is_available && is_virtual) {
86 is_available = vmIntrinsics::does_virtual_dispatch(id);
87 }
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return nullptr;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 return kit.transfer_exceptions_into_jvms();
135 }
136
137 // The intrinsic bailed out
138 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
139 assert(jvms->map() == kit.map(), "Out of sync JVM state");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dsinh:
250 case vmIntrinsics::_dtanh:
251 case vmIntrinsics::_dcbrt:
252 case vmIntrinsics::_dabs:
253 case vmIntrinsics::_fabs:
254 case vmIntrinsics::_iabs:
255 case vmIntrinsics::_labs:
256 case vmIntrinsics::_datan2:
257 case vmIntrinsics::_dsqrt:
258 case vmIntrinsics::_dsqrt_strict:
259 case vmIntrinsics::_dexp:
260 case vmIntrinsics::_dlog:
261 case vmIntrinsics::_dlog10:
262 case vmIntrinsics::_dpow:
263 case vmIntrinsics::_dcopySign:
264 case vmIntrinsics::_fcopySign:
265 case vmIntrinsics::_dsignum:
266 case vmIntrinsics::_roundF:
267 case vmIntrinsics::_roundD:
268 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
269
270 case vmIntrinsics::_notify:
271 case vmIntrinsics::_notifyAll:
272 return inline_notify(intrinsic_id());
273
274 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
275 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
276 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
277 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
278 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
279 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
280 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
281 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
282 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
283 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
284 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
285 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
286 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
287 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
288
289 case vmIntrinsics::_arraycopy: return inline_arraycopy();
290
291 case vmIntrinsics::_arraySort: return inline_array_sort();
292 case vmIntrinsics::_arrayPartition: return inline_array_partition();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309
310 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
311
312 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
313 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
314 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
315 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
316
317 case vmIntrinsics::_compressStringC:
318 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
319 case vmIntrinsics::_inflateStringC:
320 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
321
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
331
332 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
341
342 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
343 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
344 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
345 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
346 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
347 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
348 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
349 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
350 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
351
352 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
373 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
374 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
375 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
376 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
377 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
378 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
379 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
380 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
381
382 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
383 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
384 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
385 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
386 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
387 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
388 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
389 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
390 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
391
392 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
393 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
394 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
395 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
396 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
397 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
398 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
399 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
400 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
401
402 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
417
418 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
438
439 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
454
455 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
459
460 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
465
466 case vmIntrinsics::_loadFence:
467 case vmIntrinsics::_storeFence:
468 case vmIntrinsics::_storeStoreFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
474 case vmIntrinsics::_currentThread: return inline_native_currentThread();
475 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
476
477 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
478 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
479
480 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
481 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
482
483 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
484 "endFirstTransition", true);
485 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
486 "startFinalTransition", true);
487 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
488 "startTransition", false);
489 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
490 "endTransition", false);
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
508 case vmIntrinsics::_getLength: return inline_native_getLength();
509 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
510 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
511 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
512 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
513 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
514 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
515 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
516
517 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
518 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
519
520 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
521
522 case vmIntrinsics::_isInstance:
523 case vmIntrinsics::_isHidden:
524 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
535 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
536 case vmIntrinsics::_floatIsFinite:
537 case vmIntrinsics::_floatIsInfinite:
538 case vmIntrinsics::_doubleIsFinite:
539 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
540
541 case vmIntrinsics::_numberOfLeadingZeros_i:
542 case vmIntrinsics::_numberOfLeadingZeros_l:
543 case vmIntrinsics::_numberOfTrailingZeros_i:
544 case vmIntrinsics::_numberOfTrailingZeros_l:
545 case vmIntrinsics::_bitCount_i:
546 case vmIntrinsics::_bitCount_l:
547 case vmIntrinsics::_reverse_i:
548 case vmIntrinsics::_reverse_l:
549 case vmIntrinsics::_reverseBytes_i:
550 case vmIntrinsics::_reverseBytes_l:
551 case vmIntrinsics::_reverseBytes_s:
552 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
553
554 case vmIntrinsics::_compress_i:
555 case vmIntrinsics::_compress_l:
556 case vmIntrinsics::_expand_i:
557 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
558
559 case vmIntrinsics::_compareUnsigned_i:
560 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
561
562 case vmIntrinsics::_divideUnsigned_i:
563 case vmIntrinsics::_divideUnsigned_l:
564 case vmIntrinsics::_remainderUnsigned_i:
565 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
566
567 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
568
569 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
570 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
571 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
572 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
573 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
574
575 case vmIntrinsics::_Class_cast: return inline_Class_cast();
576
577 case vmIntrinsics::_aescrypt_encryptBlock:
578 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
579
580 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
581 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
582 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
583
584 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
585 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
586 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_counterMode_AESCrypt:
589 return inline_counterMode_AESCrypt(intrinsic_id());
590
591 case vmIntrinsics::_galoisCounterMode_AESCrypt:
592 return inline_galoisCounterMode_AESCrypt();
593
594 case vmIntrinsics::_md5_implCompress:
595 case vmIntrinsics::_sha_implCompress:
596 case vmIntrinsics::_sha2_implCompress:
597 case vmIntrinsics::_sha5_implCompress:
598 case vmIntrinsics::_sha3_implCompress:
599 return inline_digestBase_implCompress(intrinsic_id());
600 case vmIntrinsics::_double_keccak:
601 return inline_double_keccak();
602
603 case vmIntrinsics::_digestBase_implCompressMB:
604 return inline_digestBase_implCompressMB(predicate);
605
606 case vmIntrinsics::_multiplyToLen:
607 return inline_multiplyToLen();
608
609 case vmIntrinsics::_squareToLen:
610 return inline_squareToLen();
611
612 case vmIntrinsics::_mulAdd:
613 return inline_mulAdd();
614
615 case vmIntrinsics::_montgomeryMultiply:
616 return inline_montgomeryMultiply();
617 case vmIntrinsics::_montgomerySquare:
618 return inline_montgomerySquare();
619
620 case vmIntrinsics::_bigIntegerRightShiftWorker:
621 return inline_bigIntegerShift(true);
622 case vmIntrinsics::_bigIntegerLeftShiftWorker:
623 return inline_bigIntegerShift(false);
624
625 case vmIntrinsics::_vectorizedMismatch:
626 return inline_vectorizedMismatch();
627
628 case vmIntrinsics::_ghash_processBlocks:
629 return inline_ghash_processBlocks();
630 case vmIntrinsics::_chacha20Block:
631 return inline_chacha20Block();
632 case vmIntrinsics::_kyberNtt:
633 return inline_kyberNtt();
634 case vmIntrinsics::_kyberInverseNtt:
635 return inline_kyberInverseNtt();
636 case vmIntrinsics::_kyberNttMult:
637 return inline_kyberNttMult();
638 case vmIntrinsics::_kyberAddPoly_2:
639 return inline_kyberAddPoly_2();
640 case vmIntrinsics::_kyberAddPoly_3:
641 return inline_kyberAddPoly_3();
642 case vmIntrinsics::_kyber12To16:
643 return inline_kyber12To16();
644 case vmIntrinsics::_kyberBarrettReduce:
645 return inline_kyberBarrettReduce();
646 case vmIntrinsics::_dilithiumAlmostNtt:
647 return inline_dilithiumAlmostNtt();
648 case vmIntrinsics::_dilithiumAlmostInverseNtt:
649 return inline_dilithiumAlmostInverseNtt();
650 case vmIntrinsics::_dilithiumNttMult:
651 return inline_dilithiumNttMult();
652 case vmIntrinsics::_dilithiumMontMulByConstant:
653 return inline_dilithiumMontMulByConstant();
654 case vmIntrinsics::_dilithiumDecomposePoly:
655 return inline_dilithiumDecomposePoly();
656 case vmIntrinsics::_base64_encodeBlock:
657 return inline_base64_encodeBlock();
658 case vmIntrinsics::_base64_decodeBlock:
659 return inline_base64_decodeBlock();
660 case vmIntrinsics::_poly1305_processBlocks:
661 return inline_poly1305_processBlocks();
662 case vmIntrinsics::_intpoly_montgomeryMult_P256:
663 return inline_intpoly_montgomeryMult_P256();
664 case vmIntrinsics::_intpoly_assign:
665 return inline_intpoly_assign();
666 case vmIntrinsics::_encodeISOArray:
667 case vmIntrinsics::_encodeByteISOArray:
668 return inline_encodeISOArray(false);
669 case vmIntrinsics::_encodeAsciiArray:
670 return inline_encodeISOArray(true);
671
672 case vmIntrinsics::_updateCRC32:
673 return inline_updateCRC32();
674 case vmIntrinsics::_updateBytesCRC32:
675 return inline_updateBytesCRC32();
676 case vmIntrinsics::_updateByteBufferCRC32:
677 return inline_updateByteBufferCRC32();
678
679 case vmIntrinsics::_updateBytesCRC32C:
680 return inline_updateBytesCRC32C();
681 case vmIntrinsics::_updateDirectByteBufferCRC32C:
682 return inline_updateDirectByteBufferCRC32C();
683
684 case vmIntrinsics::_updateBytesAdler32:
685 return inline_updateBytesAdler32();
686 case vmIntrinsics::_updateByteBufferAdler32:
687 return inline_updateByteBufferAdler32();
688
689 case vmIntrinsics::_profileBoolean:
690 return inline_profileBoolean();
691 case vmIntrinsics::_isCompileConstant:
692 return inline_isCompileConstant();
693
694 case vmIntrinsics::_countPositives:
695 return inline_countPositives();
696
697 case vmIntrinsics::_fmaD:
698 case vmIntrinsics::_fmaF:
699 return inline_fma(intrinsic_id());
700
701 case vmIntrinsics::_isDigit:
702 case vmIntrinsics::_isLowerCase:
703 case vmIntrinsics::_isUpperCase:
704 case vmIntrinsics::_isWhitespace:
705 return inline_character_compare(intrinsic_id());
706
707 case vmIntrinsics::_min:
708 case vmIntrinsics::_max:
709 case vmIntrinsics::_min_strict:
710 case vmIntrinsics::_max_strict:
711 case vmIntrinsics::_minL:
712 case vmIntrinsics::_maxL:
713 case vmIntrinsics::_minF:
714 case vmIntrinsics::_maxF:
715 case vmIntrinsics::_minD:
716 case vmIntrinsics::_maxD:
717 case vmIntrinsics::_minF_strict:
718 case vmIntrinsics::_maxF_strict:
719 case vmIntrinsics::_minD_strict:
720 case vmIntrinsics::_maxD_strict:
721 return inline_min_max(intrinsic_id());
722
723 case vmIntrinsics::_VectorUnaryOp:
724 return inline_vector_nary_operation(1);
725 case vmIntrinsics::_VectorBinaryOp:
726 return inline_vector_nary_operation(2);
727 case vmIntrinsics::_VectorUnaryLibOp:
728 return inline_vector_call(1);
729 case vmIntrinsics::_VectorBinaryLibOp:
730 return inline_vector_call(2);
731 case vmIntrinsics::_VectorTernaryOp:
732 return inline_vector_nary_operation(3);
733 case vmIntrinsics::_VectorFromBitsCoerced:
734 return inline_vector_frombits_coerced();
735 case vmIntrinsics::_VectorMaskOp:
736 return inline_vector_mask_operation();
737 case vmIntrinsics::_VectorLoadOp:
738 return inline_vector_mem_operation(/*is_store=*/false);
739 case vmIntrinsics::_VectorLoadMaskedOp:
740 return inline_vector_mem_masked_operation(/*is_store*/false);
741 case vmIntrinsics::_VectorStoreOp:
742 return inline_vector_mem_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorStoreMaskedOp:
744 return inline_vector_mem_masked_operation(/*is_store=*/true);
745 case vmIntrinsics::_VectorGatherOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ false);
747 case vmIntrinsics::_VectorScatterOp:
748 return inline_vector_gather_scatter(/*is_scatter*/ true);
749 case vmIntrinsics::_VectorReductionCoerced:
750 return inline_vector_reduction();
751 case vmIntrinsics::_VectorTest:
752 return inline_vector_test();
753 case vmIntrinsics::_VectorBlend:
754 return inline_vector_blend();
755 case vmIntrinsics::_VectorRearrange:
756 return inline_vector_rearrange();
757 case vmIntrinsics::_VectorSelectFrom:
758 return inline_vector_select_from();
759 case vmIntrinsics::_VectorCompare:
760 return inline_vector_compare();
761 case vmIntrinsics::_VectorBroadcastInt:
762 return inline_vector_broadcast_int();
763 case vmIntrinsics::_VectorConvert:
764 return inline_vector_convert();
765 case vmIntrinsics::_VectorInsert:
766 return inline_vector_insert();
767 case vmIntrinsics::_VectorExtract:
768 return inline_vector_extract();
769 case vmIntrinsics::_VectorCompressExpand:
770 return inline_vector_compress_expand();
771 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
772 return inline_vector_select_from_two_vectors();
773 case vmIntrinsics::_IndexVector:
774 return inline_index_vector();
775 case vmIntrinsics::_IndexPartiallyInUpperRange:
776 return inline_index_partially_in_upper_range();
777
778 case vmIntrinsics::_getObjectSize:
779 return inline_getObjectSize();
780
781 case vmIntrinsics::_blackhole:
782 return inline_blackhole();
783
784 default:
785 // If you get here, it may be that someone has added a new intrinsic
786 // to the list in vmIntrinsics.hpp without implementing it here.
787 #ifndef PRODUCT
788 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
789 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
790 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
791 }
792 #endif
793 return false;
794 }
795 }
796
797 Node* LibraryCallKit::try_to_predicate(int predicate) {
798 if (!jvms()->has_method()) {
799 // Root JVMState has a null method.
800 assert(map()->memory()->Opcode() == Op_Parm, "");
801 // Insert the memory aliasing node
802 set_all_memory(reset_memory());
803 }
804 assert(merged_memory(), "");
805
806 switch (intrinsic_id()) {
807 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(false);
809 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt_predicate(true);
811 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(false);
813 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
814 return inline_electronicCodeBook_AESCrypt_predicate(true);
815 case vmIntrinsics::_counterMode_AESCrypt:
816 return inline_counterMode_AESCrypt_predicate();
817 case vmIntrinsics::_digestBase_implCompressMB:
818 return inline_digestBase_implCompressMB_predicate(predicate);
819 case vmIntrinsics::_galoisCounterMode_AESCrypt:
820 return inline_galoisCounterMode_AESCrypt_predicate();
821
822 default:
823 // If you get here, it may be that someone has added a new intrinsic
824 // to the list in vmIntrinsics.hpp without implementing it here.
825 #ifndef PRODUCT
826 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
827 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
828 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
829 }
830 #endif
831 Node* slow_ctl = control();
832 set_control(top()); // No fast path intrinsic
833 return slow_ctl;
834 }
835 }
836
837 //------------------------------set_result-------------------------------
838 // Helper function for finishing intrinsics.
839 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
840 record_for_igvn(region);
841 set_control(_gvn.transform(region));
842 set_result( _gvn.transform(value));
843 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
844 }
845
846 //------------------------------generate_guard---------------------------
847 // Helper function for generating guarded fast-slow graph structures.
848 // The given 'test', if true, guards a slow path. If the test fails
849 // then a fast path can be taken. (We generally hope it fails.)
850 // In all cases, GraphKit::control() is updated to the fast path.
851 // The returned value represents the control for the slow path.
852 // The return value is never 'top'; it is either a valid control
853 // or null if it is obvious that the slow path can never be taken.
854 // Also, if region and the slow control are not null, the slow edge
855 // is appended to the region.
856 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
857 if (stopped()) {
858 // Already short circuited.
859 return nullptr;
860 }
861
862 // Build an if node and its projections.
863 // If test is true we take the slow path, which we assume is uncommon.
864 if (_gvn.type(test) == TypeInt::ZERO) {
865 // The slow branch is never taken. No need to build this guard.
866 return nullptr;
867 }
868
869 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
870
871 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
872 if (if_slow == top()) {
873 // The slow branch is never taken. No need to build this guard.
874 return nullptr;
875 }
876
877 if (region != nullptr)
878 region->add_req(if_slow);
879
880 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
881 set_control(if_fast);
882
883 return if_slow;
884 }
885
886 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
887 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
888 }
889 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
890 return generate_guard(test, region, PROB_FAIR);
891 }
892
893 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
894 Node* *pos_index) {
895 if (stopped())
896 return nullptr; // already stopped
897 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
898 return nullptr; // index is already adequately typed
899 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
900 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
901 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
902 if (is_neg != nullptr && pos_index != nullptr) {
903 // Emulate effect of Parse::adjust_map_after_if.
904 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
905 (*pos_index) = _gvn.transform(ccast);
906 }
907 return is_neg;
908 }
909
910 // Make sure that 'position' is a valid limit index, in [0..length].
911 // There are two equivalent plans for checking this:
912 // A. (offset + copyLength) unsigned<= arrayLength
913 // B. offset <= (arrayLength - copyLength)
914 // We require that all of the values above, except for the sum and
915 // difference, are already known to be non-negative.
916 // Plan A is robust in the face of overflow, if offset and copyLength
917 // are both hugely positive.
918 //
919 // Plan B is less direct and intuitive, but it does not overflow at
920 // all, since the difference of two non-negatives is always
921 // representable. Whenever Java methods must perform the equivalent
922 // check they generally use Plan B instead of Plan A.
923 // For the moment we use Plan A.
924 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
925 Node* subseq_length,
926 Node* array_length,
927 RegionNode* region) {
928 if (stopped())
929 return nullptr; // already stopped
930 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
931 if (zero_offset && subseq_length->eqv_uncast(array_length))
932 return nullptr; // common case of whole-array copy
933 Node* last = subseq_length;
934 if (!zero_offset) // last += offset
935 last = _gvn.transform(new AddINode(last, offset));
936 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
937 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
938 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
939 return is_over;
940 }
941
942 // Emit range checks for the given String.value byte array
943 void LibraryCallKit::generate_string_range_check(Node* array,
944 Node* offset,
945 Node* count,
946 bool char_count,
947 bool halt_on_oob) {
948 if (stopped()) {
949 return; // already stopped
950 }
951 RegionNode* bailout = new RegionNode(1);
952 record_for_igvn(bailout);
953 if (char_count) {
954 // Convert char count to byte count
955 count = _gvn.transform(new LShiftINode(count, intcon(1)));
956 }
957
958 // Offset and count must not be negative
959 generate_negative_guard(offset, bailout);
960 generate_negative_guard(count, bailout);
961 // Offset + count must not exceed length of array
962 generate_limit_guard(offset, count, load_array_length(array), bailout);
963
964 if (bailout->req() > 1) {
965 if (halt_on_oob) {
966 bailout = _gvn.transform(bailout)->as_Region();
967 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
968 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
969 C->root()->add_req(halt);
970 } else {
971 PreserveJVMState pjvms(this);
972 set_control(_gvn.transform(bailout));
973 uncommon_trap(Deoptimization::Reason_intrinsic,
974 Deoptimization::Action_maybe_recompile);
975 }
976 }
977 }
978
979 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
980 bool is_immutable) {
981 ciKlass* thread_klass = env()->Thread_klass();
982 const Type* thread_type
983 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
984
985 Node* thread = _gvn.transform(new ThreadLocalNode());
986 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
987 tls_output = thread;
988
989 Node* thread_obj_handle
990 = (is_immutable
991 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
992 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
993 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
994 thread_obj_handle = _gvn.transform(thread_obj_handle);
995
996 DecoratorSet decorators = IN_NATIVE;
997 if (is_immutable) {
998 decorators |= C2_IMMUTABLE_MEMORY;
999 }
1000 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1001 }
1002
1003 //--------------------------generate_current_thread--------------------
1004 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1005 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1006 /*is_immutable*/false);
1007 }
1008
1009 //--------------------------generate_virtual_thread--------------------
1010 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1011 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1012 !C->method()->changes_current_thread());
1013 }
1014
1015 //------------------------------make_string_method_node------------------------
1016 // Helper method for String intrinsic functions. This version is called with
1017 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1018 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1019 // containing the lengths of str1 and str2.
1020 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1021 Node* result = nullptr;
1022 switch (opcode) {
1023 case Op_StrIndexOf:
1024 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1025 str1_start, cnt1, str2_start, cnt2, ae);
1026 break;
1027 case Op_StrComp:
1028 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1029 str1_start, cnt1, str2_start, cnt2, ae);
1030 break;
1031 case Op_StrEquals:
1032 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1033 // Use the constant length if there is one because optimized match rule may exist.
1034 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1035 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1036 break;
1037 default:
1038 ShouldNotReachHere();
1039 return nullptr;
1040 }
1041
1042 // All these intrinsics have checks.
1043 C->set_has_split_ifs(true); // Has chance for split-if optimization
1044 clear_upper_avx();
1045
1046 return _gvn.transform(result);
1047 }
1048
1049 //------------------------------inline_string_compareTo------------------------
1050 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1051 Node* arg1 = argument(0);
1052 Node* arg2 = argument(1);
1053
1054 arg1 = must_be_not_null(arg1, true);
1055 arg2 = must_be_not_null(arg2, true);
1056
1057 // Get start addr and length of first argument
1058 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1059 Node* arg1_cnt = load_array_length(arg1);
1060
1061 // Get start addr and length of second argument
1062 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1063 Node* arg2_cnt = load_array_length(arg2);
1064
1065 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1066 set_result(result);
1067 return true;
1068 }
1069
1070 //------------------------------inline_string_equals------------------------
1071 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1072 Node* arg1 = argument(0);
1073 Node* arg2 = argument(1);
1074
1075 // paths (plus control) merge
1076 RegionNode* region = new RegionNode(3);
1077 Node* phi = new PhiNode(region, TypeInt::BOOL);
1078
1079 if (!stopped()) {
1080
1081 arg1 = must_be_not_null(arg1, true);
1082 arg2 = must_be_not_null(arg2, true);
1083
1084 // Get start addr and length of first argument
1085 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1086 Node* arg1_cnt = load_array_length(arg1);
1087
1088 // Get start addr and length of second argument
1089 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1090 Node* arg2_cnt = load_array_length(arg2);
1091
1092 // Check for arg1_cnt != arg2_cnt
1093 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1094 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1095 Node* if_ne = generate_slow_guard(bol, nullptr);
1096 if (if_ne != nullptr) {
1097 phi->init_req(2, intcon(0));
1098 region->init_req(2, if_ne);
1099 }
1100
1101 // Check for count == 0 is done by assembler code for StrEquals.
1102
1103 if (!stopped()) {
1104 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1105 phi->init_req(1, equals);
1106 region->init_req(1, control());
1107 }
1108 }
1109
1110 // post merge
1111 set_control(_gvn.transform(region));
1112 record_for_igvn(region);
1113
1114 set_result(_gvn.transform(phi));
1115 return true;
1116 }
1117
1118 //------------------------------inline_array_equals----------------------------
1119 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1120 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1121 Node* arg1 = argument(0);
1122 Node* arg2 = argument(1);
1123
1124 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1125 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1126 clear_upper_avx();
1127
1128 return true;
1129 }
1130
1131
1132 //------------------------------inline_countPositives------------------------------
1133 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1134 bool LibraryCallKit::inline_countPositives() {
1135 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1136 return false;
1137 }
1138
1139 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1140 // no receiver since it is static method
1141 Node* ba = argument(0);
1142 Node* offset = argument(1);
1143 Node* len = argument(2);
1144
1145 if (VerifyIntrinsicChecks) {
1146 ba = must_be_not_null(ba, true);
1147 generate_string_range_check(ba, offset, len, false, true);
1148 if (stopped()) {
1149 return true;
1150 }
1151 }
1152
1153 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1154 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1155 set_result(_gvn.transform(result));
1156 clear_upper_avx();
1157 return true;
1158 }
1159
1160 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1161 Node* index = argument(0);
1162 Node* length = bt == T_INT ? argument(1) : argument(2);
1163 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1164 return false;
1165 }
1166
1167 // check that length is positive
1168 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1169 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1170
1171 {
1172 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1173 uncommon_trap(Deoptimization::Reason_intrinsic,
1174 Deoptimization::Action_make_not_entrant);
1175 }
1176
1177 if (stopped()) {
1178 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1179 return true;
1180 }
1181
1182 // length is now known positive, add a cast node to make this explicit
1183 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1184 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1185 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1186 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1187 casted_length = _gvn.transform(casted_length);
1188 replace_in_map(length, casted_length);
1189 length = casted_length;
1190
1191 // Use an unsigned comparison for the range check itself
1192 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1193 BoolTest::mask btest = BoolTest::lt;
1194 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1195 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1196 _gvn.set_type(rc, rc->Value(&_gvn));
1197 if (!rc_bool->is_Con()) {
1198 record_for_igvn(rc);
1199 }
1200 set_control(_gvn.transform(new IfTrueNode(rc)));
1201 {
1202 PreserveJVMState pjvms(this);
1203 set_control(_gvn.transform(new IfFalseNode(rc)));
1204 uncommon_trap(Deoptimization::Reason_range_check,
1205 Deoptimization::Action_make_not_entrant);
1206 }
1207
1208 if (stopped()) {
1209 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1210 return true;
1211 }
1212
1213 // index is now known to be >= 0 and < length, cast it
1214 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1215 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1216 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1217 result = _gvn.transform(result);
1218 set_result(result);
1219 replace_in_map(index, result);
1220 return true;
1221 }
1222
1223 //------------------------------inline_string_indexOf------------------------
1224 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1225 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1226 return false;
1227 }
1228 Node* src = argument(0);
1229 Node* tgt = argument(1);
1230
1231 // Make the merge point
1232 RegionNode* result_rgn = new RegionNode(4);
1233 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1234
1235 src = must_be_not_null(src, true);
1236 tgt = must_be_not_null(tgt, true);
1237
1238 // Get start addr and length of source string
1239 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1240 Node* src_count = load_array_length(src);
1241
1242 // Get start addr and length of substring
1243 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1244 Node* tgt_count = load_array_length(tgt);
1245
1246 Node* result = nullptr;
1247 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1248
1249 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1250 // Divide src size by 2 if String is UTF16 encoded
1251 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1252 }
1253 if (ae == StrIntrinsicNode::UU) {
1254 // Divide substring size by 2 if String is UTF16 encoded
1255 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1256 }
1257
1258 if (call_opt_stub) {
1259 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1260 StubRoutines::_string_indexof_array[ae],
1261 "stringIndexOf", TypePtr::BOTTOM, src_start,
1262 src_count, tgt_start, tgt_count);
1263 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1264 } else {
1265 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1266 result_rgn, result_phi, ae);
1267 }
1268 if (result != nullptr) {
1269 result_phi->init_req(3, result);
1270 result_rgn->init_req(3, control());
1271 }
1272 set_control(_gvn.transform(result_rgn));
1273 record_for_igvn(result_rgn);
1274 set_result(_gvn.transform(result_phi));
1275
1276 return true;
1277 }
1278
1279 //-----------------------------inline_string_indexOfI-----------------------
1280 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1281 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1282 return false;
1283 }
1284 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1285 return false;
1286 }
1287
1288 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1289 Node* src = argument(0); // byte[]
1290 Node* src_count = argument(1); // char count
1291 Node* tgt = argument(2); // byte[]
1292 Node* tgt_count = argument(3); // char count
1293 Node* from_index = argument(4); // char index
1294
1295 src = must_be_not_null(src, true);
1296 tgt = must_be_not_null(tgt, true);
1297
1298 // Multiply byte array index by 2 if String is UTF16 encoded
1299 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1300 src_count = _gvn.transform(new SubINode(src_count, from_index));
1301 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1302 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1303
1304 // Range checks
1305 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1306 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1307 if (stopped()) {
1308 return true;
1309 }
1310
1311 RegionNode* region = new RegionNode(5);
1312 Node* phi = new PhiNode(region, TypeInt::INT);
1313 Node* result = nullptr;
1314
1315 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1316
1317 if (call_opt_stub) {
1318 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1319 StubRoutines::_string_indexof_array[ae],
1320 "stringIndexOf", TypePtr::BOTTOM, src_start,
1321 src_count, tgt_start, tgt_count);
1322 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1323 } else {
1324 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1325 region, phi, ae);
1326 }
1327 if (result != nullptr) {
1328 // The result is index relative to from_index if substring was found, -1 otherwise.
1329 // Generate code which will fold into cmove.
1330 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1331 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1332
1333 Node* if_lt = generate_slow_guard(bol, nullptr);
1334 if (if_lt != nullptr) {
1335 // result == -1
1336 phi->init_req(3, result);
1337 region->init_req(3, if_lt);
1338 }
1339 if (!stopped()) {
1340 result = _gvn.transform(new AddINode(result, from_index));
1341 phi->init_req(4, result);
1342 region->init_req(4, control());
1343 }
1344 }
1345
1346 set_control(_gvn.transform(region));
1347 record_for_igvn(region);
1348 set_result(_gvn.transform(phi));
1349 clear_upper_avx();
1350
1351 return true;
1352 }
1353
1354 // Create StrIndexOfNode with fast path checks
1355 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1356 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1357 // Check for substr count > string count
1358 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1359 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1360 Node* if_gt = generate_slow_guard(bol, nullptr);
1361 if (if_gt != nullptr) {
1362 phi->init_req(1, intcon(-1));
1363 region->init_req(1, if_gt);
1364 }
1365 if (!stopped()) {
1366 // Check for substr count == 0
1367 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1368 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1369 Node* if_zero = generate_slow_guard(bol, nullptr);
1370 if (if_zero != nullptr) {
1371 phi->init_req(2, intcon(0));
1372 region->init_req(2, if_zero);
1373 }
1374 }
1375 if (!stopped()) {
1376 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1377 }
1378 return nullptr;
1379 }
1380
1381 //-----------------------------inline_string_indexOfChar-----------------------
1382 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1383 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1384 return false;
1385 }
1386 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1387 return false;
1388 }
1389 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1390 Node* src = argument(0); // byte[]
1391 Node* int_ch = argument(1);
1392 Node* from_index = argument(2);
1393 Node* max = argument(3);
1394
1395 src = must_be_not_null(src, true);
1396
1397 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1398 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1399 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1400
1401 // Range checks
1402 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1403
1404 // Check for int_ch >= 0
1405 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1406 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1407 {
1408 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1409 uncommon_trap(Deoptimization::Reason_intrinsic,
1410 Deoptimization::Action_maybe_recompile);
1411 }
1412 if (stopped()) {
1413 return true;
1414 }
1415
1416 RegionNode* region = new RegionNode(3);
1417 Node* phi = new PhiNode(region, TypeInt::INT);
1418
1419 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1420 C->set_has_split_ifs(true); // Has chance for split-if optimization
1421 _gvn.transform(result);
1422
1423 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1424 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1425
1426 Node* if_lt = generate_slow_guard(bol, nullptr);
1427 if (if_lt != nullptr) {
1428 // result == -1
1429 phi->init_req(2, result);
1430 region->init_req(2, if_lt);
1431 }
1432 if (!stopped()) {
1433 result = _gvn.transform(new AddINode(result, from_index));
1434 phi->init_req(1, result);
1435 region->init_req(1, control());
1436 }
1437 set_control(_gvn.transform(region));
1438 record_for_igvn(region);
1439 set_result(_gvn.transform(phi));
1440 clear_upper_avx();
1441
1442 return true;
1443 }
1444 //---------------------------inline_string_copy---------------------
1445 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1446 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1447 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1448 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1449 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1450 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1451 bool LibraryCallKit::inline_string_copy(bool compress) {
1452 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1453 return false;
1454 }
1455 int nargs = 5; // 2 oops, 3 ints
1456 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1457
1458 Node* src = argument(0);
1459 Node* src_offset = argument(1);
1460 Node* dst = argument(2);
1461 Node* dst_offset = argument(3);
1462 Node* length = argument(4);
1463
1464 // Check for allocation before we add nodes that would confuse
1465 // tightly_coupled_allocation()
1466 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1467
1468 // Figure out the size and type of the elements we will be copying.
1469 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1470 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1471 if (src_type == nullptr || dst_type == nullptr) {
1472 return false;
1473 }
1474 BasicType src_elem = src_type->elem()->array_element_basic_type();
1475 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1476 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1477 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1478 "Unsupported array types for inline_string_copy");
1479
1480 src = must_be_not_null(src, true);
1481 dst = must_be_not_null(dst, true);
1482
1483 // Convert char[] offsets to byte[] offsets
1484 bool convert_src = (compress && src_elem == T_BYTE);
1485 bool convert_dst = (!compress && dst_elem == T_BYTE);
1486 if (convert_src) {
1487 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1488 } else if (convert_dst) {
1489 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1490 }
1491
1492 // Range checks
1493 generate_string_range_check(src, src_offset, length, convert_src);
1494 generate_string_range_check(dst, dst_offset, length, convert_dst);
1495 if (stopped()) {
1496 return true;
1497 }
1498
1499 Node* src_start = array_element_address(src, src_offset, src_elem);
1500 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1501 // 'src_start' points to src array + scaled offset
1502 // 'dst_start' points to dst array + scaled offset
1503 Node* count = nullptr;
1504 if (compress) {
1505 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1506 } else {
1507 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1508 }
1509
1510 if (alloc != nullptr) {
1511 if (alloc->maybe_set_complete(&_gvn)) {
1512 // "You break it, you buy it."
1513 InitializeNode* init = alloc->initialization();
1514 assert(init->is_complete(), "we just did this");
1515 init->set_complete_with_arraycopy();
1516 assert(dst->is_CheckCastPP(), "sanity");
1517 assert(dst->in(0)->in(0) == init, "dest pinned");
1518 }
1519 // Do not let stores that initialize this object be reordered with
1520 // a subsequent store that would make this object accessible by
1521 // other threads.
1522 // Record what AllocateNode this StoreStore protects so that
1523 // escape analysis can go from the MemBarStoreStoreNode to the
1524 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1525 // based on the escape status of the AllocateNode.
1526 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1527 }
1528 if (compress) {
1529 set_result(_gvn.transform(count));
1530 }
1531 clear_upper_avx();
1532
1533 return true;
1534 }
1535
1536 #ifdef _LP64
1537 #define XTOP ,top() /*additional argument*/
1538 #else //_LP64
1539 #define XTOP /*no additional argument*/
1540 #endif //_LP64
1541
1542 //------------------------inline_string_toBytesU--------------------------
1543 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1544 bool LibraryCallKit::inline_string_toBytesU() {
1545 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1546 return false;
1547 }
1548 // Get the arguments.
1549 Node* value = argument(0);
1550 Node* offset = argument(1);
1551 Node* length = argument(2);
1552
1553 Node* newcopy = nullptr;
1554
1555 // Set the original stack and the reexecute bit for the interpreter to reexecute
1556 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1557 { PreserveReexecuteState preexecs(this);
1558 jvms()->set_should_reexecute(true);
1559
1560 // Check if a null path was taken unconditionally.
1561 value = null_check(value);
1562
1563 RegionNode* bailout = new RegionNode(1);
1564 record_for_igvn(bailout);
1565
1566 // Range checks
1567 generate_negative_guard(offset, bailout);
1568 generate_negative_guard(length, bailout);
1569 generate_limit_guard(offset, length, load_array_length(value), bailout);
1570 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1571 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1572
1573 if (bailout->req() > 1) {
1574 PreserveJVMState pjvms(this);
1575 set_control(_gvn.transform(bailout));
1576 uncommon_trap(Deoptimization::Reason_intrinsic,
1577 Deoptimization::Action_maybe_recompile);
1578 }
1579 if (stopped()) {
1580 return true;
1581 }
1582
1583 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1584 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1585 newcopy = new_array(klass_node, size, 0); // no arguments to push
1586 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1587 guarantee(alloc != nullptr, "created above");
1588
1589 // Calculate starting addresses.
1590 Node* src_start = array_element_address(value, offset, T_CHAR);
1591 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1592
1593 // Check if dst array address is aligned to HeapWordSize
1594 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1595 // If true, then check if src array address is aligned to HeapWordSize
1596 if (aligned) {
1597 const TypeInt* toffset = gvn().type(offset)->is_int();
1598 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1599 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1600 }
1601
1602 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1603 const char* copyfunc_name = "arraycopy";
1604 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1605 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1606 OptoRuntime::fast_arraycopy_Type(),
1607 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1608 src_start, dst_start, ConvI2X(length) XTOP);
1609 // Do not let reads from the cloned object float above the arraycopy.
1610 if (alloc->maybe_set_complete(&_gvn)) {
1611 // "You break it, you buy it."
1612 InitializeNode* init = alloc->initialization();
1613 assert(init->is_complete(), "we just did this");
1614 init->set_complete_with_arraycopy();
1615 assert(newcopy->is_CheckCastPP(), "sanity");
1616 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1617 }
1618 // Do not let stores that initialize this object be reordered with
1619 // a subsequent store that would make this object accessible by
1620 // other threads.
1621 // Record what AllocateNode this StoreStore protects so that
1622 // escape analysis can go from the MemBarStoreStoreNode to the
1623 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1624 // based on the escape status of the AllocateNode.
1625 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1626 } // original reexecute is set back here
1627
1628 C->set_has_split_ifs(true); // Has chance for split-if optimization
1629 if (!stopped()) {
1630 set_result(newcopy);
1631 }
1632 clear_upper_avx();
1633
1634 return true;
1635 }
1636
1637 //------------------------inline_string_getCharsU--------------------------
1638 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1639 bool LibraryCallKit::inline_string_getCharsU() {
1640 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1641 return false;
1642 }
1643
1644 // Get the arguments.
1645 Node* src = argument(0);
1646 Node* src_begin = argument(1);
1647 Node* src_end = argument(2); // exclusive offset (i < src_end)
1648 Node* dst = argument(3);
1649 Node* dst_begin = argument(4);
1650
1651 // Check for allocation before we add nodes that would confuse
1652 // tightly_coupled_allocation()
1653 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1654
1655 // Check if a null path was taken unconditionally.
1656 src = null_check(src);
1657 dst = null_check(dst);
1658 if (stopped()) {
1659 return true;
1660 }
1661
1662 // Get length and convert char[] offset to byte[] offset
1663 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1664 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1665
1666 // Range checks
1667 generate_string_range_check(src, src_begin, length, true);
1668 generate_string_range_check(dst, dst_begin, length, false);
1669 if (stopped()) {
1670 return true;
1671 }
1672
1673 if (!stopped()) {
1674 // Calculate starting addresses.
1675 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1676 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1677
1678 // Check if array addresses are aligned to HeapWordSize
1679 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1680 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1681 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1682 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1683
1684 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1685 const char* copyfunc_name = "arraycopy";
1686 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1687 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1688 OptoRuntime::fast_arraycopy_Type(),
1689 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1690 src_start, dst_start, ConvI2X(length) XTOP);
1691 // Do not let reads from the cloned object float above the arraycopy.
1692 if (alloc != nullptr) {
1693 if (alloc->maybe_set_complete(&_gvn)) {
1694 // "You break it, you buy it."
1695 InitializeNode* init = alloc->initialization();
1696 assert(init->is_complete(), "we just did this");
1697 init->set_complete_with_arraycopy();
1698 assert(dst->is_CheckCastPP(), "sanity");
1699 assert(dst->in(0)->in(0) == init, "dest pinned");
1700 }
1701 // Do not let stores that initialize this object be reordered with
1702 // a subsequent store that would make this object accessible by
1703 // other threads.
1704 // Record what AllocateNode this StoreStore protects so that
1705 // escape analysis can go from the MemBarStoreStoreNode to the
1706 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1707 // based on the escape status of the AllocateNode.
1708 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1709 } else {
1710 insert_mem_bar(Op_MemBarCPUOrder);
1711 }
1712 }
1713
1714 C->set_has_split_ifs(true); // Has chance for split-if optimization
1715 return true;
1716 }
1717
1718 //----------------------inline_string_char_access----------------------------
1719 // Store/Load char to/from byte[] array.
1720 // static void StringUTF16.putChar(byte[] val, int index, int c)
1721 // static char StringUTF16.getChar(byte[] val, int index)
1722 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1723 Node* value = argument(0);
1724 Node* index = argument(1);
1725 Node* ch = is_store ? argument(2) : nullptr;
1726
1727 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1728 // correctly requires matched array shapes.
1729 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1730 "sanity: byte[] and char[] bases agree");
1731 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1732 "sanity: byte[] and char[] scales agree");
1733
1734 // Bail when getChar over constants is requested: constant folding would
1735 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1736 // Java method would constant fold nicely instead.
1737 if (!is_store && value->is_Con() && index->is_Con()) {
1738 return false;
1739 }
1740
1741 // Save state and restore on bailout
1742 SavedState old_state(this);
1743
1744 value = must_be_not_null(value, true);
1745
1746 Node* adr = array_element_address(value, index, T_CHAR);
1747 if (adr->is_top()) {
1748 return false;
1749 }
1750 old_state.discard();
1751 if (is_store) {
1752 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1753 } else {
1754 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);
1755 set_result(ch);
1756 }
1757 return true;
1758 }
1759
1760
1761 //------------------------------inline_math-----------------------------------
1762 // public static double Math.abs(double)
1763 // public static double Math.sqrt(double)
1764 // public static double Math.log(double)
1765 // public static double Math.log10(double)
1766 // public static double Math.round(double)
1767 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1768 Node* arg = argument(0);
1769 Node* n = nullptr;
1770 switch (id) {
1771 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1772 case vmIntrinsics::_dsqrt:
1773 case vmIntrinsics::_dsqrt_strict:
1774 n = new SqrtDNode(C, control(), arg); break;
1775 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1776 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1777 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1778 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1779 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1780 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1781 default: fatal_unexpected_iid(id); break;
1782 }
1783 set_result(_gvn.transform(n));
1784 return true;
1785 }
1786
1787 //------------------------------inline_math-----------------------------------
1788 // public static float Math.abs(float)
1789 // public static int Math.abs(int)
1790 // public static long Math.abs(long)
1791 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1792 Node* arg = argument(0);
1793 Node* n = nullptr;
1794 switch (id) {
1795 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1796 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1797 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1798 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1799 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1800 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1801 default: fatal_unexpected_iid(id); break;
1802 }
1803 set_result(_gvn.transform(n));
1804 return true;
1805 }
1806
1807 //------------------------------runtime_math-----------------------------
1808 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1809 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1810 "must be (DD)D or (D)D type");
1811
1812 // Inputs
1813 Node* a = argument(0);
1814 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1815
1816 const TypePtr* no_memory_effects = nullptr;
1817 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1818 no_memory_effects,
1819 a, top(), b, b ? top() : nullptr);
1820 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1821 #ifdef ASSERT
1822 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1823 assert(value_top == top(), "second value must be top");
1824 #endif
1825
1826 set_result(value);
1827 return true;
1828 }
1829
1830 //------------------------------inline_math_pow-----------------------------
1831 bool LibraryCallKit::inline_math_pow() {
1832 Node* exp = argument(2);
1833 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1834 if (d != nullptr) {
1835 if (d->getd() == 2.0) {
1836 // Special case: pow(x, 2.0) => x * x
1837 Node* base = argument(0);
1838 set_result(_gvn.transform(new MulDNode(base, base)));
1839 return true;
1840 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1841 // Special case: pow(x, 0.5) => sqrt(x)
1842 Node* base = argument(0);
1843 Node* zero = _gvn.zerocon(T_DOUBLE);
1844
1845 RegionNode* region = new RegionNode(3);
1846 Node* phi = new PhiNode(region, Type::DOUBLE);
1847
1848 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1849 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1850 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1851 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1852 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1853
1854 Node* if_pow = generate_slow_guard(test, nullptr);
1855 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1856 phi->init_req(1, value_sqrt);
1857 region->init_req(1, control());
1858
1859 if (if_pow != nullptr) {
1860 set_control(if_pow);
1861 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1862 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1863 const TypePtr* no_memory_effects = nullptr;
1864 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1865 no_memory_effects, base, top(), exp, top());
1866 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1867 #ifdef ASSERT
1868 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1869 assert(value_top == top(), "second value must be top");
1870 #endif
1871 phi->init_req(2, value_pow);
1872 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1873 }
1874
1875 C->set_has_split_ifs(true); // Has chance for split-if optimization
1876 set_control(_gvn.transform(region));
1877 record_for_igvn(region);
1878 set_result(_gvn.transform(phi));
1879
1880 return true;
1881 }
1882 }
1883
1884 return StubRoutines::dpow() != nullptr ?
1885 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1886 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1887 }
1888
1889 //------------------------------inline_math_native-----------------------------
1890 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1891 switch (id) {
1892 case vmIntrinsics::_dsin:
1893 return StubRoutines::dsin() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1895 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1896 case vmIntrinsics::_dcos:
1897 return StubRoutines::dcos() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1900 case vmIntrinsics::_dtan:
1901 return StubRoutines::dtan() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1904 case vmIntrinsics::_dsinh:
1905 return StubRoutines::dsinh() != nullptr ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1907 case vmIntrinsics::_dtanh:
1908 return StubRoutines::dtanh() != nullptr ?
1909 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1910 case vmIntrinsics::_dcbrt:
1911 return StubRoutines::dcbrt() != nullptr ?
1912 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1913 case vmIntrinsics::_dexp:
1914 return StubRoutines::dexp() != nullptr ?
1915 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1916 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1917 case vmIntrinsics::_dlog:
1918 return StubRoutines::dlog() != nullptr ?
1919 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1920 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1921 case vmIntrinsics::_dlog10:
1922 return StubRoutines::dlog10() != nullptr ?
1923 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1924 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1925
1926 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1927 case vmIntrinsics::_ceil:
1928 case vmIntrinsics::_floor:
1929 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1930
1931 case vmIntrinsics::_dsqrt:
1932 case vmIntrinsics::_dsqrt_strict:
1933 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1934 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1935 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1936 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1937 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1938
1939 case vmIntrinsics::_dpow: return inline_math_pow();
1940 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1941 case vmIntrinsics::_fcopySign: return inline_math(id);
1942 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1943 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1944 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1945
1946 // These intrinsics are not yet correctly implemented
1947 case vmIntrinsics::_datan2:
1948 return false;
1949
1950 default:
1951 fatal_unexpected_iid(id);
1952 return false;
1953 }
1954 }
1955
1956 //----------------------------inline_notify-----------------------------------*
1957 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1958 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1959 address func;
1960 if (id == vmIntrinsics::_notify) {
1961 func = OptoRuntime::monitor_notify_Java();
1962 } else {
1963 func = OptoRuntime::monitor_notifyAll_Java();
1964 }
1965 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1966 make_slow_call_ex(call, env()->Throwable_klass(), false);
1967 return true;
1968 }
1969
1970
1971 //----------------------------inline_min_max-----------------------------------
1972 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1973 Node* a = nullptr;
1974 Node* b = nullptr;
1975 Node* n = nullptr;
1976 switch (id) {
1977 case vmIntrinsics::_min:
1978 case vmIntrinsics::_max:
1979 case vmIntrinsics::_minF:
1980 case vmIntrinsics::_maxF:
1981 case vmIntrinsics::_minF_strict:
1982 case vmIntrinsics::_maxF_strict:
1983 case vmIntrinsics::_min_strict:
1984 case vmIntrinsics::_max_strict:
1985 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1986 a = argument(0);
1987 b = argument(1);
1988 break;
1989 case vmIntrinsics::_minD:
1990 case vmIntrinsics::_maxD:
1991 case vmIntrinsics::_minD_strict:
1992 case vmIntrinsics::_maxD_strict:
1993 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1994 a = argument(0);
1995 b = argument(2);
1996 break;
1997 case vmIntrinsics::_minL:
1998 case vmIntrinsics::_maxL:
1999 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2000 a = argument(0);
2001 b = argument(2);
2002 break;
2003 default:
2004 fatal_unexpected_iid(id);
2005 break;
2006 }
2007
2008 switch (id) {
2009 case vmIntrinsics::_min:
2010 case vmIntrinsics::_min_strict:
2011 n = new MinINode(a, b);
2012 break;
2013 case vmIntrinsics::_max:
2014 case vmIntrinsics::_max_strict:
2015 n = new MaxINode(a, b);
2016 break;
2017 case vmIntrinsics::_minF:
2018 case vmIntrinsics::_minF_strict:
2019 n = new MinFNode(a, b);
2020 break;
2021 case vmIntrinsics::_maxF:
2022 case vmIntrinsics::_maxF_strict:
2023 n = new MaxFNode(a, b);
2024 break;
2025 case vmIntrinsics::_minD:
2026 case vmIntrinsics::_minD_strict:
2027 n = new MinDNode(a, b);
2028 break;
2029 case vmIntrinsics::_maxD:
2030 case vmIntrinsics::_maxD_strict:
2031 n = new MaxDNode(a, b);
2032 break;
2033 case vmIntrinsics::_minL:
2034 n = new MinLNode(_gvn.C, a, b);
2035 break;
2036 case vmIntrinsics::_maxL:
2037 n = new MaxLNode(_gvn.C, a, b);
2038 break;
2039 default:
2040 fatal_unexpected_iid(id);
2041 break;
2042 }
2043
2044 set_result(_gvn.transform(n));
2045 return true;
2046 }
2047
2048 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2049 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2050 env()->ArithmeticException_instance())) {
2051 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2052 // so let's bail out intrinsic rather than risking deopting again.
2053 return false;
2054 }
2055
2056 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2057 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2058 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2059 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2060
2061 {
2062 PreserveJVMState pjvms(this);
2063 PreserveReexecuteState preexecs(this);
2064 jvms()->set_should_reexecute(true);
2065
2066 set_control(slow_path);
2067 set_i_o(i_o());
2068
2069 builtin_throw(Deoptimization::Reason_intrinsic,
2070 env()->ArithmeticException_instance(),
2071 /*allow_too_many_traps*/ false);
2072 }
2073
2074 set_control(fast_path);
2075 set_result(math);
2076 return true;
2077 }
2078
2079 template <typename OverflowOp>
2080 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2081 typedef typename OverflowOp::MathOp MathOp;
2082
2083 MathOp* mathOp = new MathOp(arg1, arg2);
2084 Node* operation = _gvn.transform( mathOp );
2085 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2086 return inline_math_mathExact(operation, ofcheck);
2087 }
2088
2089 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2090 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2091 }
2092
2093 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2094 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2095 }
2096
2097 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2098 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2099 }
2100
2101 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2102 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2103 }
2104
2105 bool LibraryCallKit::inline_math_negateExactI() {
2106 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2107 }
2108
2109 bool LibraryCallKit::inline_math_negateExactL() {
2110 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2111 }
2112
2113 bool LibraryCallKit::inline_math_multiplyExactI() {
2114 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2115 }
2116
2117 bool LibraryCallKit::inline_math_multiplyExactL() {
2118 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2119 }
2120
2121 bool LibraryCallKit::inline_math_multiplyHigh() {
2122 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2123 return true;
2124 }
2125
2126 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2127 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2128 return true;
2129 }
2130
2131 inline int
2132 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2133 const TypePtr* base_type = TypePtr::NULL_PTR;
2134 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2135 if (base_type == nullptr) {
2136 // Unknown type.
2137 return Type::AnyPtr;
2138 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2139 // Since this is a null+long form, we have to switch to a rawptr.
2140 base = _gvn.transform(new CastX2PNode(offset));
2141 offset = MakeConX(0);
2142 return Type::RawPtr;
2143 } else if (base_type->base() == Type::RawPtr) {
2144 return Type::RawPtr;
2145 } else if (base_type->isa_oopptr()) {
2146 // Base is never null => always a heap address.
2147 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2148 return Type::OopPtr;
2149 }
2150 // Offset is small => always a heap address.
2151 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2152 if (offset_type != nullptr &&
2153 base_type->offset() == 0 && // (should always be?)
2154 offset_type->_lo >= 0 &&
2155 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2156 return Type::OopPtr;
2157 } else if (type == T_OBJECT) {
2158 // off heap access to an oop doesn't make any sense. Has to be on
2159 // heap.
2160 return Type::OopPtr;
2161 }
2162 // Otherwise, it might either be oop+off or null+addr.
2163 return Type::AnyPtr;
2164 } else {
2165 // No information:
2166 return Type::AnyPtr;
2167 }
2168 }
2169
2170 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2171 Node* uncasted_base = base;
2172 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2173 if (kind == Type::RawPtr) {
2174 return basic_plus_adr(top(), uncasted_base, offset);
2175 } else if (kind == Type::AnyPtr) {
2176 assert(base == uncasted_base, "unexpected base change");
2177 if (can_cast) {
2178 if (!_gvn.type(base)->speculative_maybe_null() &&
2179 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2180 // According to profiling, this access is always on
2181 // heap. Casting the base to not null and thus avoiding membars
2182 // around the access should allow better optimizations
2183 Node* null_ctl = top();
2184 base = null_check_oop(base, &null_ctl, true, true, true);
2185 assert(null_ctl->is_top(), "no null control here");
2186 return basic_plus_adr(base, offset);
2187 } else if (_gvn.type(base)->speculative_always_null() &&
2188 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2189 // According to profiling, this access is always off
2190 // heap.
2191 base = null_assert(base);
2192 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2193 offset = MakeConX(0);
2194 return basic_plus_adr(top(), raw_base, offset);
2195 }
2196 }
2197 // We don't know if it's an on heap or off heap access. Fall back
2198 // to raw memory access.
2199 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2200 return basic_plus_adr(top(), raw, offset);
2201 } else {
2202 assert(base == uncasted_base, "unexpected base change");
2203 // We know it's an on heap access so base can't be null
2204 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2205 base = must_be_not_null(base, true);
2206 }
2207 return basic_plus_adr(base, offset);
2208 }
2209 }
2210
2211 //--------------------------inline_number_methods-----------------------------
2212 // inline int Integer.numberOfLeadingZeros(int)
2213 // inline int Long.numberOfLeadingZeros(long)
2214 //
2215 // inline int Integer.numberOfTrailingZeros(int)
2216 // inline int Long.numberOfTrailingZeros(long)
2217 //
2218 // inline int Integer.bitCount(int)
2219 // inline int Long.bitCount(long)
2220 //
2221 // inline char Character.reverseBytes(char)
2222 // inline short Short.reverseBytes(short)
2223 // inline int Integer.reverseBytes(int)
2224 // inline long Long.reverseBytes(long)
2225 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2226 Node* arg = argument(0);
2227 Node* n = nullptr;
2228 switch (id) {
2229 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2230 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2231 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2232 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2233 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2234 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2235 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2236 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2237 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2238 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2239 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2240 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2241 default: fatal_unexpected_iid(id); break;
2242 }
2243 set_result(_gvn.transform(n));
2244 return true;
2245 }
2246
2247 //--------------------------inline_bitshuffle_methods-----------------------------
2248 // inline int Integer.compress(int, int)
2249 // inline int Integer.expand(int, int)
2250 // inline long Long.compress(long, long)
2251 // inline long Long.expand(long, long)
2252 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2253 Node* n = nullptr;
2254 switch (id) {
2255 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2256 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2257 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2258 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2259 default: fatal_unexpected_iid(id); break;
2260 }
2261 set_result(_gvn.transform(n));
2262 return true;
2263 }
2264
2265 //--------------------------inline_number_methods-----------------------------
2266 // inline int Integer.compareUnsigned(int, int)
2267 // inline int Long.compareUnsigned(long, long)
2268 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2269 Node* arg1 = argument(0);
2270 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2271 Node* n = nullptr;
2272 switch (id) {
2273 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2274 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2275 default: fatal_unexpected_iid(id); break;
2276 }
2277 set_result(_gvn.transform(n));
2278 return true;
2279 }
2280
2281 //--------------------------inline_unsigned_divmod_methods-----------------------------
2282 // inline int Integer.divideUnsigned(int, int)
2283 // inline int Integer.remainderUnsigned(int, int)
2284 // inline long Long.divideUnsigned(long, long)
2285 // inline long Long.remainderUnsigned(long, long)
2286 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2287 Node* n = nullptr;
2288 switch (id) {
2289 case vmIntrinsics::_divideUnsigned_i: {
2290 zero_check_int(argument(1));
2291 // Compile-time detect of null-exception
2292 if (stopped()) {
2293 return true; // keep the graph constructed so far
2294 }
2295 n = new UDivINode(control(), argument(0), argument(1));
2296 break;
2297 }
2298 case vmIntrinsics::_divideUnsigned_l: {
2299 zero_check_long(argument(2));
2300 // Compile-time detect of null-exception
2301 if (stopped()) {
2302 return true; // keep the graph constructed so far
2303 }
2304 n = new UDivLNode(control(), argument(0), argument(2));
2305 break;
2306 }
2307 case vmIntrinsics::_remainderUnsigned_i: {
2308 zero_check_int(argument(1));
2309 // Compile-time detect of null-exception
2310 if (stopped()) {
2311 return true; // keep the graph constructed so far
2312 }
2313 n = new UModINode(control(), argument(0), argument(1));
2314 break;
2315 }
2316 case vmIntrinsics::_remainderUnsigned_l: {
2317 zero_check_long(argument(2));
2318 // Compile-time detect of null-exception
2319 if (stopped()) {
2320 return true; // keep the graph constructed so far
2321 }
2322 n = new UModLNode(control(), argument(0), argument(2));
2323 break;
2324 }
2325 default: fatal_unexpected_iid(id); break;
2326 }
2327 set_result(_gvn.transform(n));
2328 return true;
2329 }
2330
2331 //----------------------------inline_unsafe_access----------------------------
2332
2333 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2334 // Attempt to infer a sharper value type from the offset and base type.
2335 ciKlass* sharpened_klass = nullptr;
2336
2337 // See if it is an instance field, with an object type.
2338 if (alias_type->field() != nullptr) {
2339 if (alias_type->field()->type()->is_klass()) {
2340 sharpened_klass = alias_type->field()->type()->as_klass();
2341 }
2342 }
2343
2344 const TypeOopPtr* result = nullptr;
2345 // See if it is a narrow oop array.
2346 if (adr_type->isa_aryptr()) {
2347 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2348 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2349 if (elem_type != nullptr && elem_type->is_loaded()) {
2350 // Sharpen the value type.
2351 result = elem_type;
2352 }
2353 }
2354 }
2355
2356 // The sharpened class might be unloaded if there is no class loader
2357 // contraint in place.
2358 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2359 // Sharpen the value type.
2360 result = TypeOopPtr::make_from_klass(sharpened_klass);
2361 }
2362 if (result != nullptr) {
2363 #ifndef PRODUCT
2364 if (C->print_intrinsics() || C->print_inlining()) {
2365 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2366 tty->print(" sharpened value: "); result->dump(); tty->cr();
2367 }
2368 #endif
2369 }
2370 return result;
2371 }
2372
2373 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2374 switch (kind) {
2375 case Relaxed:
2376 return MO_UNORDERED;
2377 case Opaque:
2378 return MO_RELAXED;
2379 case Acquire:
2380 return MO_ACQUIRE;
2381 case Release:
2382 return MO_RELEASE;
2383 case Volatile:
2384 return MO_SEQ_CST;
2385 default:
2386 ShouldNotReachHere();
2387 return 0;
2388 }
2389 }
2390
2391 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2392 _kit(kit),
2393 _sp(kit->sp()),
2394 _jvms(kit->jvms()),
2395 _map(kit->clone_map()),
2396 _discarded(false)
2397 {
2398 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2399 Node* out = kit->control()->fast_out(i);
2400 if (out->is_CFG()) {
2401 _ctrl_succ.push(out);
2402 }
2403 }
2404 }
2405
2406 LibraryCallKit::SavedState::~SavedState() {
2407 if (_discarded) {
2408 _kit->destruct_map_clone(_map);
2409 return;
2410 }
2411 _kit->jvms()->set_map(_map);
2412 _kit->jvms()->set_sp(_sp);
2413 _map->set_jvms(_kit->jvms());
2414 _kit->set_map(_map);
2415 _kit->set_sp(_sp);
2416 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2417 Node* out = _kit->control()->fast_out(i);
2418 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2419 _kit->_gvn.hash_delete(out);
2420 out->set_req(0, _kit->C->top());
2421 _kit->C->record_for_igvn(out);
2422 --i; --imax;
2423 _kit->_gvn.hash_find_insert(out);
2424 }
2425 }
2426 }
2427
2428 void LibraryCallKit::SavedState::discard() {
2429 _discarded = true;
2430 }
2431
2432 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2433 if (callee()->is_static()) return false; // caller must have the capability!
2434 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2435 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2436 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2437 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2438
2439 if (is_reference_type(type)) {
2440 decorators |= ON_UNKNOWN_OOP_REF;
2441 }
2442
2443 if (unaligned) {
2444 decorators |= C2_UNALIGNED;
2445 }
2446
2447 #ifndef PRODUCT
2448 {
2449 ResourceMark rm;
2450 // Check the signatures.
2451 ciSignature* sig = callee()->signature();
2452 #ifdef ASSERT
2453 if (!is_store) {
2454 // Object getReference(Object base, int/long offset), etc.
2455 BasicType rtype = sig->return_type()->basic_type();
2456 assert(rtype == type, "getter must return the expected value");
2457 assert(sig->count() == 2, "oop getter has 2 arguments");
2458 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2459 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2460 } else {
2461 // void putReference(Object base, int/long offset, Object x), etc.
2462 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2463 assert(sig->count() == 3, "oop putter has 3 arguments");
2464 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2465 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2466 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2467 assert(vtype == type, "putter must accept the expected value");
2468 }
2469 #endif // ASSERT
2470 }
2471 #endif //PRODUCT
2472
2473 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2474
2475 Node* receiver = argument(0); // type: oop
2476
2477 // Build address expression.
2478 Node* heap_base_oop = top();
2479
2480 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2481 Node* base = argument(1); // type: oop
2482 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2483 Node* offset = argument(2); // type: long
2484 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2485 // to be plain byte offsets, which are also the same as those accepted
2486 // by oopDesc::field_addr.
2487 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2488 "fieldOffset must be byte-scaled");
2489 // 32-bit machines ignore the high half!
2490 offset = ConvL2X(offset);
2491
2492 // Save state and restore on bailout
2493 SavedState old_state(this);
2494
2495 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2496 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2497
2498 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2499 if (type != T_OBJECT) {
2500 decorators |= IN_NATIVE; // off-heap primitive access
2501 } else {
2502 return false; // off-heap oop accesses are not supported
2503 }
2504 } else {
2505 heap_base_oop = base; // on-heap or mixed access
2506 }
2507
2508 // Can base be null? Otherwise, always on-heap access.
2509 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2510
2511 if (!can_access_non_heap) {
2512 decorators |= IN_HEAP;
2513 }
2514
2515 Node* val = is_store ? argument(4) : nullptr;
2516
2517 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2518 if (adr_type == TypePtr::NULL_PTR) {
2519 return false; // off-heap access with zero address
2520 }
2521
2522 // Try to categorize the address.
2523 Compile::AliasType* alias_type = C->alias_type(adr_type);
2524 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2525
2526 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2527 alias_type->adr_type() == TypeAryPtr::RANGE) {
2528 return false; // not supported
2529 }
2530
2531 bool mismatched = false;
2532 BasicType bt = alias_type->basic_type();
2533 if (bt != T_ILLEGAL) {
2534 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2535 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2536 // Alias type doesn't differentiate between byte[] and boolean[]).
2537 // Use address type to get the element type.
2538 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2539 }
2540 if (is_reference_type(bt, true)) {
2541 // accessing an array field with getReference is not a mismatch
2542 bt = T_OBJECT;
2543 }
2544 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2545 // Don't intrinsify mismatched object accesses
2546 return false;
2547 }
2548 mismatched = (bt != type);
2549 } else if (alias_type->adr_type()->isa_oopptr()) {
2550 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2551 }
2552
2553 old_state.discard();
2554 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2555
2556 if (mismatched) {
2557 decorators |= C2_MISMATCHED;
2558 }
2559
2560 // First guess at the value type.
2561 const Type *value_type = Type::get_const_basic_type(type);
2562
2563 // Figure out the memory ordering.
2564 decorators |= mo_decorator_for_access_kind(kind);
2565
2566 if (!is_store && type == T_OBJECT) {
2567 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2568 if (tjp != nullptr) {
2569 value_type = tjp;
2570 }
2571 }
2572
2573 receiver = null_check(receiver);
2574 if (stopped()) {
2575 return true;
2576 }
2577 // Heap pointers get a null-check from the interpreter,
2578 // as a courtesy. However, this is not guaranteed by Unsafe,
2579 // and it is not possible to fully distinguish unintended nulls
2580 // from intended ones in this API.
2581
2582 if (!is_store) {
2583 Node* p = nullptr;
2584 // Try to constant fold a load from a constant field
2585 ciField* field = alias_type->field();
2586 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2587 // final or stable field
2588 p = make_constant_from_field(field, heap_base_oop);
2589 }
2590
2591 if (p == nullptr) { // Could not constant fold the load
2592 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2593 // Normalize the value returned by getBoolean in the following cases
2594 if (type == T_BOOLEAN &&
2595 (mismatched ||
2596 heap_base_oop == top() || // - heap_base_oop is null or
2597 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2598 // and the unsafe access is made to large offset
2599 // (i.e., larger than the maximum offset necessary for any
2600 // field access)
2601 ) {
2602 IdealKit ideal = IdealKit(this);
2603 #define __ ideal.
2604 IdealVariable normalized_result(ideal);
2605 __ declarations_done();
2606 __ set(normalized_result, p);
2607 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2608 __ set(normalized_result, ideal.ConI(1));
2609 ideal.end_if();
2610 final_sync(ideal);
2611 p = __ value(normalized_result);
2612 #undef __
2613 }
2614 }
2615 if (type == T_ADDRESS) {
2616 p = gvn().transform(new CastP2XNode(nullptr, p));
2617 p = ConvX2UL(p);
2618 }
2619 // The load node has the control of the preceding MemBarCPUOrder. All
2620 // following nodes will have the control of the MemBarCPUOrder inserted at
2621 // the end of this method. So, pushing the load onto the stack at a later
2622 // point is fine.
2623 set_result(p);
2624 } else {
2625 if (bt == T_ADDRESS) {
2626 // Repackage the long as a pointer.
2627 val = ConvL2X(val);
2628 val = gvn().transform(new CastX2PNode(val));
2629 }
2630 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2631 }
2632
2633 return true;
2634 }
2635
2636 //----------------------------inline_unsafe_load_store----------------------------
2637 // This method serves a couple of different customers (depending on LoadStoreKind):
2638 //
2639 // LS_cmp_swap:
2640 //
2641 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2642 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2643 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2644 //
2645 // LS_cmp_swap_weak:
2646 //
2647 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2648 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2649 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2650 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2651 //
2652 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2653 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2654 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2655 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2656 //
2657 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2658 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2659 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2660 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2661 //
2662 // LS_cmp_exchange:
2663 //
2664 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2665 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2666 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2667 //
2668 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2669 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2670 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2671 //
2672 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2673 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2674 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2675 //
2676 // LS_get_add:
2677 //
2678 // int getAndAddInt( Object o, long offset, int delta)
2679 // long getAndAddLong(Object o, long offset, long delta)
2680 //
2681 // LS_get_set:
2682 //
2683 // int getAndSet(Object o, long offset, int newValue)
2684 // long getAndSet(Object o, long offset, long newValue)
2685 // Object getAndSet(Object o, long offset, Object newValue)
2686 //
2687 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2688 // This basic scheme here is the same as inline_unsafe_access, but
2689 // differs in enough details that combining them would make the code
2690 // overly confusing. (This is a true fact! I originally combined
2691 // them, but even I was confused by it!) As much code/comments as
2692 // possible are retained from inline_unsafe_access though to make
2693 // the correspondences clearer. - dl
2694
2695 if (callee()->is_static()) return false; // caller must have the capability!
2696
2697 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2698 decorators |= mo_decorator_for_access_kind(access_kind);
2699
2700 #ifndef PRODUCT
2701 BasicType rtype;
2702 {
2703 ResourceMark rm;
2704 // Check the signatures.
2705 ciSignature* sig = callee()->signature();
2706 rtype = sig->return_type()->basic_type();
2707 switch(kind) {
2708 case LS_get_add:
2709 case LS_get_set: {
2710 // Check the signatures.
2711 #ifdef ASSERT
2712 assert(rtype == type, "get and set must return the expected type");
2713 assert(sig->count() == 3, "get and set has 3 arguments");
2714 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2715 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2716 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2717 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2718 #endif // ASSERT
2719 break;
2720 }
2721 case LS_cmp_swap:
2722 case LS_cmp_swap_weak: {
2723 // Check the signatures.
2724 #ifdef ASSERT
2725 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2726 assert(sig->count() == 4, "CAS has 4 arguments");
2727 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2728 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2729 #endif // ASSERT
2730 break;
2731 }
2732 case LS_cmp_exchange: {
2733 // Check the signatures.
2734 #ifdef ASSERT
2735 assert(rtype == type, "CAS must return the expected type");
2736 assert(sig->count() == 4, "CAS has 4 arguments");
2737 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2738 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2739 #endif // ASSERT
2740 break;
2741 }
2742 default:
2743 ShouldNotReachHere();
2744 }
2745 }
2746 #endif //PRODUCT
2747
2748 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2749
2750 // Get arguments:
2751 Node* receiver = nullptr;
2752 Node* base = nullptr;
2753 Node* offset = nullptr;
2754 Node* oldval = nullptr;
2755 Node* newval = nullptr;
2756 switch(kind) {
2757 case LS_cmp_swap:
2758 case LS_cmp_swap_weak:
2759 case LS_cmp_exchange: {
2760 const bool two_slot_type = type2size[type] == 2;
2761 receiver = argument(0); // type: oop
2762 base = argument(1); // type: oop
2763 offset = argument(2); // type: long
2764 oldval = argument(4); // type: oop, int, or long
2765 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2766 break;
2767 }
2768 case LS_get_add:
2769 case LS_get_set: {
2770 receiver = argument(0); // type: oop
2771 base = argument(1); // type: oop
2772 offset = argument(2); // type: long
2773 oldval = nullptr;
2774 newval = argument(4); // type: oop, int, or long
2775 break;
2776 }
2777 default:
2778 ShouldNotReachHere();
2779 }
2780
2781 // Build field offset expression.
2782 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2783 // to be plain byte offsets, which are also the same as those accepted
2784 // by oopDesc::field_addr.
2785 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2786 // 32-bit machines ignore the high half of long offsets
2787 offset = ConvL2X(offset);
2788 // Save state and restore on bailout
2789 SavedState old_state(this);
2790 Node* adr = make_unsafe_address(base, offset,type, false);
2791 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2792
2793 Compile::AliasType* alias_type = C->alias_type(adr_type);
2794 BasicType bt = alias_type->basic_type();
2795 if (bt != T_ILLEGAL &&
2796 (is_reference_type(bt) != (type == T_OBJECT))) {
2797 // Don't intrinsify mismatched object accesses.
2798 return false;
2799 }
2800
2801 old_state.discard();
2802
2803 // For CAS, unlike inline_unsafe_access, there seems no point in
2804 // trying to refine types. Just use the coarse types here.
2805 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2806 const Type *value_type = Type::get_const_basic_type(type);
2807
2808 switch (kind) {
2809 case LS_get_set:
2810 case LS_cmp_exchange: {
2811 if (type == T_OBJECT) {
2812 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2813 if (tjp != nullptr) {
2814 value_type = tjp;
2815 }
2816 }
2817 break;
2818 }
2819 case LS_cmp_swap:
2820 case LS_cmp_swap_weak:
2821 case LS_get_add:
2822 break;
2823 default:
2824 ShouldNotReachHere();
2825 }
2826
2827 // Null check receiver.
2828 receiver = null_check(receiver);
2829 if (stopped()) {
2830 return true;
2831 }
2832
2833 int alias_idx = C->get_alias_index(adr_type);
2834
2835 if (is_reference_type(type)) {
2836 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2837
2838 // Transformation of a value which could be null pointer (CastPP #null)
2839 // could be delayed during Parse (for example, in adjust_map_after_if()).
2840 // Execute transformation here to avoid barrier generation in such case.
2841 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2842 newval = _gvn.makecon(TypePtr::NULL_PTR);
2843
2844 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2845 // Refine the value to a null constant, when it is known to be null
2846 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2847 }
2848 }
2849
2850 Node* result = nullptr;
2851 switch (kind) {
2852 case LS_cmp_exchange: {
2853 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2854 oldval, newval, value_type, type, decorators);
2855 break;
2856 }
2857 case LS_cmp_swap_weak:
2858 decorators |= C2_WEAK_CMPXCHG;
2859 case LS_cmp_swap: {
2860 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2861 oldval, newval, value_type, type, decorators);
2862 break;
2863 }
2864 case LS_get_set: {
2865 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2866 newval, value_type, type, decorators);
2867 break;
2868 }
2869 case LS_get_add: {
2870 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2871 newval, value_type, type, decorators);
2872 break;
2873 }
2874 default:
2875 ShouldNotReachHere();
2876 }
2877
2878 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2879 set_result(result);
2880 return true;
2881 }
2882
2883 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2884 // Regardless of form, don't allow previous ld/st to move down,
2885 // then issue acquire, release, or volatile mem_bar.
2886 insert_mem_bar(Op_MemBarCPUOrder);
2887 switch(id) {
2888 case vmIntrinsics::_loadFence:
2889 insert_mem_bar(Op_LoadFence);
2890 return true;
2891 case vmIntrinsics::_storeFence:
2892 insert_mem_bar(Op_StoreFence);
2893 return true;
2894 case vmIntrinsics::_storeStoreFence:
2895 insert_mem_bar(Op_StoreStoreFence);
2896 return true;
2897 case vmIntrinsics::_fullFence:
2898 insert_mem_bar(Op_MemBarVolatile);
2899 return true;
2900 default:
2901 fatal_unexpected_iid(id);
2902 return false;
2903 }
2904 }
2905
2906 bool LibraryCallKit::inline_onspinwait() {
2907 insert_mem_bar(Op_OnSpinWait);
2908 return true;
2909 }
2910
2911 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2912 if (!kls->is_Con()) {
2913 return true;
2914 }
2915 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2916 if (klsptr == nullptr) {
2917 return true;
2918 }
2919 ciInstanceKlass* ik = klsptr->instance_klass();
2920 // don't need a guard for a klass that is already initialized
2921 return !ik->is_initialized();
2922 }
2923
2924 //----------------------------inline_unsafe_writeback0-------------------------
2925 // public native void Unsafe.writeback0(long address)
2926 bool LibraryCallKit::inline_unsafe_writeback0() {
2927 if (!Matcher::has_match_rule(Op_CacheWB)) {
2928 return false;
2929 }
2930 #ifndef PRODUCT
2931 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2932 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2933 ciSignature* sig = callee()->signature();
2934 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2935 #endif
2936 null_check_receiver(); // null-check, then ignore
2937 Node *addr = argument(1);
2938 addr = new CastX2PNode(addr);
2939 addr = _gvn.transform(addr);
2940 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2941 flush = _gvn.transform(flush);
2942 set_memory(flush, TypeRawPtr::BOTTOM);
2943 return true;
2944 }
2945
2946 //----------------------------inline_unsafe_writeback0-------------------------
2947 // public native void Unsafe.writeback0(long address)
2948 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2949 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2950 return false;
2951 }
2952 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2953 return false;
2954 }
2955 #ifndef PRODUCT
2956 assert(Matcher::has_match_rule(Op_CacheWB),
2957 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2958 : "found match rule for CacheWBPostSync but not CacheWB"));
2959
2960 #endif
2961 null_check_receiver(); // null-check, then ignore
2962 Node *sync;
2963 if (is_pre) {
2964 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2965 } else {
2966 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2967 }
2968 sync = _gvn.transform(sync);
2969 set_memory(sync, TypeRawPtr::BOTTOM);
2970 return true;
2971 }
2972
2973 //----------------------------inline_unsafe_allocate---------------------------
2974 // public native Object Unsafe.allocateInstance(Class<?> cls);
2975 bool LibraryCallKit::inline_unsafe_allocate() {
2976
2977 #if INCLUDE_JVMTI
2978 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2979 return false;
2980 }
2981 #endif //INCLUDE_JVMTI
2982
2983 if (callee()->is_static()) return false; // caller must have the capability!
2984
2985 null_check_receiver(); // null-check, then ignore
2986 Node* cls = null_check(argument(1));
2987 if (stopped()) return true;
2988
2989 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2990 kls = null_check(kls);
2991 if (stopped()) return true; // argument was like int.class
2992
2993 #if INCLUDE_JVMTI
2994 // Don't try to access new allocated obj in the intrinsic.
2995 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2996 // Deoptimize and allocate in interpreter instead.
2997 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2998 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2999 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3000 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3001 {
3002 BuildCutout unless(this, tst, PROB_MAX);
3003 uncommon_trap(Deoptimization::Reason_intrinsic,
3004 Deoptimization::Action_make_not_entrant);
3005 }
3006 if (stopped()) {
3007 return true;
3008 }
3009 #endif //INCLUDE_JVMTI
3010
3011 Node* test = nullptr;
3012 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3013 // Note: The argument might still be an illegal value like
3014 // Serializable.class or Object[].class. The runtime will handle it.
3015 // But we must make an explicit check for initialization.
3016 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3017 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3018 // can generate code to load it as unsigned byte.
3019 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3020 Node* bits = intcon(InstanceKlass::fully_initialized);
3021 test = _gvn.transform(new SubINode(inst, bits));
3022 // The 'test' is non-zero if we need to take a slow path.
3023 }
3024
3025 Node* obj = new_instance(kls, test);
3026 set_result(obj);
3027 return true;
3028 }
3029
3030 //------------------------inline_native_time_funcs--------------
3031 // inline code for System.currentTimeMillis() and System.nanoTime()
3032 // these have the same type and signature
3033 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3034 const TypeFunc* tf = OptoRuntime::void_long_Type();
3035 const TypePtr* no_memory_effects = nullptr;
3036 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3037 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3038 #ifdef ASSERT
3039 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3040 assert(value_top == top(), "second value must be top");
3041 #endif
3042 set_result(value);
3043 return true;
3044 }
3045
3046 //--------------------inline_native_vthread_start_transition--------------------
3047 // inline void startTransition(boolean is_mount);
3048 // inline void startFinalTransition();
3049 // Pseudocode of implementation:
3050 //
3051 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3052 // carrier->set_is_in_vthread_transition(true);
3053 // OrderAccess::storeload();
3054 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3055 // + global_vthread_transition_disable_count();
3056 // if (disable_requests > 0) {
3057 // slow path: runtime call
3058 // }
3059 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3060 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3061 IdealKit ideal(this);
3062
3063 Node* thread = ideal.thread();
3064 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3065 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3066 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3067 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3068 insert_mem_bar(Op_MemBarVolatile);
3069 ideal.sync_kit(this);
3070
3071 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3072 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3073 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3074 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3075 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3076
3077 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3078 sync_kit(ideal);
3079 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3080 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3081 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3082 ideal.sync_kit(this);
3083 }
3084 ideal.end_if();
3085
3086 final_sync(ideal);
3087 return true;
3088 }
3089
3090 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3091 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3092 IdealKit ideal(this);
3093
3094 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3095 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3096
3097 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3098 sync_kit(ideal);
3099 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3100 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3101 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3102 ideal.sync_kit(this);
3103 } ideal.else_(); {
3104 Node* thread = ideal.thread();
3105 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3106 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3107
3108 sync_kit(ideal);
3109 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3110 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3111 ideal.sync_kit(this);
3112 } ideal.end_if();
3113
3114 final_sync(ideal);
3115 return true;
3116 }
3117
3118 #if INCLUDE_JVMTI
3119
3120 // Always update the is_disable_suspend bit.
3121 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3122 if (!DoJVMTIVirtualThreadTransitions) {
3123 return true;
3124 }
3125 IdealKit ideal(this);
3126
3127 {
3128 // unconditionally update the is_disable_suspend bit in current JavaThread
3129 Node* thread = ideal.thread();
3130 Node* arg = _gvn.transform(argument(0)); // argument for notification
3131 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3132 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3133
3134 sync_kit(ideal);
3135 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3136 ideal.sync_kit(this);
3137 }
3138 final_sync(ideal);
3139
3140 return true;
3141 }
3142
3143 #endif // INCLUDE_JVMTI
3144
3145 #ifdef JFR_HAVE_INTRINSICS
3146
3147 /**
3148 * if oop->klass != null
3149 * // normal class
3150 * epoch = _epoch_state ? 2 : 1
3151 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3152 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3153 * }
3154 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3155 * else
3156 * // primitive class
3157 * if oop->array_klass != null
3158 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3159 * else
3160 * id = LAST_TYPE_ID + 1 // void class path
3161 * if (!signaled)
3162 * signaled = true
3163 */
3164 bool LibraryCallKit::inline_native_classID() {
3165 Node* cls = argument(0);
3166
3167 IdealKit ideal(this);
3168 #define __ ideal.
3169 IdealVariable result(ideal); __ declarations_done();
3170 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3171 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3172 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3173
3174
3175 __ if_then(kls, BoolTest::ne, null()); {
3176 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3177 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3178
3179 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3180 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3181 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3182 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3183 mask = _gvn.transform(new OrLNode(mask, epoch));
3184 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3185
3186 float unlikely = PROB_UNLIKELY(0.999);
3187 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3188 sync_kit(ideal);
3189 make_runtime_call(RC_LEAF,
3190 OptoRuntime::class_id_load_barrier_Type(),
3191 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3192 "class id load barrier",
3193 TypePtr::BOTTOM,
3194 kls);
3195 ideal.sync_kit(this);
3196 } __ end_if();
3197
3198 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3199 } __ else_(); {
3200 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3201 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3202 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3203 __ if_then(array_kls, BoolTest::ne, null()); {
3204 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3205 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3206 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3207 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3208 } __ else_(); {
3209 // void class case
3210 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3211 } __ end_if();
3212
3213 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3214 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3215 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3216 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3217 } __ end_if();
3218 } __ end_if();
3219
3220 final_sync(ideal);
3221 set_result(ideal.value(result));
3222 #undef __
3223 return true;
3224 }
3225
3226 //------------------------inline_native_jvm_commit------------------
3227 bool LibraryCallKit::inline_native_jvm_commit() {
3228 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3229
3230 // Save input memory and i_o state.
3231 Node* input_memory_state = reset_memory();
3232 set_all_memory(input_memory_state);
3233 Node* input_io_state = i_o();
3234
3235 // TLS.
3236 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3237 // Jfr java buffer.
3238 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3239 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3240 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3241
3242 // Load the current value of the notified field in the JfrThreadLocal.
3243 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3244 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3245
3246 // Test for notification.
3247 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3248 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3249 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3250
3251 // True branch, is notified.
3252 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3253 set_control(is_notified);
3254
3255 // Reset notified state.
3256 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3257 Node* notified_reset_memory = reset_memory();
3258
3259 // 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.
3260 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3261 // Convert the machine-word to a long.
3262 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3263
3264 // False branch, not notified.
3265 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3266 set_control(not_notified);
3267 set_all_memory(input_memory_state);
3268
3269 // Arg is the next position as a long.
3270 Node* arg = argument(0);
3271 // Convert long to machine-word.
3272 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3273
3274 // Store the next_position to the underlying jfr java buffer.
3275 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3276
3277 Node* commit_memory = reset_memory();
3278 set_all_memory(commit_memory);
3279
3280 // 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.
3281 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3282 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3283 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3284
3285 // And flags with lease constant.
3286 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3287
3288 // Branch on lease to conditionalize returning the leased java buffer.
3289 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3290 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3291 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3292
3293 // False branch, not a lease.
3294 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3295
3296 // True branch, is lease.
3297 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3298 set_control(is_lease);
3299
3300 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3301 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3302 OptoRuntime::void_void_Type(),
3303 SharedRuntime::jfr_return_lease(),
3304 "return_lease", TypePtr::BOTTOM);
3305 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3306
3307 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3308 record_for_igvn(lease_compare_rgn);
3309 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3310 record_for_igvn(lease_compare_mem);
3311 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3312 record_for_igvn(lease_compare_io);
3313 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3314 record_for_igvn(lease_result_value);
3315
3316 // Update control and phi nodes.
3317 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3318 lease_compare_rgn->init_req(_false_path, not_lease);
3319
3320 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3321 lease_compare_mem->init_req(_false_path, commit_memory);
3322
3323 lease_compare_io->init_req(_true_path, i_o());
3324 lease_compare_io->init_req(_false_path, input_io_state);
3325
3326 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3327 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3328
3329 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3330 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3331 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3332 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3333
3334 // Update control and phi nodes.
3335 result_rgn->init_req(_true_path, is_notified);
3336 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3337
3338 result_mem->init_req(_true_path, notified_reset_memory);
3339 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3340
3341 result_io->init_req(_true_path, input_io_state);
3342 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3343
3344 result_value->init_req(_true_path, current_pos);
3345 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3346
3347 // Set output state.
3348 set_control(_gvn.transform(result_rgn));
3349 set_all_memory(_gvn.transform(result_mem));
3350 set_i_o(_gvn.transform(result_io));
3351 set_result(result_rgn, result_value);
3352 return true;
3353 }
3354
3355 /*
3356 * The intrinsic is a model of this pseudo-code:
3357 *
3358 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3359 * jobject h_event_writer = tl->java_event_writer();
3360 * if (h_event_writer == nullptr) {
3361 * return nullptr;
3362 * }
3363 * oop threadObj = Thread::threadObj();
3364 * oop vthread = java_lang_Thread::vthread(threadObj);
3365 * traceid tid;
3366 * bool pinVirtualThread;
3367 * bool excluded;
3368 * if (vthread != threadObj) { // i.e. current thread is virtual
3369 * tid = java_lang_Thread::tid(vthread);
3370 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3371 * pinVirtualThread = VMContinuations;
3372 * excluded = vthread_epoch_raw & excluded_mask;
3373 * if (!excluded) {
3374 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3375 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3376 * if (vthread_epoch != current_epoch) {
3377 * write_checkpoint();
3378 * }
3379 * }
3380 * } else {
3381 * tid = java_lang_Thread::tid(threadObj);
3382 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3383 * pinVirtualThread = false;
3384 * excluded = thread_epoch_raw & excluded_mask;
3385 * }
3386 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3387 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3388 * if (tid_in_event_writer != tid) {
3389 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3390 * setField(event_writer, "excluded", excluded);
3391 * setField(event_writer, "threadID", tid);
3392 * }
3393 * return event_writer
3394 */
3395 bool LibraryCallKit::inline_native_getEventWriter() {
3396 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3397
3398 // Save input memory and i_o state.
3399 Node* input_memory_state = reset_memory();
3400 set_all_memory(input_memory_state);
3401 Node* input_io_state = i_o();
3402
3403 // The most significant bit of the u2 is used to denote thread exclusion
3404 Node* excluded_shift = _gvn.intcon(15);
3405 Node* excluded_mask = _gvn.intcon(1 << 15);
3406 // The epoch generation is the range [1-32767]
3407 Node* epoch_mask = _gvn.intcon(32767);
3408
3409 // TLS
3410 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3411
3412 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3413 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3414
3415 // Load the eventwriter jobject handle.
3416 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3417
3418 // Null check the jobject handle.
3419 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3420 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3421 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3422
3423 // False path, jobj is null.
3424 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3425
3426 // True path, jobj is not null.
3427 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3428
3429 set_control(jobj_is_not_null);
3430
3431 // Load the threadObj for the CarrierThread.
3432 Node* threadObj = generate_current_thread(tls_ptr);
3433
3434 // Load the vthread.
3435 Node* vthread = generate_virtual_thread(tls_ptr);
3436
3437 // If vthread != threadObj, this is a virtual thread.
3438 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3439 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3440 IfNode* iff_vthread_not_equal_threadObj =
3441 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3442
3443 // False branch, fallback to threadObj.
3444 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3445 set_control(vthread_equal_threadObj);
3446
3447 // Load the tid field from the vthread object.
3448 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3449
3450 // Load the raw epoch value from the threadObj.
3451 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3452 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3453 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3454 TypeInt::CHAR, T_CHAR,
3455 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3456
3457 // Mask off the excluded information from the epoch.
3458 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3459
3460 // True branch, this is a virtual thread.
3461 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3462 set_control(vthread_not_equal_threadObj);
3463
3464 // Load the tid field from the vthread object.
3465 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3466
3467 // Continuation support determines if a virtual thread should be pinned.
3468 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3469 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3470
3471 // Load the raw epoch value from the vthread.
3472 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3473 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3474 TypeInt::CHAR, T_CHAR,
3475 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3476
3477 // Mask off the excluded information from the epoch.
3478 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3479
3480 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3481 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3482 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3483 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3484
3485 // False branch, vthread is excluded, no need to write epoch info.
3486 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3487
3488 // True branch, vthread is included, update epoch info.
3489 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3490 set_control(included);
3491
3492 // Get epoch value.
3493 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3494
3495 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3496 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3497 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3498
3499 // Compare the epoch in the vthread to the current epoch generation.
3500 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3501 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3502 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3503
3504 // False path, epoch is equal, checkpoint information is valid.
3505 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3506
3507 // True path, epoch is not equal, write a checkpoint for the vthread.
3508 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3509
3510 set_control(epoch_is_not_equal);
3511
3512 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3513 // The call also updates the native thread local thread id and the vthread with the current epoch.
3514 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3515 OptoRuntime::jfr_write_checkpoint_Type(),
3516 SharedRuntime::jfr_write_checkpoint(),
3517 "write_checkpoint", TypePtr::BOTTOM);
3518 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3519
3520 // vthread epoch != current epoch
3521 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3522 record_for_igvn(epoch_compare_rgn);
3523 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3524 record_for_igvn(epoch_compare_mem);
3525 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3526 record_for_igvn(epoch_compare_io);
3527
3528 // Update control and phi nodes.
3529 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3530 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3531 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3532 epoch_compare_mem->init_req(_false_path, input_memory_state);
3533 epoch_compare_io->init_req(_true_path, i_o());
3534 epoch_compare_io->init_req(_false_path, input_io_state);
3535
3536 // excluded != true
3537 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3538 record_for_igvn(exclude_compare_rgn);
3539 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3540 record_for_igvn(exclude_compare_mem);
3541 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3542 record_for_igvn(exclude_compare_io);
3543
3544 // Update control and phi nodes.
3545 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3546 exclude_compare_rgn->init_req(_false_path, excluded);
3547 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3548 exclude_compare_mem->init_req(_false_path, input_memory_state);
3549 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3550 exclude_compare_io->init_req(_false_path, input_io_state);
3551
3552 // vthread != threadObj
3553 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3554 record_for_igvn(vthread_compare_rgn);
3555 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3556 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3557 record_for_igvn(vthread_compare_io);
3558 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3559 record_for_igvn(tid);
3560 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3561 record_for_igvn(exclusion);
3562 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3563 record_for_igvn(pinVirtualThread);
3564
3565 // Update control and phi nodes.
3566 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3567 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3568 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3569 vthread_compare_mem->init_req(_false_path, input_memory_state);
3570 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3571 vthread_compare_io->init_req(_false_path, input_io_state);
3572 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3573 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3574 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3575 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3576 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3577 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3578
3579 // Update branch state.
3580 set_control(_gvn.transform(vthread_compare_rgn));
3581 set_all_memory(_gvn.transform(vthread_compare_mem));
3582 set_i_o(_gvn.transform(vthread_compare_io));
3583
3584 // Load the event writer oop by dereferencing the jobject handle.
3585 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3586 assert(klass_EventWriter->is_loaded(), "invariant");
3587 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3588 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3589 const TypeOopPtr* const xtype = aklass->as_instance_type();
3590 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3591 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3592
3593 // Load the current thread id from the event writer object.
3594 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3595 // Get the field offset to, conditionally, store an updated tid value later.
3596 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3597 // Get the field offset to, conditionally, store an updated exclusion value later.
3598 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3599 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3600 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3601
3602 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3603 record_for_igvn(event_writer_tid_compare_rgn);
3604 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3605 record_for_igvn(event_writer_tid_compare_mem);
3606 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3607 record_for_igvn(event_writer_tid_compare_io);
3608
3609 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3610 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3611 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3612 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3613
3614 // False path, tids are the same.
3615 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3616
3617 // True path, tid is not equal, need to update the tid in the event writer.
3618 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3619 record_for_igvn(tid_is_not_equal);
3620
3621 // Store the pin state to the event writer.
3622 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3623
3624 // Store the exclusion state to the event writer.
3625 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3626 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3627
3628 // Store the tid to the event writer.
3629 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3630
3631 // Update control and phi nodes.
3632 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3633 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3634 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3635 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3636 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3637 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3638
3639 // Result of top level CFG, Memory, IO and Value.
3640 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3641 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3642 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3643 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3644
3645 // Result control.
3646 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3647 result_rgn->init_req(_false_path, jobj_is_null);
3648
3649 // Result memory.
3650 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3651 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3652
3653 // Result IO.
3654 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3655 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3656
3657 // Result value.
3658 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3659 result_value->init_req(_false_path, null()); // return null
3660
3661 // Set output state.
3662 set_control(_gvn.transform(result_rgn));
3663 set_all_memory(_gvn.transform(result_mem));
3664 set_i_o(_gvn.transform(result_io));
3665 set_result(result_rgn, result_value);
3666 return true;
3667 }
3668
3669 /*
3670 * The intrinsic is a model of this pseudo-code:
3671 *
3672 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3673 * if (carrierThread != thread) { // is virtual thread
3674 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3675 * bool excluded = vthread_epoch_raw & excluded_mask;
3676 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3677 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3678 * if (!excluded) {
3679 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3680 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3681 * }
3682 * AtomicAccess::release_store(&tl->_vthread, true);
3683 * return;
3684 * }
3685 * AtomicAccess::release_store(&tl->_vthread, false);
3686 */
3687 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3688 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3689
3690 Node* input_memory_state = reset_memory();
3691 set_all_memory(input_memory_state);
3692
3693 // The most significant bit of the u2 is used to denote thread exclusion
3694 Node* excluded_mask = _gvn.intcon(1 << 15);
3695 // The epoch generation is the range [1-32767]
3696 Node* epoch_mask = _gvn.intcon(32767);
3697
3698 Node* const carrierThread = generate_current_thread(jt);
3699 // If thread != carrierThread, this is a virtual thread.
3700 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3701 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3702 IfNode* iff_thread_not_equal_carrierThread =
3703 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3704
3705 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3706
3707 // False branch, is carrierThread.
3708 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3709 // Store release
3710 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3711
3712 set_all_memory(input_memory_state);
3713
3714 // True branch, is virtual thread.
3715 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3716 set_control(thread_not_equal_carrierThread);
3717
3718 // Load the raw epoch value from the vthread.
3719 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3720 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3721 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3722
3723 // Mask off the excluded information from the epoch.
3724 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3725
3726 // Load the tid field from the thread.
3727 Node* tid = load_field_from_object(thread, "tid", "J");
3728
3729 // Store the vthread tid to the jfr thread local.
3730 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3731 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3732
3733 // Branch is_excluded to conditionalize updating the epoch .
3734 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3735 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3736 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3737
3738 // True branch, vthread is excluded, no need to write epoch info.
3739 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3740 set_control(excluded);
3741 Node* vthread_is_excluded = _gvn.intcon(1);
3742
3743 // False branch, vthread is included, update epoch info.
3744 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3745 set_control(included);
3746 Node* vthread_is_included = _gvn.intcon(0);
3747
3748 // Get epoch value.
3749 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3750
3751 // Store the vthread epoch to the jfr thread local.
3752 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3753 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3754
3755 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3756 record_for_igvn(excluded_rgn);
3757 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3758 record_for_igvn(excluded_mem);
3759 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3760 record_for_igvn(exclusion);
3761
3762 // Merge the excluded control and memory.
3763 excluded_rgn->init_req(_true_path, excluded);
3764 excluded_rgn->init_req(_false_path, included);
3765 excluded_mem->init_req(_true_path, tid_memory);
3766 excluded_mem->init_req(_false_path, included_memory);
3767 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3768 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3769
3770 // Set intermediate state.
3771 set_control(_gvn.transform(excluded_rgn));
3772 set_all_memory(excluded_mem);
3773
3774 // Store the vthread exclusion state to the jfr thread local.
3775 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3776 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3777
3778 // Store release
3779 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3780
3781 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3782 record_for_igvn(thread_compare_rgn);
3783 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3784 record_for_igvn(thread_compare_mem);
3785 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3786 record_for_igvn(vthread);
3787
3788 // Merge the thread_compare control and memory.
3789 thread_compare_rgn->init_req(_true_path, control());
3790 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3791 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3792 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3793
3794 // Set output state.
3795 set_control(_gvn.transform(thread_compare_rgn));
3796 set_all_memory(_gvn.transform(thread_compare_mem));
3797 }
3798
3799 #endif // JFR_HAVE_INTRINSICS
3800
3801 //------------------------inline_native_currentCarrierThread------------------
3802 bool LibraryCallKit::inline_native_currentCarrierThread() {
3803 Node* junk = nullptr;
3804 set_result(generate_current_thread(junk));
3805 return true;
3806 }
3807
3808 //------------------------inline_native_currentThread------------------
3809 bool LibraryCallKit::inline_native_currentThread() {
3810 Node* junk = nullptr;
3811 set_result(generate_virtual_thread(junk));
3812 return true;
3813 }
3814
3815 //------------------------inline_native_setVthread------------------
3816 bool LibraryCallKit::inline_native_setCurrentThread() {
3817 assert(C->method()->changes_current_thread(),
3818 "method changes current Thread but is not annotated ChangesCurrentThread");
3819 Node* arr = argument(1);
3820 Node* thread = _gvn.transform(new ThreadLocalNode());
3821 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3822 Node* thread_obj_handle
3823 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3824 thread_obj_handle = _gvn.transform(thread_obj_handle);
3825 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3826 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3827
3828 // Change the _monitor_owner_id of the JavaThread
3829 Node* tid = load_field_from_object(arr, "tid", "J");
3830 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3831 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3832
3833 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3834 return true;
3835 }
3836
3837 const Type* LibraryCallKit::scopedValueCache_type() {
3838 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3839 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3840 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3841
3842 // Because we create the scopedValue cache lazily we have to make the
3843 // type of the result BotPTR.
3844 bool xk = etype->klass_is_exact();
3845 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3846 return objects_type;
3847 }
3848
3849 Node* LibraryCallKit::scopedValueCache_helper() {
3850 Node* thread = _gvn.transform(new ThreadLocalNode());
3851 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3852 // We cannot use immutable_memory() because we might flip onto a
3853 // different carrier thread, at which point we'll need to use that
3854 // carrier thread's cache.
3855 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3856 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3857 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3858 }
3859
3860 //------------------------inline_native_scopedValueCache------------------
3861 bool LibraryCallKit::inline_native_scopedValueCache() {
3862 Node* cache_obj_handle = scopedValueCache_helper();
3863 const Type* objects_type = scopedValueCache_type();
3864 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3865
3866 return true;
3867 }
3868
3869 //------------------------inline_native_setScopedValueCache------------------
3870 bool LibraryCallKit::inline_native_setScopedValueCache() {
3871 Node* arr = argument(0);
3872 Node* cache_obj_handle = scopedValueCache_helper();
3873 const Type* objects_type = scopedValueCache_type();
3874
3875 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3876 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3877
3878 return true;
3879 }
3880
3881 //------------------------inline_native_Continuation_pin and unpin-----------
3882
3883 // Shared implementation routine for both pin and unpin.
3884 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3885 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3886
3887 // Save input memory.
3888 Node* input_memory_state = reset_memory();
3889 set_all_memory(input_memory_state);
3890
3891 // TLS
3892 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3893 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3894 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3895
3896 // Null check the last continuation object.
3897 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3898 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3899 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3900
3901 // False path, last continuation is null.
3902 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3903
3904 // True path, last continuation is not null.
3905 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3906
3907 set_control(continuation_is_not_null);
3908
3909 // Load the pin count from the last continuation.
3910 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3911 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3912
3913 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3914 Node* pin_count_rhs;
3915 if (unpin) {
3916 pin_count_rhs = _gvn.intcon(0);
3917 } else {
3918 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3919 }
3920 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3921 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3922 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3923
3924 // True branch, pin count over/underflow.
3925 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3926 {
3927 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3928 // which will throw IllegalStateException for pin count over/underflow.
3929 // No memory changed so far - we can use memory create by reset_memory()
3930 // at the beginning of this intrinsic. No need to call reset_memory() again.
3931 PreserveJVMState pjvms(this);
3932 set_control(pin_count_over_underflow);
3933 uncommon_trap(Deoptimization::Reason_intrinsic,
3934 Deoptimization::Action_none);
3935 assert(stopped(), "invariant");
3936 }
3937
3938 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3939 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3940 set_control(valid_pin_count);
3941
3942 Node* next_pin_count;
3943 if (unpin) {
3944 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3945 } else {
3946 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3947 }
3948
3949 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3950
3951 // Result of top level CFG and Memory.
3952 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3953 record_for_igvn(result_rgn);
3954 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3955 record_for_igvn(result_mem);
3956
3957 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3958 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3959 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3960 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3961
3962 // Set output state.
3963 set_control(_gvn.transform(result_rgn));
3964 set_all_memory(_gvn.transform(result_mem));
3965
3966 return true;
3967 }
3968
3969 //---------------------------load_mirror_from_klass----------------------------
3970 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3971 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3972 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3973 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3974 // mirror = ((OopHandle)mirror)->resolve();
3975 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3976 }
3977
3978 //-----------------------load_klass_from_mirror_common-------------------------
3979 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3980 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3981 // and branch to the given path on the region.
3982 // If never_see_null, take an uncommon trap on null, so we can optimistically
3983 // compile for the non-null case.
3984 // If the region is null, force never_see_null = true.
3985 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3986 bool never_see_null,
3987 RegionNode* region,
3988 int null_path,
3989 int offset) {
3990 if (region == nullptr) never_see_null = true;
3991 Node* p = basic_plus_adr(mirror, offset);
3992 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3993 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3994 Node* null_ctl = top();
3995 kls = null_check_oop(kls, &null_ctl, never_see_null);
3996 if (region != nullptr) {
3997 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3998 region->init_req(null_path, null_ctl);
3999 } else {
4000 assert(null_ctl == top(), "no loose ends");
4001 }
4002 return kls;
4003 }
4004
4005 //--------------------(inline_native_Class_query helpers)---------------------
4006 // Use this for JVM_ACC_INTERFACE.
4007 // Fall through if (mods & mask) == bits, take the guard otherwise.
4008 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4009 ByteSize offset, const Type* type, BasicType bt) {
4010 // Branch around if the given klass has the given modifier bit set.
4011 // Like generate_guard, adds a new path onto the region.
4012 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4013 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4014 Node* mask = intcon(modifier_mask);
4015 Node* bits = intcon(modifier_bits);
4016 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4017 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4018 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4019 return generate_fair_guard(bol, region);
4020 }
4021 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4022 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4023 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4024 }
4025
4026 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4027 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4028 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4029 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4030 }
4031
4032 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4033 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4034 }
4035
4036 //-------------------------inline_native_Class_query-------------------
4037 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4038 const Type* return_type = TypeInt::BOOL;
4039 Node* prim_return_value = top(); // what happens if it's a primitive class?
4040 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4041 bool expect_prim = false; // most of these guys expect to work on refs
4042
4043 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4044
4045 Node* mirror = argument(0);
4046 Node* obj = top();
4047
4048 switch (id) {
4049 case vmIntrinsics::_isInstance:
4050 // nothing is an instance of a primitive type
4051 prim_return_value = intcon(0);
4052 obj = argument(1);
4053 break;
4054 case vmIntrinsics::_isHidden:
4055 prim_return_value = intcon(0);
4056 break;
4057 case vmIntrinsics::_getSuperclass:
4058 prim_return_value = null();
4059 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4060 break;
4061 default:
4062 fatal_unexpected_iid(id);
4063 break;
4064 }
4065
4066 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4067 if (mirror_con == nullptr) return false; // cannot happen?
4068
4069 #ifndef PRODUCT
4070 if (C->print_intrinsics() || C->print_inlining()) {
4071 ciType* k = mirror_con->java_mirror_type();
4072 if (k) {
4073 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4074 k->print_name();
4075 tty->cr();
4076 }
4077 }
4078 #endif
4079
4080 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4081 RegionNode* region = new RegionNode(PATH_LIMIT);
4082 record_for_igvn(region);
4083 PhiNode* phi = new PhiNode(region, return_type);
4084
4085 // The mirror will never be null of Reflection.getClassAccessFlags, however
4086 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4087 // if it is. See bug 4774291.
4088
4089 // For Reflection.getClassAccessFlags(), the null check occurs in
4090 // the wrong place; see inline_unsafe_access(), above, for a similar
4091 // situation.
4092 mirror = null_check(mirror);
4093 // If mirror or obj is dead, only null-path is taken.
4094 if (stopped()) return true;
4095
4096 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4097
4098 // Now load the mirror's klass metaobject, and null-check it.
4099 // Side-effects region with the control path if the klass is null.
4100 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4101 // If kls is null, we have a primitive mirror.
4102 phi->init_req(_prim_path, prim_return_value);
4103 if (stopped()) { set_result(region, phi); return true; }
4104 bool safe_for_replace = (region->in(_prim_path) == top());
4105
4106 Node* p; // handy temp
4107 Node* null_ctl;
4108
4109 // Now that we have the non-null klass, we can perform the real query.
4110 // For constant classes, the query will constant-fold in LoadNode::Value.
4111 Node* query_value = top();
4112 switch (id) {
4113 case vmIntrinsics::_isInstance:
4114 // nothing is an instance of a primitive type
4115 query_value = gen_instanceof(obj, kls, safe_for_replace);
4116 break;
4117
4118 case vmIntrinsics::_isHidden:
4119 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4120 if (generate_hidden_class_guard(kls, region) != nullptr)
4121 // A guard was added. If the guard is taken, it was an hidden class.
4122 phi->add_req(intcon(1));
4123 // If we fall through, it's a plain class.
4124 query_value = intcon(0);
4125 break;
4126
4127
4128 case vmIntrinsics::_getSuperclass:
4129 // The rules here are somewhat unfortunate, but we can still do better
4130 // with random logic than with a JNI call.
4131 // Interfaces store null or Object as _super, but must report null.
4132 // Arrays store an intermediate super as _super, but must report Object.
4133 // Other types can report the actual _super.
4134 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4135 if (generate_array_guard(kls, region) != nullptr) {
4136 // A guard was added. If the guard is taken, it was an array.
4137 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4138 }
4139 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4140 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4141 if (generate_interface_guard(kls, region) != nullptr) {
4142 // A guard was added. If the guard is taken, it was an interface.
4143 phi->add_req(null());
4144 }
4145 // If we fall through, it's a plain class. Get its _super.
4146 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4147 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4148 null_ctl = top();
4149 kls = null_check_oop(kls, &null_ctl);
4150 if (null_ctl != top()) {
4151 // If the guard is taken, Object.superClass is null (both klass and mirror).
4152 region->add_req(null_ctl);
4153 phi ->add_req(null());
4154 }
4155 if (!stopped()) {
4156 query_value = load_mirror_from_klass(kls);
4157 }
4158 break;
4159
4160 default:
4161 fatal_unexpected_iid(id);
4162 break;
4163 }
4164
4165 // Fall-through is the normal case of a query to a real class.
4166 phi->init_req(1, query_value);
4167 region->init_req(1, control());
4168
4169 C->set_has_split_ifs(true); // Has chance for split-if optimization
4170 set_result(region, phi);
4171 return true;
4172 }
4173
4174 //-------------------------inline_Class_cast-------------------
4175 bool LibraryCallKit::inline_Class_cast() {
4176 Node* mirror = argument(0); // Class
4177 Node* obj = argument(1);
4178 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4179 if (mirror_con == nullptr) {
4180 return false; // dead path (mirror->is_top()).
4181 }
4182 if (obj == nullptr || obj->is_top()) {
4183 return false; // dead path
4184 }
4185 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4186
4187 // First, see if Class.cast() can be folded statically.
4188 // java_mirror_type() returns non-null for compile-time Class constants.
4189 ciType* tm = mirror_con->java_mirror_type();
4190 if (tm != nullptr && tm->is_klass() &&
4191 tp != nullptr) {
4192 if (!tp->is_loaded()) {
4193 // Don't use intrinsic when class is not loaded.
4194 return false;
4195 } else {
4196 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4197 if (static_res == Compile::SSC_always_true) {
4198 // isInstance() is true - fold the code.
4199 set_result(obj);
4200 return true;
4201 } else if (static_res == Compile::SSC_always_false) {
4202 // Don't use intrinsic, have to throw ClassCastException.
4203 // If the reference is null, the non-intrinsic bytecode will
4204 // be optimized appropriately.
4205 return false;
4206 }
4207 }
4208 }
4209
4210 // Bailout intrinsic and do normal inlining if exception path is frequent.
4211 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4212 return false;
4213 }
4214
4215 // Generate dynamic checks.
4216 // Class.cast() is java implementation of _checkcast bytecode.
4217 // Do checkcast (Parse::do_checkcast()) optimizations here.
4218
4219 mirror = null_check(mirror);
4220 // If mirror is dead, only null-path is taken.
4221 if (stopped()) {
4222 return true;
4223 }
4224
4225 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4226 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4227 RegionNode* region = new RegionNode(PATH_LIMIT);
4228 record_for_igvn(region);
4229
4230 // Now load the mirror's klass metaobject, and null-check it.
4231 // If kls is null, we have a primitive mirror and
4232 // nothing is an instance of a primitive type.
4233 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4234
4235 Node* res = top();
4236 if (!stopped()) {
4237 Node* bad_type_ctrl = top();
4238 // Do checkcast optimizations.
4239 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4240 region->init_req(_bad_type_path, bad_type_ctrl);
4241 }
4242 if (region->in(_prim_path) != top() ||
4243 region->in(_bad_type_path) != top()) {
4244 // Let Interpreter throw ClassCastException.
4245 PreserveJVMState pjvms(this);
4246 set_control(_gvn.transform(region));
4247 uncommon_trap(Deoptimization::Reason_intrinsic,
4248 Deoptimization::Action_maybe_recompile);
4249 }
4250 if (!stopped()) {
4251 set_result(res);
4252 }
4253 return true;
4254 }
4255
4256
4257 //--------------------------inline_native_subtype_check------------------------
4258 // This intrinsic takes the JNI calls out of the heart of
4259 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4260 bool LibraryCallKit::inline_native_subtype_check() {
4261 // Pull both arguments off the stack.
4262 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4263 args[0] = argument(0);
4264 args[1] = argument(1);
4265 Node* klasses[2]; // corresponding Klasses: superk, subk
4266 klasses[0] = klasses[1] = top();
4267
4268 enum {
4269 // A full decision tree on {superc is prim, subc is prim}:
4270 _prim_0_path = 1, // {P,N} => false
4271 // {P,P} & superc!=subc => false
4272 _prim_same_path, // {P,P} & superc==subc => true
4273 _prim_1_path, // {N,P} => false
4274 _ref_subtype_path, // {N,N} & subtype check wins => true
4275 _both_ref_path, // {N,N} & subtype check loses => false
4276 PATH_LIMIT
4277 };
4278
4279 RegionNode* region = new RegionNode(PATH_LIMIT);
4280 Node* phi = new PhiNode(region, TypeInt::BOOL);
4281 record_for_igvn(region);
4282
4283 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4284 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4285 int class_klass_offset = java_lang_Class::klass_offset();
4286
4287 // First null-check both mirrors and load each mirror's klass metaobject.
4288 int which_arg;
4289 for (which_arg = 0; which_arg <= 1; which_arg++) {
4290 Node* arg = args[which_arg];
4291 arg = null_check(arg);
4292 if (stopped()) break;
4293 args[which_arg] = arg;
4294
4295 Node* p = basic_plus_adr(arg, class_klass_offset);
4296 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4297 klasses[which_arg] = _gvn.transform(kls);
4298 }
4299
4300 // Having loaded both klasses, test each for null.
4301 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4302 for (which_arg = 0; which_arg <= 1; which_arg++) {
4303 Node* kls = klasses[which_arg];
4304 Node* null_ctl = top();
4305 kls = null_check_oop(kls, &null_ctl, never_see_null);
4306 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4307 region->init_req(prim_path, null_ctl);
4308 if (stopped()) break;
4309 klasses[which_arg] = kls;
4310 }
4311
4312 if (!stopped()) {
4313 // now we have two reference types, in klasses[0..1]
4314 Node* subk = klasses[1]; // the argument to isAssignableFrom
4315 Node* superk = klasses[0]; // the receiver
4316 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4317 // now we have a successful reference subtype check
4318 region->set_req(_ref_subtype_path, control());
4319 }
4320
4321 // If both operands are primitive (both klasses null), then
4322 // we must return true when they are identical primitives.
4323 // It is convenient to test this after the first null klass check.
4324 set_control(region->in(_prim_0_path)); // go back to first null check
4325 if (!stopped()) {
4326 // Since superc is primitive, make a guard for the superc==subc case.
4327 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4328 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4329 generate_guard(bol_eq, region, PROB_FAIR);
4330 if (region->req() == PATH_LIMIT+1) {
4331 // A guard was added. If the added guard is taken, superc==subc.
4332 region->swap_edges(PATH_LIMIT, _prim_same_path);
4333 region->del_req(PATH_LIMIT);
4334 }
4335 region->set_req(_prim_0_path, control()); // Not equal after all.
4336 }
4337
4338 // these are the only paths that produce 'true':
4339 phi->set_req(_prim_same_path, intcon(1));
4340 phi->set_req(_ref_subtype_path, intcon(1));
4341
4342 // pull together the cases:
4343 assert(region->req() == PATH_LIMIT, "sane region");
4344 for (uint i = 1; i < region->req(); i++) {
4345 Node* ctl = region->in(i);
4346 if (ctl == nullptr || ctl == top()) {
4347 region->set_req(i, top());
4348 phi ->set_req(i, top());
4349 } else if (phi->in(i) == nullptr) {
4350 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4351 }
4352 }
4353
4354 set_control(_gvn.transform(region));
4355 set_result(_gvn.transform(phi));
4356 return true;
4357 }
4358
4359 //---------------------generate_array_guard_common------------------------
4360 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4361 bool obj_array, bool not_array, Node** obj) {
4362
4363 if (stopped()) {
4364 return nullptr;
4365 }
4366
4367 // If obj_array/non_array==false/false:
4368 // Branch around if the given klass is in fact an array (either obj or prim).
4369 // If obj_array/non_array==false/true:
4370 // Branch around if the given klass is not an array klass of any kind.
4371 // If obj_array/non_array==true/true:
4372 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4373 // If obj_array/non_array==true/false:
4374 // Branch around if the kls is an oop array (Object[] or subtype)
4375 //
4376 // Like generate_guard, adds a new path onto the region.
4377 jint layout_con = 0;
4378 Node* layout_val = get_layout_helper(kls, layout_con);
4379 if (layout_val == nullptr) {
4380 bool query = (obj_array
4381 ? Klass::layout_helper_is_objArray(layout_con)
4382 : Klass::layout_helper_is_array(layout_con));
4383 if (query == not_array) {
4384 return nullptr; // never a branch
4385 } else { // always a branch
4386 Node* always_branch = control();
4387 if (region != nullptr)
4388 region->add_req(always_branch);
4389 set_control(top());
4390 return always_branch;
4391 }
4392 }
4393 // Now test the correct condition.
4394 jint nval = (obj_array
4395 ? (jint)(Klass::_lh_array_tag_type_value
4396 << Klass::_lh_array_tag_shift)
4397 : Klass::_lh_neutral_value);
4398 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4399 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4400 // invert the test if we are looking for a non-array
4401 if (not_array) btest = BoolTest(btest).negate();
4402 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4403 Node* ctrl = generate_fair_guard(bol, region);
4404 Node* is_array_ctrl = not_array ? control() : ctrl;
4405 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4406 // Keep track of the fact that 'obj' is an array to prevent
4407 // array specific accesses from floating above the guard.
4408 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4409 }
4410 return ctrl;
4411 }
4412
4413
4414 //-----------------------inline_native_newArray--------------------------
4415 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4416 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4417 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4418 Node* mirror;
4419 Node* count_val;
4420 if (uninitialized) {
4421 null_check_receiver();
4422 mirror = argument(1);
4423 count_val = argument(2);
4424 } else {
4425 mirror = argument(0);
4426 count_val = argument(1);
4427 }
4428
4429 mirror = null_check(mirror);
4430 // If mirror or obj is dead, only null-path is taken.
4431 if (stopped()) return true;
4432
4433 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4434 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4435 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4436 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4437 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4438
4439 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4440 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4441 result_reg, _slow_path);
4442 Node* normal_ctl = control();
4443 Node* no_array_ctl = result_reg->in(_slow_path);
4444
4445 // Generate code for the slow case. We make a call to newArray().
4446 set_control(no_array_ctl);
4447 if (!stopped()) {
4448 // Either the input type is void.class, or else the
4449 // array klass has not yet been cached. Either the
4450 // ensuing call will throw an exception, or else it
4451 // will cache the array klass for next time.
4452 PreserveJVMState pjvms(this);
4453 CallJavaNode* slow_call = nullptr;
4454 if (uninitialized) {
4455 // Generate optimized virtual call (holder class 'Unsafe' is final)
4456 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4457 } else {
4458 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4459 }
4460 Node* slow_result = set_results_for_java_call(slow_call);
4461 // this->control() comes from set_results_for_java_call
4462 result_reg->set_req(_slow_path, control());
4463 result_val->set_req(_slow_path, slow_result);
4464 result_io ->set_req(_slow_path, i_o());
4465 result_mem->set_req(_slow_path, reset_memory());
4466 }
4467
4468 set_control(normal_ctl);
4469 if (!stopped()) {
4470 // Normal case: The array type has been cached in the java.lang.Class.
4471 // The following call works fine even if the array type is polymorphic.
4472 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4473 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4474 result_reg->init_req(_normal_path, control());
4475 result_val->init_req(_normal_path, obj);
4476 result_io ->init_req(_normal_path, i_o());
4477 result_mem->init_req(_normal_path, reset_memory());
4478
4479 if (uninitialized) {
4480 // Mark the allocation so that zeroing is skipped
4481 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4482 alloc->maybe_set_complete(&_gvn);
4483 }
4484 }
4485
4486 // Return the combined state.
4487 set_i_o( _gvn.transform(result_io) );
4488 set_all_memory( _gvn.transform(result_mem));
4489
4490 C->set_has_split_ifs(true); // Has chance for split-if optimization
4491 set_result(result_reg, result_val);
4492 return true;
4493 }
4494
4495 //----------------------inline_native_getLength--------------------------
4496 // public static native int java.lang.reflect.Array.getLength(Object array);
4497 bool LibraryCallKit::inline_native_getLength() {
4498 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4499
4500 Node* array = null_check(argument(0));
4501 // If array is dead, only null-path is taken.
4502 if (stopped()) return true;
4503
4504 // Deoptimize if it is a non-array.
4505 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4506
4507 if (non_array != nullptr) {
4508 PreserveJVMState pjvms(this);
4509 set_control(non_array);
4510 uncommon_trap(Deoptimization::Reason_intrinsic,
4511 Deoptimization::Action_maybe_recompile);
4512 }
4513
4514 // If control is dead, only non-array-path is taken.
4515 if (stopped()) return true;
4516
4517 // The works fine even if the array type is polymorphic.
4518 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4519 Node* result = load_array_length(array);
4520
4521 C->set_has_split_ifs(true); // Has chance for split-if optimization
4522 set_result(result);
4523 return true;
4524 }
4525
4526 //------------------------inline_array_copyOf----------------------------
4527 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4528 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4529 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4530 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4531
4532 // Get the arguments.
4533 Node* original = argument(0);
4534 Node* start = is_copyOfRange? argument(1): intcon(0);
4535 Node* end = is_copyOfRange? argument(2): argument(1);
4536 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4537
4538 Node* newcopy = nullptr;
4539
4540 // Set the original stack and the reexecute bit for the interpreter to reexecute
4541 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4542 { PreserveReexecuteState preexecs(this);
4543 jvms()->set_should_reexecute(true);
4544
4545 array_type_mirror = null_check(array_type_mirror);
4546 original = null_check(original);
4547
4548 // Check if a null path was taken unconditionally.
4549 if (stopped()) return true;
4550
4551 Node* orig_length = load_array_length(original);
4552
4553 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4554 klass_node = null_check(klass_node);
4555
4556 RegionNode* bailout = new RegionNode(1);
4557 record_for_igvn(bailout);
4558
4559 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4560 // Bail out if that is so.
4561 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4562 if (not_objArray != nullptr) {
4563 // Improve the klass node's type from the new optimistic assumption:
4564 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4565 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4566 Node* cast = new CastPPNode(control(), klass_node, akls);
4567 klass_node = _gvn.transform(cast);
4568 }
4569
4570 // Bail out if either start or end is negative.
4571 generate_negative_guard(start, bailout, &start);
4572 generate_negative_guard(end, bailout, &end);
4573
4574 Node* length = end;
4575 if (_gvn.type(start) != TypeInt::ZERO) {
4576 length = _gvn.transform(new SubINode(end, start));
4577 }
4578
4579 // Bail out if length is negative (i.e., if start > end).
4580 // Without this the new_array would throw
4581 // NegativeArraySizeException but IllegalArgumentException is what
4582 // should be thrown
4583 generate_negative_guard(length, bailout, &length);
4584
4585 // Bail out if start is larger than the original length
4586 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4587 generate_negative_guard(orig_tail, bailout, &orig_tail);
4588
4589 if (bailout->req() > 1) {
4590 PreserveJVMState pjvms(this);
4591 set_control(_gvn.transform(bailout));
4592 uncommon_trap(Deoptimization::Reason_intrinsic,
4593 Deoptimization::Action_maybe_recompile);
4594 }
4595
4596 if (!stopped()) {
4597 // How many elements will we copy from the original?
4598 // The answer is MinI(orig_tail, length).
4599 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4600
4601 // Generate a direct call to the right arraycopy function(s).
4602 // We know the copy is disjoint but we might not know if the
4603 // oop stores need checking.
4604 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4605 // This will fail a store-check if x contains any non-nulls.
4606
4607 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4608 // loads/stores but it is legal only if we're sure the
4609 // Arrays.copyOf would succeed. So we need all input arguments
4610 // to the copyOf to be validated, including that the copy to the
4611 // new array won't trigger an ArrayStoreException. That subtype
4612 // check can be optimized if we know something on the type of
4613 // the input array from type speculation.
4614 if (_gvn.type(klass_node)->singleton()) {
4615 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4616 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4617
4618 int test = C->static_subtype_check(superk, subk);
4619 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4620 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4621 if (t_original->speculative_type() != nullptr) {
4622 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4623 }
4624 }
4625 }
4626
4627 bool validated = false;
4628 // Reason_class_check rather than Reason_intrinsic because we
4629 // want to intrinsify even if this traps.
4630 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4631 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4632
4633 if (not_subtype_ctrl != top()) {
4634 PreserveJVMState pjvms(this);
4635 set_control(not_subtype_ctrl);
4636 uncommon_trap(Deoptimization::Reason_class_check,
4637 Deoptimization::Action_make_not_entrant);
4638 assert(stopped(), "Should be stopped");
4639 }
4640 validated = true;
4641 }
4642
4643 if (!stopped()) {
4644 newcopy = new_array(klass_node, length, 0); // no arguments to push
4645
4646 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4647 load_object_klass(original), klass_node);
4648 if (!is_copyOfRange) {
4649 ac->set_copyof(validated);
4650 } else {
4651 ac->set_copyofrange(validated);
4652 }
4653 Node* n = _gvn.transform(ac);
4654 if (n == ac) {
4655 ac->connect_outputs(this);
4656 } else {
4657 assert(validated, "shouldn't transform if all arguments not validated");
4658 set_all_memory(n);
4659 }
4660 }
4661 }
4662 } // original reexecute is set back here
4663
4664 C->set_has_split_ifs(true); // Has chance for split-if optimization
4665 if (!stopped()) {
4666 set_result(newcopy);
4667 }
4668 return true;
4669 }
4670
4671
4672 //----------------------generate_virtual_guard---------------------------
4673 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4674 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4675 RegionNode* slow_region) {
4676 ciMethod* method = callee();
4677 int vtable_index = method->vtable_index();
4678 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4679 "bad index %d", vtable_index);
4680 // Get the Method* out of the appropriate vtable entry.
4681 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4682 vtable_index*vtableEntry::size_in_bytes() +
4683 in_bytes(vtableEntry::method_offset());
4684 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4685 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4686
4687 // Compare the target method with the expected method (e.g., Object.hashCode).
4688 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4689
4690 Node* native_call = makecon(native_call_addr);
4691 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4692 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4693
4694 return generate_slow_guard(test_native, slow_region);
4695 }
4696
4697 //-----------------------generate_method_call----------------------------
4698 // Use generate_method_call to make a slow-call to the real
4699 // method if the fast path fails. An alternative would be to
4700 // use a stub like OptoRuntime::slow_arraycopy_Java.
4701 // This only works for expanding the current library call,
4702 // not another intrinsic. (E.g., don't use this for making an
4703 // arraycopy call inside of the copyOf intrinsic.)
4704 CallJavaNode*
4705 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4706 // When compiling the intrinsic method itself, do not use this technique.
4707 guarantee(callee() != C->method(), "cannot make slow-call to self");
4708
4709 ciMethod* method = callee();
4710 // ensure the JVMS we have will be correct for this call
4711 guarantee(method_id == method->intrinsic_id(), "must match");
4712
4713 const TypeFunc* tf = TypeFunc::make(method);
4714 if (res_not_null) {
4715 assert(tf->return_type() == T_OBJECT, "");
4716 const TypeTuple* range = tf->range();
4717 const Type** fields = TypeTuple::fields(range->cnt());
4718 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4719 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4720 tf = TypeFunc::make(tf->domain(), new_range);
4721 }
4722 CallJavaNode* slow_call;
4723 if (is_static) {
4724 assert(!is_virtual, "");
4725 slow_call = new CallStaticJavaNode(C, tf,
4726 SharedRuntime::get_resolve_static_call_stub(), method);
4727 } else if (is_virtual) {
4728 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4729 int vtable_index = Method::invalid_vtable_index;
4730 if (UseInlineCaches) {
4731 // Suppress the vtable call
4732 } else {
4733 // hashCode and clone are not a miranda methods,
4734 // so the vtable index is fixed.
4735 // No need to use the linkResolver to get it.
4736 vtable_index = method->vtable_index();
4737 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4738 "bad index %d", vtable_index);
4739 }
4740 slow_call = new CallDynamicJavaNode(tf,
4741 SharedRuntime::get_resolve_virtual_call_stub(),
4742 method, vtable_index);
4743 } else { // neither virtual nor static: opt_virtual
4744 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4745 slow_call = new CallStaticJavaNode(C, tf,
4746 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4747 slow_call->set_optimized_virtual(true);
4748 }
4749 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4750 // To be able to issue a direct call (optimized virtual or virtual)
4751 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4752 // about the method being invoked should be attached to the call site to
4753 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4754 slow_call->set_override_symbolic_info(true);
4755 }
4756 set_arguments_for_java_call(slow_call);
4757 set_edges_for_java_call(slow_call);
4758 return slow_call;
4759 }
4760
4761
4762 /**
4763 * Build special case code for calls to hashCode on an object. This call may
4764 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4765 * slightly different code.
4766 */
4767 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4768 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4769 assert(!(is_virtual && is_static), "either virtual, special, or static");
4770
4771 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4772
4773 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4774 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4775 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4776 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4777 Node* obj = nullptr;
4778 if (!is_static) {
4779 // Check for hashing null object
4780 obj = null_check_receiver();
4781 if (stopped()) return true; // unconditionally null
4782 result_reg->init_req(_null_path, top());
4783 result_val->init_req(_null_path, top());
4784 } else {
4785 // Do a null check, and return zero if null.
4786 // System.identityHashCode(null) == 0
4787 obj = argument(0);
4788 Node* null_ctl = top();
4789 obj = null_check_oop(obj, &null_ctl);
4790 result_reg->init_req(_null_path, null_ctl);
4791 result_val->init_req(_null_path, _gvn.intcon(0));
4792 }
4793
4794 // Unconditionally null? Then return right away.
4795 if (stopped()) {
4796 set_control( result_reg->in(_null_path));
4797 if (!stopped())
4798 set_result(result_val->in(_null_path));
4799 return true;
4800 }
4801
4802 // We only go to the fast case code if we pass a number of guards. The
4803 // paths which do not pass are accumulated in the slow_region.
4804 RegionNode* slow_region = new RegionNode(1);
4805 record_for_igvn(slow_region);
4806
4807 // If this is a virtual call, we generate a funny guard. We pull out
4808 // the vtable entry corresponding to hashCode() from the target object.
4809 // If the target method which we are calling happens to be the native
4810 // Object hashCode() method, we pass the guard. We do not need this
4811 // guard for non-virtual calls -- the caller is known to be the native
4812 // Object hashCode().
4813 if (is_virtual) {
4814 // After null check, get the object's klass.
4815 Node* obj_klass = load_object_klass(obj);
4816 generate_virtual_guard(obj_klass, slow_region);
4817 }
4818
4819 // Get the header out of the object, use LoadMarkNode when available
4820 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4821 // The control of the load must be null. Otherwise, the load can move before
4822 // the null check after castPP removal.
4823 Node* no_ctrl = nullptr;
4824 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4825
4826 if (!UseObjectMonitorTable) {
4827 // Test the header to see if it is safe to read w.r.t. locking.
4828 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4829 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4830 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4831 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4832 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4833
4834 generate_slow_guard(test_monitor, slow_region);
4835 }
4836
4837 // Get the hash value and check to see that it has been properly assigned.
4838 // We depend on hash_mask being at most 32 bits and avoid the use of
4839 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4840 // vm: see markWord.hpp.
4841 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4842 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4843 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4844 // This hack lets the hash bits live anywhere in the mark object now, as long
4845 // as the shift drops the relevant bits into the low 32 bits. Note that
4846 // Java spec says that HashCode is an int so there's no point in capturing
4847 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4848 hshifted_header = ConvX2I(hshifted_header);
4849 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4850
4851 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4852 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4853 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4854
4855 generate_slow_guard(test_assigned, slow_region);
4856
4857 Node* init_mem = reset_memory();
4858 // fill in the rest of the null path:
4859 result_io ->init_req(_null_path, i_o());
4860 result_mem->init_req(_null_path, init_mem);
4861
4862 result_val->init_req(_fast_path, hash_val);
4863 result_reg->init_req(_fast_path, control());
4864 result_io ->init_req(_fast_path, i_o());
4865 result_mem->init_req(_fast_path, init_mem);
4866
4867 // Generate code for the slow case. We make a call to hashCode().
4868 set_control(_gvn.transform(slow_region));
4869 if (!stopped()) {
4870 // No need for PreserveJVMState, because we're using up the present state.
4871 set_all_memory(init_mem);
4872 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4873 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4874 Node* slow_result = set_results_for_java_call(slow_call);
4875 // this->control() comes from set_results_for_java_call
4876 result_reg->init_req(_slow_path, control());
4877 result_val->init_req(_slow_path, slow_result);
4878 result_io ->set_req(_slow_path, i_o());
4879 result_mem ->set_req(_slow_path, reset_memory());
4880 }
4881
4882 // Return the combined state.
4883 set_i_o( _gvn.transform(result_io) );
4884 set_all_memory( _gvn.transform(result_mem));
4885
4886 set_result(result_reg, result_val);
4887 return true;
4888 }
4889
4890 //---------------------------inline_native_getClass----------------------------
4891 // public final native Class<?> java.lang.Object.getClass();
4892 //
4893 // Build special case code for calls to getClass on an object.
4894 bool LibraryCallKit::inline_native_getClass() {
4895 Node* obj = null_check_receiver();
4896 if (stopped()) return true;
4897 set_result(load_mirror_from_klass(load_object_klass(obj)));
4898 return true;
4899 }
4900
4901 //-----------------inline_native_Reflection_getCallerClass---------------------
4902 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4903 //
4904 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4905 //
4906 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4907 // in that it must skip particular security frames and checks for
4908 // caller sensitive methods.
4909 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4910 #ifndef PRODUCT
4911 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4912 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4913 }
4914 #endif
4915
4916 if (!jvms()->has_method()) {
4917 #ifndef PRODUCT
4918 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4919 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4920 }
4921 #endif
4922 return false;
4923 }
4924
4925 // Walk back up the JVM state to find the caller at the required
4926 // depth.
4927 JVMState* caller_jvms = jvms();
4928
4929 // Cf. JVM_GetCallerClass
4930 // NOTE: Start the loop at depth 1 because the current JVM state does
4931 // not include the Reflection.getCallerClass() frame.
4932 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4933 ciMethod* m = caller_jvms->method();
4934 switch (n) {
4935 case 0:
4936 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4937 break;
4938 case 1:
4939 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4940 if (!m->caller_sensitive()) {
4941 #ifndef PRODUCT
4942 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4943 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4944 }
4945 #endif
4946 return false; // bail-out; let JVM_GetCallerClass do the work
4947 }
4948 break;
4949 default:
4950 if (!m->is_ignored_by_security_stack_walk()) {
4951 // We have reached the desired frame; return the holder class.
4952 // Acquire method holder as java.lang.Class and push as constant.
4953 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4954 ciInstance* caller_mirror = caller_klass->java_mirror();
4955 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4956
4957 #ifndef PRODUCT
4958 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4959 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());
4960 tty->print_cr(" JVM state at this point:");
4961 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4962 ciMethod* m = jvms()->of_depth(i)->method();
4963 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4964 }
4965 }
4966 #endif
4967 return true;
4968 }
4969 break;
4970 }
4971 }
4972
4973 #ifndef PRODUCT
4974 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4975 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4976 tty->print_cr(" JVM state at this point:");
4977 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4978 ciMethod* m = jvms()->of_depth(i)->method();
4979 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4980 }
4981 }
4982 #endif
4983
4984 return false; // bail-out; let JVM_GetCallerClass do the work
4985 }
4986
4987 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4988 Node* arg = argument(0);
4989 Node* result = nullptr;
4990
4991 switch (id) {
4992 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4993 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4994 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4995 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4996 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4997 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4998
4999 case vmIntrinsics::_doubleToLongBits: {
5000 // two paths (plus control) merge in a wood
5001 RegionNode *r = new RegionNode(3);
5002 Node *phi = new PhiNode(r, TypeLong::LONG);
5003
5004 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5005 // Build the boolean node
5006 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5007
5008 // Branch either way.
5009 // NaN case is less traveled, which makes all the difference.
5010 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5011 Node *opt_isnan = _gvn.transform(ifisnan);
5012 assert( opt_isnan->is_If(), "Expect an IfNode");
5013 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5014 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5015
5016 set_control(iftrue);
5017
5018 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5019 Node *slow_result = longcon(nan_bits); // return NaN
5020 phi->init_req(1, _gvn.transform( slow_result ));
5021 r->init_req(1, iftrue);
5022
5023 // Else fall through
5024 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5025 set_control(iffalse);
5026
5027 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5028 r->init_req(2, iffalse);
5029
5030 // Post merge
5031 set_control(_gvn.transform(r));
5032 record_for_igvn(r);
5033
5034 C->set_has_split_ifs(true); // Has chance for split-if optimization
5035 result = phi;
5036 assert(result->bottom_type()->isa_long(), "must be");
5037 break;
5038 }
5039
5040 case vmIntrinsics::_floatToIntBits: {
5041 // two paths (plus control) merge in a wood
5042 RegionNode *r = new RegionNode(3);
5043 Node *phi = new PhiNode(r, TypeInt::INT);
5044
5045 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5046 // Build the boolean node
5047 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5048
5049 // Branch either way.
5050 // NaN case is less traveled, which makes all the difference.
5051 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5052 Node *opt_isnan = _gvn.transform(ifisnan);
5053 assert( opt_isnan->is_If(), "Expect an IfNode");
5054 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5055 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5056
5057 set_control(iftrue);
5058
5059 static const jint nan_bits = 0x7fc00000;
5060 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5061 phi->init_req(1, _gvn.transform( slow_result ));
5062 r->init_req(1, iftrue);
5063
5064 // Else fall through
5065 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5066 set_control(iffalse);
5067
5068 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5069 r->init_req(2, iffalse);
5070
5071 // Post merge
5072 set_control(_gvn.transform(r));
5073 record_for_igvn(r);
5074
5075 C->set_has_split_ifs(true); // Has chance for split-if optimization
5076 result = phi;
5077 assert(result->bottom_type()->isa_int(), "must be");
5078 break;
5079 }
5080
5081 default:
5082 fatal_unexpected_iid(id);
5083 break;
5084 }
5085 set_result(_gvn.transform(result));
5086 return true;
5087 }
5088
5089 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5090 Node* arg = argument(0);
5091 Node* result = nullptr;
5092
5093 switch (id) {
5094 case vmIntrinsics::_floatIsInfinite:
5095 result = new IsInfiniteFNode(arg);
5096 break;
5097 case vmIntrinsics::_floatIsFinite:
5098 result = new IsFiniteFNode(arg);
5099 break;
5100 case vmIntrinsics::_doubleIsInfinite:
5101 result = new IsInfiniteDNode(arg);
5102 break;
5103 case vmIntrinsics::_doubleIsFinite:
5104 result = new IsFiniteDNode(arg);
5105 break;
5106 default:
5107 fatal_unexpected_iid(id);
5108 break;
5109 }
5110 set_result(_gvn.transform(result));
5111 return true;
5112 }
5113
5114 //----------------------inline_unsafe_copyMemory-------------------------
5115 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5116
5117 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5118 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5119 const Type* base_t = gvn.type(base);
5120
5121 bool in_native = (base_t == TypePtr::NULL_PTR);
5122 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5123 bool is_mixed = !in_heap && !in_native;
5124
5125 if (is_mixed) {
5126 return true; // mixed accesses can touch both on-heap and off-heap memory
5127 }
5128 if (in_heap) {
5129 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5130 if (!is_prim_array) {
5131 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5132 // there's not enough type information available to determine proper memory slice for it.
5133 return true;
5134 }
5135 }
5136 return false;
5137 }
5138
5139 bool LibraryCallKit::inline_unsafe_copyMemory() {
5140 if (callee()->is_static()) return false; // caller must have the capability!
5141 null_check_receiver(); // null-check receiver
5142 if (stopped()) return true;
5143
5144 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5145
5146 Node* src_base = argument(1); // type: oop
5147 Node* src_off = ConvL2X(argument(2)); // type: long
5148 Node* dst_base = argument(4); // type: oop
5149 Node* dst_off = ConvL2X(argument(5)); // type: long
5150 Node* size = ConvL2X(argument(7)); // type: long
5151
5152 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5153 "fieldOffset must be byte-scaled");
5154
5155 Node* src_addr = make_unsafe_address(src_base, src_off);
5156 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5157
5158 Node* thread = _gvn.transform(new ThreadLocalNode());
5159 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5160 BasicType doing_unsafe_access_bt = T_BYTE;
5161 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5162
5163 // update volatile field
5164 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5165
5166 int flags = RC_LEAF | RC_NO_FP;
5167
5168 const TypePtr* dst_type = TypePtr::BOTTOM;
5169
5170 // Adjust memory effects of the runtime call based on input values.
5171 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5172 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5173 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5174
5175 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5176 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5177 flags |= RC_NARROW_MEM; // narrow in memory
5178 }
5179 }
5180
5181 // Call it. Note that the length argument is not scaled.
5182 make_runtime_call(flags,
5183 OptoRuntime::fast_arraycopy_Type(),
5184 StubRoutines::unsafe_arraycopy(),
5185 "unsafe_arraycopy",
5186 dst_type,
5187 src_addr, dst_addr, size XTOP);
5188
5189 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5190
5191 return true;
5192 }
5193
5194 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5195 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5196 bool LibraryCallKit::inline_unsafe_setMemory() {
5197 if (callee()->is_static()) return false; // caller must have the capability!
5198 null_check_receiver(); // null-check receiver
5199 if (stopped()) return true;
5200
5201 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5202
5203 Node* dst_base = argument(1); // type: oop
5204 Node* dst_off = ConvL2X(argument(2)); // type: long
5205 Node* size = ConvL2X(argument(4)); // type: long
5206 Node* byte = argument(6); // type: byte
5207
5208 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5209 "fieldOffset must be byte-scaled");
5210
5211 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5212
5213 Node* thread = _gvn.transform(new ThreadLocalNode());
5214 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5215 BasicType doing_unsafe_access_bt = T_BYTE;
5216 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5217
5218 // update volatile field
5219 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5220
5221 int flags = RC_LEAF | RC_NO_FP;
5222
5223 const TypePtr* dst_type = TypePtr::BOTTOM;
5224
5225 // Adjust memory effects of the runtime call based on input values.
5226 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5227 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5228
5229 flags |= RC_NARROW_MEM; // narrow in memory
5230 }
5231
5232 // Call it. Note that the length argument is not scaled.
5233 make_runtime_call(flags,
5234 OptoRuntime::unsafe_setmemory_Type(),
5235 StubRoutines::unsafe_setmemory(),
5236 "unsafe_setmemory",
5237 dst_type,
5238 dst_addr, size XTOP, byte);
5239
5240 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5241
5242 return true;
5243 }
5244
5245 #undef XTOP
5246
5247 //------------------------clone_coping-----------------------------------
5248 // Helper function for inline_native_clone.
5249 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5250 assert(obj_size != nullptr, "");
5251 Node* raw_obj = alloc_obj->in(1);
5252 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5253
5254 AllocateNode* alloc = nullptr;
5255 if (ReduceBulkZeroing &&
5256 // If we are implementing an array clone without knowing its source type
5257 // (can happen when compiling the array-guarded branch of a reflective
5258 // Object.clone() invocation), initialize the array within the allocation.
5259 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5260 // to a runtime clone call that assumes fully initialized source arrays.
5261 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5262 // We will be completely responsible for initializing this object -
5263 // mark Initialize node as complete.
5264 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5265 // The object was just allocated - there should be no any stores!
5266 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5267 // Mark as complete_with_arraycopy so that on AllocateNode
5268 // expansion, we know this AllocateNode is initialized by an array
5269 // copy and a StoreStore barrier exists after the array copy.
5270 alloc->initialization()->set_complete_with_arraycopy();
5271 }
5272
5273 Node* size = _gvn.transform(obj_size);
5274 access_clone(obj, alloc_obj, size, is_array);
5275
5276 // Do not let reads from the cloned object float above the arraycopy.
5277 if (alloc != nullptr) {
5278 // Do not let stores that initialize this object be reordered with
5279 // a subsequent store that would make this object accessible by
5280 // other threads.
5281 // Record what AllocateNode this StoreStore protects so that
5282 // escape analysis can go from the MemBarStoreStoreNode to the
5283 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5284 // based on the escape status of the AllocateNode.
5285 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5286 } else {
5287 insert_mem_bar(Op_MemBarCPUOrder);
5288 }
5289 }
5290
5291 //------------------------inline_native_clone----------------------------
5292 // protected native Object java.lang.Object.clone();
5293 //
5294 // Here are the simple edge cases:
5295 // null receiver => normal trap
5296 // virtual and clone was overridden => slow path to out-of-line clone
5297 // not cloneable or finalizer => slow path to out-of-line Object.clone
5298 //
5299 // The general case has two steps, allocation and copying.
5300 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5301 //
5302 // Copying also has two cases, oop arrays and everything else.
5303 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5304 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5305 //
5306 // These steps fold up nicely if and when the cloned object's klass
5307 // can be sharply typed as an object array, a type array, or an instance.
5308 //
5309 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5310 PhiNode* result_val;
5311
5312 // Set the reexecute bit for the interpreter to reexecute
5313 // the bytecode that invokes Object.clone if deoptimization happens.
5314 { PreserveReexecuteState preexecs(this);
5315 jvms()->set_should_reexecute(true);
5316
5317 Node* obj = null_check_receiver();
5318 if (stopped()) return true;
5319
5320 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5321
5322 // If we are going to clone an instance, we need its exact type to
5323 // know the number and types of fields to convert the clone to
5324 // loads/stores. Maybe a speculative type can help us.
5325 if (!obj_type->klass_is_exact() &&
5326 obj_type->speculative_type() != nullptr &&
5327 obj_type->speculative_type()->is_instance_klass()) {
5328 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5329 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5330 !spec_ik->has_injected_fields()) {
5331 if (!obj_type->isa_instptr() ||
5332 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5333 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5334 }
5335 }
5336 }
5337
5338 // Conservatively insert a memory barrier on all memory slices.
5339 // Do not let writes into the original float below the clone.
5340 insert_mem_bar(Op_MemBarCPUOrder);
5341
5342 // paths into result_reg:
5343 enum {
5344 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5345 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5346 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5347 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5348 PATH_LIMIT
5349 };
5350 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5351 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5352 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5353 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5354 record_for_igvn(result_reg);
5355
5356 Node* obj_klass = load_object_klass(obj);
5357 Node* array_obj = obj;
5358 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5359 if (array_ctl != nullptr) {
5360 // It's an array.
5361 PreserveJVMState pjvms(this);
5362 set_control(array_ctl);
5363 Node* obj_length = load_array_length(array_obj);
5364 Node* array_size = nullptr; // Size of the array without object alignment padding.
5365 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5366
5367 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5368 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5369 // If it is an oop array, it requires very special treatment,
5370 // because gc barriers are required when accessing the array.
5371 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5372 if (is_obja != nullptr) {
5373 PreserveJVMState pjvms2(this);
5374 set_control(is_obja);
5375 // Generate a direct call to the right arraycopy function(s).
5376 // Clones are always tightly coupled.
5377 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5378 ac->set_clone_oop_array();
5379 Node* n = _gvn.transform(ac);
5380 assert(n == ac, "cannot disappear");
5381 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5382
5383 result_reg->init_req(_objArray_path, control());
5384 result_val->init_req(_objArray_path, alloc_obj);
5385 result_i_o ->set_req(_objArray_path, i_o());
5386 result_mem ->set_req(_objArray_path, reset_memory());
5387 }
5388 }
5389 // Otherwise, there are no barriers to worry about.
5390 // (We can dispense with card marks if we know the allocation
5391 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5392 // causes the non-eden paths to take compensating steps to
5393 // simulate a fresh allocation, so that no further
5394 // card marks are required in compiled code to initialize
5395 // the object.)
5396
5397 if (!stopped()) {
5398 copy_to_clone(array_obj, alloc_obj, array_size, true);
5399
5400 // Present the results of the copy.
5401 result_reg->init_req(_array_path, control());
5402 result_val->init_req(_array_path, alloc_obj);
5403 result_i_o ->set_req(_array_path, i_o());
5404 result_mem ->set_req(_array_path, reset_memory());
5405 }
5406 }
5407
5408 // We only go to the instance fast case code if we pass a number of guards.
5409 // The paths which do not pass are accumulated in the slow_region.
5410 RegionNode* slow_region = new RegionNode(1);
5411 record_for_igvn(slow_region);
5412 if (!stopped()) {
5413 // It's an instance (we did array above). Make the slow-path tests.
5414 // If this is a virtual call, we generate a funny guard. We grab
5415 // the vtable entry corresponding to clone() from the target object.
5416 // If the target method which we are calling happens to be the
5417 // Object clone() method, we pass the guard. We do not need this
5418 // guard for non-virtual calls; the caller is known to be the native
5419 // Object clone().
5420 if (is_virtual) {
5421 generate_virtual_guard(obj_klass, slow_region);
5422 }
5423
5424 // The object must be easily cloneable and must not have a finalizer.
5425 // Both of these conditions may be checked in a single test.
5426 // We could optimize the test further, but we don't care.
5427 generate_misc_flags_guard(obj_klass,
5428 // Test both conditions:
5429 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5430 // Must be cloneable but not finalizer:
5431 KlassFlags::_misc_is_cloneable_fast,
5432 slow_region);
5433 }
5434
5435 if (!stopped()) {
5436 // It's an instance, and it passed the slow-path tests.
5437 PreserveJVMState pjvms(this);
5438 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5439 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5440 // is reexecuted if deoptimization occurs and there could be problems when merging
5441 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5442 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5443
5444 copy_to_clone(obj, alloc_obj, obj_size, false);
5445
5446 // Present the results of the slow call.
5447 result_reg->init_req(_instance_path, control());
5448 result_val->init_req(_instance_path, alloc_obj);
5449 result_i_o ->set_req(_instance_path, i_o());
5450 result_mem ->set_req(_instance_path, reset_memory());
5451 }
5452
5453 // Generate code for the slow case. We make a call to clone().
5454 set_control(_gvn.transform(slow_region));
5455 if (!stopped()) {
5456 PreserveJVMState pjvms(this);
5457 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5458 // We need to deoptimize on exception (see comment above)
5459 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5460 // this->control() comes from set_results_for_java_call
5461 result_reg->init_req(_slow_path, control());
5462 result_val->init_req(_slow_path, slow_result);
5463 result_i_o ->set_req(_slow_path, i_o());
5464 result_mem ->set_req(_slow_path, reset_memory());
5465 }
5466
5467 // Return the combined state.
5468 set_control( _gvn.transform(result_reg));
5469 set_i_o( _gvn.transform(result_i_o));
5470 set_all_memory( _gvn.transform(result_mem));
5471 } // original reexecute is set back here
5472
5473 set_result(_gvn.transform(result_val));
5474 return true;
5475 }
5476
5477 // If we have a tightly coupled allocation, the arraycopy may take care
5478 // of the array initialization. If one of the guards we insert between
5479 // the allocation and the arraycopy causes a deoptimization, an
5480 // uninitialized array will escape the compiled method. To prevent that
5481 // we set the JVM state for uncommon traps between the allocation and
5482 // the arraycopy to the state before the allocation so, in case of
5483 // deoptimization, we'll reexecute the allocation and the
5484 // initialization.
5485 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5486 if (alloc != nullptr) {
5487 ciMethod* trap_method = alloc->jvms()->method();
5488 int trap_bci = alloc->jvms()->bci();
5489
5490 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5491 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5492 // Make sure there's no store between the allocation and the
5493 // arraycopy otherwise visible side effects could be rexecuted
5494 // in case of deoptimization and cause incorrect execution.
5495 bool no_interfering_store = true;
5496 Node* mem = alloc->in(TypeFunc::Memory);
5497 if (mem->is_MergeMem()) {
5498 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5499 Node* n = mms.memory();
5500 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5501 assert(n->is_Store(), "what else?");
5502 no_interfering_store = false;
5503 break;
5504 }
5505 }
5506 } else {
5507 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5508 Node* n = mms.memory();
5509 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5510 assert(n->is_Store(), "what else?");
5511 no_interfering_store = false;
5512 break;
5513 }
5514 }
5515 }
5516
5517 if (no_interfering_store) {
5518 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5519
5520 JVMState* saved_jvms = jvms();
5521 saved_reexecute_sp = _reexecute_sp;
5522
5523 set_jvms(sfpt->jvms());
5524 _reexecute_sp = jvms()->sp();
5525
5526 return saved_jvms;
5527 }
5528 }
5529 }
5530 return nullptr;
5531 }
5532
5533 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5534 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5535 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5536 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5537 uint size = alloc->req();
5538 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5539 old_jvms->set_map(sfpt);
5540 for (uint i = 0; i < size; i++) {
5541 sfpt->init_req(i, alloc->in(i));
5542 }
5543 // re-push array length for deoptimization
5544 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5545 old_jvms->set_sp(old_jvms->sp()+1);
5546 old_jvms->set_monoff(old_jvms->monoff()+1);
5547 old_jvms->set_scloff(old_jvms->scloff()+1);
5548 old_jvms->set_endoff(old_jvms->endoff()+1);
5549 old_jvms->set_should_reexecute(true);
5550
5551 sfpt->set_i_o(map()->i_o());
5552 sfpt->set_memory(map()->memory());
5553 sfpt->set_control(map()->control());
5554 return sfpt;
5555 }
5556
5557 // In case of a deoptimization, we restart execution at the
5558 // allocation, allocating a new array. We would leave an uninitialized
5559 // array in the heap that GCs wouldn't expect. Move the allocation
5560 // after the traps so we don't allocate the array if we
5561 // deoptimize. This is possible because tightly_coupled_allocation()
5562 // guarantees there's no observer of the allocated array at this point
5563 // and the control flow is simple enough.
5564 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5565 int saved_reexecute_sp, uint new_idx) {
5566 if (saved_jvms_before_guards != nullptr && !stopped()) {
5567 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5568
5569 assert(alloc != nullptr, "only with a tightly coupled allocation");
5570 // restore JVM state to the state at the arraycopy
5571 saved_jvms_before_guards->map()->set_control(map()->control());
5572 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5573 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5574 // If we've improved the types of some nodes (null check) while
5575 // emitting the guards, propagate them to the current state
5576 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5577 set_jvms(saved_jvms_before_guards);
5578 _reexecute_sp = saved_reexecute_sp;
5579
5580 // Remove the allocation from above the guards
5581 CallProjections callprojs;
5582 alloc->extract_projections(&callprojs, true);
5583 InitializeNode* init = alloc->initialization();
5584 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5585 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5586 init->replace_mem_projs_by(alloc_mem, C);
5587
5588 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5589 // the allocation (i.e. is only valid if the allocation succeeds):
5590 // 1) replace CastIINode with AllocateArrayNode's length here
5591 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5592 //
5593 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5594 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5595 Node* init_control = init->proj_out(TypeFunc::Control);
5596 Node* alloc_length = alloc->Ideal_length();
5597 #ifdef ASSERT
5598 Node* prev_cast = nullptr;
5599 #endif
5600 for (uint i = 0; i < init_control->outcnt(); i++) {
5601 Node* init_out = init_control->raw_out(i);
5602 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5603 #ifdef ASSERT
5604 if (prev_cast == nullptr) {
5605 prev_cast = init_out;
5606 } else {
5607 if (prev_cast->cmp(*init_out) == false) {
5608 prev_cast->dump();
5609 init_out->dump();
5610 assert(false, "not equal CastIINode");
5611 }
5612 }
5613 #endif
5614 C->gvn_replace_by(init_out, alloc_length);
5615 }
5616 }
5617 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5618
5619 // move the allocation here (after the guards)
5620 _gvn.hash_delete(alloc);
5621 alloc->set_req(TypeFunc::Control, control());
5622 alloc->set_req(TypeFunc::I_O, i_o());
5623 Node *mem = reset_memory();
5624 set_all_memory(mem);
5625 alloc->set_req(TypeFunc::Memory, mem);
5626 set_control(init->proj_out_or_null(TypeFunc::Control));
5627 set_i_o(callprojs.fallthrough_ioproj);
5628
5629 // Update memory as done in GraphKit::set_output_for_allocation()
5630 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5631 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5632 if (ary_type->isa_aryptr() && length_type != nullptr) {
5633 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5634 }
5635 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5636 int elemidx = C->get_alias_index(telemref);
5637 // Need to properly move every memory projection for the Initialize
5638 #ifdef ASSERT
5639 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5640 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
5641 #endif
5642 auto move_proj = [&](ProjNode* proj) {
5643 int alias_idx = C->get_alias_index(proj->adr_type());
5644 assert(alias_idx == Compile::AliasIdxRaw ||
5645 alias_idx == elemidx ||
5646 alias_idx == mark_idx ||
5647 alias_idx == klass_idx, "should be raw memory or array element type");
5648 set_memory(proj, alias_idx);
5649 };
5650 init->for_each_proj(move_proj, TypeFunc::Memory);
5651
5652 Node* allocx = _gvn.transform(alloc);
5653 assert(allocx == alloc, "where has the allocation gone?");
5654 assert(dest->is_CheckCastPP(), "not an allocation result?");
5655
5656 _gvn.hash_delete(dest);
5657 dest->set_req(0, control());
5658 Node* destx = _gvn.transform(dest);
5659 assert(destx == dest, "where has the allocation result gone?");
5660
5661 array_ideal_length(alloc, ary_type, true);
5662 }
5663 }
5664
5665 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5666 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5667 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5668 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5669 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5670 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5671 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5672 JVMState* saved_jvms_before_guards) {
5673 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5674 // There is at least one unrelated uncommon trap which needs to be replaced.
5675 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5676
5677 JVMState* saved_jvms = jvms();
5678 const int saved_reexecute_sp = _reexecute_sp;
5679 set_jvms(sfpt->jvms());
5680 _reexecute_sp = jvms()->sp();
5681
5682 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5683
5684 // Restore state
5685 set_jvms(saved_jvms);
5686 _reexecute_sp = saved_reexecute_sp;
5687 }
5688 }
5689
5690 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5691 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5692 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5693 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5694 while (if_proj->is_IfProj()) {
5695 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5696 if (uncommon_trap != nullptr) {
5697 create_new_uncommon_trap(uncommon_trap);
5698 }
5699 assert(if_proj->in(0)->is_If(), "must be If");
5700 if_proj = if_proj->in(0)->in(0);
5701 }
5702 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5703 "must have reached control projection of init node");
5704 }
5705
5706 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5707 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5708 assert(trap_request != 0, "no valid UCT trap request");
5709 PreserveJVMState pjvms(this);
5710 set_control(uncommon_trap_call->in(0));
5711 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5712 Deoptimization::trap_request_action(trap_request));
5713 assert(stopped(), "Should be stopped");
5714 _gvn.hash_delete(uncommon_trap_call);
5715 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5716 }
5717
5718 // Common checks for array sorting intrinsics arguments.
5719 // Returns `true` if checks passed.
5720 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5721 // check address of the class
5722 if (elementType == nullptr || elementType->is_top()) {
5723 return false; // dead path
5724 }
5725 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5726 if (elem_klass == nullptr) {
5727 return false; // dead path
5728 }
5729 // java_mirror_type() returns non-null for compile-time Class constants only
5730 ciType* elem_type = elem_klass->java_mirror_type();
5731 if (elem_type == nullptr) {
5732 return false;
5733 }
5734 bt = elem_type->basic_type();
5735 // Disable the intrinsic if the CPU does not support SIMD sort
5736 if (!Matcher::supports_simd_sort(bt)) {
5737 return false;
5738 }
5739 // check address of the array
5740 if (obj == nullptr || obj->is_top()) {
5741 return false; // dead path
5742 }
5743 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5744 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5745 return false; // failed input validation
5746 }
5747 return true;
5748 }
5749
5750 //------------------------------inline_array_partition-----------------------
5751 bool LibraryCallKit::inline_array_partition() {
5752 address stubAddr = StubRoutines::select_array_partition_function();
5753 if (stubAddr == nullptr) {
5754 return false; // Intrinsic's stub is not implemented on this platform
5755 }
5756 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5757
5758 // no receiver because it is a static method
5759 Node* elementType = argument(0);
5760 Node* obj = argument(1);
5761 Node* offset = argument(2); // long
5762 Node* fromIndex = argument(4);
5763 Node* toIndex = argument(5);
5764 Node* indexPivot1 = argument(6);
5765 Node* indexPivot2 = argument(7);
5766 // PartitionOperation: argument(8) is ignored
5767
5768 Node* pivotIndices = nullptr;
5769 BasicType bt = T_ILLEGAL;
5770
5771 if (!check_array_sort_arguments(elementType, obj, bt)) {
5772 return false;
5773 }
5774 null_check(obj);
5775 // If obj is dead, only null-path is taken.
5776 if (stopped()) {
5777 return true;
5778 }
5779 // Set the original stack and the reexecute bit for the interpreter to reexecute
5780 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5781 { PreserveReexecuteState preexecs(this);
5782 jvms()->set_should_reexecute(true);
5783
5784 Node* obj_adr = make_unsafe_address(obj, offset);
5785
5786 // create the pivotIndices array of type int and size = 2
5787 Node* size = intcon(2);
5788 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5789 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5790 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5791 guarantee(alloc != nullptr, "created above");
5792 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5793
5794 // pass the basic type enum to the stub
5795 Node* elemType = intcon(bt);
5796
5797 // Call the stub
5798 const char *stubName = "array_partition_stub";
5799 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5800 stubAddr, stubName, TypePtr::BOTTOM,
5801 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5802 indexPivot1, indexPivot2);
5803
5804 } // original reexecute is set back here
5805
5806 if (!stopped()) {
5807 set_result(pivotIndices);
5808 }
5809
5810 return true;
5811 }
5812
5813
5814 //------------------------------inline_array_sort-----------------------
5815 bool LibraryCallKit::inline_array_sort() {
5816 address stubAddr = StubRoutines::select_arraysort_function();
5817 if (stubAddr == nullptr) {
5818 return false; // Intrinsic's stub is not implemented on this platform
5819 }
5820 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5821
5822 // no receiver because it is a static method
5823 Node* elementType = argument(0);
5824 Node* obj = argument(1);
5825 Node* offset = argument(2); // long
5826 Node* fromIndex = argument(4);
5827 Node* toIndex = argument(5);
5828 // SortOperation: argument(6) is ignored
5829
5830 BasicType bt = T_ILLEGAL;
5831
5832 if (!check_array_sort_arguments(elementType, obj, bt)) {
5833 return false;
5834 }
5835 null_check(obj);
5836 // If obj is dead, only null-path is taken.
5837 if (stopped()) {
5838 return true;
5839 }
5840 Node* obj_adr = make_unsafe_address(obj, offset);
5841
5842 // pass the basic type enum to the stub
5843 Node* elemType = intcon(bt);
5844
5845 // Call the stub.
5846 const char *stubName = "arraysort_stub";
5847 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5848 stubAddr, stubName, TypePtr::BOTTOM,
5849 obj_adr, elemType, fromIndex, toIndex);
5850
5851 return true;
5852 }
5853
5854
5855 //------------------------------inline_arraycopy-----------------------
5856 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5857 // Object dest, int destPos,
5858 // int length);
5859 bool LibraryCallKit::inline_arraycopy() {
5860 // Get the arguments.
5861 Node* src = argument(0); // type: oop
5862 Node* src_offset = argument(1); // type: int
5863 Node* dest = argument(2); // type: oop
5864 Node* dest_offset = argument(3); // type: int
5865 Node* length = argument(4); // type: int
5866
5867 uint new_idx = C->unique();
5868
5869 // Check for allocation before we add nodes that would confuse
5870 // tightly_coupled_allocation()
5871 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5872
5873 int saved_reexecute_sp = -1;
5874 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5875 // See arraycopy_restore_alloc_state() comment
5876 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5877 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5878 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5879 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5880
5881 // The following tests must be performed
5882 // (1) src and dest are arrays.
5883 // (2) src and dest arrays must have elements of the same BasicType
5884 // (3) src and dest must not be null.
5885 // (4) src_offset must not be negative.
5886 // (5) dest_offset must not be negative.
5887 // (6) length must not be negative.
5888 // (7) src_offset + length must not exceed length of src.
5889 // (8) dest_offset + length must not exceed length of dest.
5890 // (9) each element of an oop array must be assignable
5891
5892 // (3) src and dest must not be null.
5893 // always do this here because we need the JVM state for uncommon traps
5894 Node* null_ctl = top();
5895 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5896 assert(null_ctl->is_top(), "no null control here");
5897 dest = null_check(dest, T_ARRAY);
5898
5899 if (!can_emit_guards) {
5900 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5901 // guards but the arraycopy node could still take advantage of a
5902 // tightly allocated allocation. tightly_coupled_allocation() is
5903 // called again to make sure it takes the null check above into
5904 // account: the null check is mandatory and if it caused an
5905 // uncommon trap to be emitted then the allocation can't be
5906 // considered tightly coupled in this context.
5907 alloc = tightly_coupled_allocation(dest);
5908 }
5909
5910 bool validated = false;
5911
5912 const Type* src_type = _gvn.type(src);
5913 const Type* dest_type = _gvn.type(dest);
5914 const TypeAryPtr* top_src = src_type->isa_aryptr();
5915 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5916
5917 // Do we have the type of src?
5918 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5919 // Do we have the type of dest?
5920 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5921 // Is the type for src from speculation?
5922 bool src_spec = false;
5923 // Is the type for dest from speculation?
5924 bool dest_spec = false;
5925
5926 if ((!has_src || !has_dest) && can_emit_guards) {
5927 // We don't have sufficient type information, let's see if
5928 // speculative types can help. We need to have types for both src
5929 // and dest so that it pays off.
5930
5931 // Do we already have or could we have type information for src
5932 bool could_have_src = has_src;
5933 // Do we already have or could we have type information for dest
5934 bool could_have_dest = has_dest;
5935
5936 ciKlass* src_k = nullptr;
5937 if (!has_src) {
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
5944 ciKlass* dest_k = nullptr;
5945 if (!has_dest) {
5946 dest_k = dest_type->speculative_type_not_null();
5947 if (dest_k != nullptr && dest_k->is_array_klass()) {
5948 could_have_dest = true;
5949 }
5950 }
5951
5952 if (could_have_src && could_have_dest) {
5953 // This is going to pay off so emit the required guards
5954 if (!has_src) {
5955 src = maybe_cast_profiled_obj(src, src_k, true);
5956 src_type = _gvn.type(src);
5957 top_src = src_type->isa_aryptr();
5958 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5959 src_spec = true;
5960 }
5961 if (!has_dest) {
5962 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5963 dest_type = _gvn.type(dest);
5964 top_dest = dest_type->isa_aryptr();
5965 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5966 dest_spec = true;
5967 }
5968 }
5969 }
5970
5971 if (has_src && has_dest && can_emit_guards) {
5972 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5973 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5974 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5975 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5976
5977 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5978 // If both arrays are object arrays then having the exact types
5979 // for both will remove the need for a subtype check at runtime
5980 // before the call and may make it possible to pick a faster copy
5981 // routine (without a subtype check on every element)
5982 // Do we have the exact type of src?
5983 bool could_have_src = src_spec;
5984 // Do we have the exact type of dest?
5985 bool could_have_dest = dest_spec;
5986 ciKlass* src_k = nullptr;
5987 ciKlass* dest_k = nullptr;
5988 if (!src_spec) {
5989 src_k = src_type->speculative_type_not_null();
5990 if (src_k != nullptr && src_k->is_array_klass()) {
5991 could_have_src = true;
5992 }
5993 }
5994 if (!dest_spec) {
5995 dest_k = dest_type->speculative_type_not_null();
5996 if (dest_k != nullptr && dest_k->is_array_klass()) {
5997 could_have_dest = true;
5998 }
5999 }
6000 if (could_have_src && could_have_dest) {
6001 // If we can have both exact types, emit the missing guards
6002 if (could_have_src && !src_spec) {
6003 src = maybe_cast_profiled_obj(src, src_k, true);
6004 }
6005 if (could_have_dest && !dest_spec) {
6006 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6007 }
6008 }
6009 }
6010 }
6011
6012 ciMethod* trap_method = method();
6013 int trap_bci = bci();
6014 if (saved_jvms_before_guards != nullptr) {
6015 trap_method = alloc->jvms()->method();
6016 trap_bci = alloc->jvms()->bci();
6017 }
6018
6019 bool negative_length_guard_generated = false;
6020
6021 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6022 can_emit_guards &&
6023 !src->is_top() && !dest->is_top()) {
6024 // validate arguments: enables transformation the ArrayCopyNode
6025 validated = true;
6026
6027 RegionNode* slow_region = new RegionNode(1);
6028 record_for_igvn(slow_region);
6029
6030 // (1) src and dest are arrays.
6031 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6032 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6033
6034 // (2) src and dest arrays must have elements of the same BasicType
6035 // done at macro expansion or at Ideal transformation time
6036
6037 // (4) src_offset must not be negative.
6038 generate_negative_guard(src_offset, slow_region);
6039
6040 // (5) dest_offset must not be negative.
6041 generate_negative_guard(dest_offset, slow_region);
6042
6043 // (7) src_offset + length must not exceed length of src.
6044 generate_limit_guard(src_offset, length,
6045 load_array_length(src),
6046 slow_region);
6047
6048 // (8) dest_offset + length must not exceed length of dest.
6049 generate_limit_guard(dest_offset, length,
6050 load_array_length(dest),
6051 slow_region);
6052
6053 // (6) length must not be negative.
6054 // This is also checked in generate_arraycopy() during macro expansion, but
6055 // we also have to check it here for the case where the ArrayCopyNode will
6056 // be eliminated by Escape Analysis.
6057 if (EliminateAllocations) {
6058 generate_negative_guard(length, slow_region);
6059 negative_length_guard_generated = true;
6060 }
6061
6062 // (9) each element of an oop array must be assignable
6063 Node* dest_klass = load_object_klass(dest);
6064 if (src != dest) {
6065 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6066
6067 if (not_subtype_ctrl != top()) {
6068 PreserveJVMState pjvms(this);
6069 set_control(not_subtype_ctrl);
6070 uncommon_trap(Deoptimization::Reason_intrinsic,
6071 Deoptimization::Action_make_not_entrant);
6072 assert(stopped(), "Should be stopped");
6073 }
6074 }
6075 {
6076 PreserveJVMState pjvms(this);
6077 set_control(_gvn.transform(slow_region));
6078 uncommon_trap(Deoptimization::Reason_intrinsic,
6079 Deoptimization::Action_make_not_entrant);
6080 assert(stopped(), "Should be stopped");
6081 }
6082
6083 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6084 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6085 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6086 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6087 }
6088
6089 if (stopped()) {
6090 return true;
6091 }
6092
6093 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6094 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6095 // so the compiler has a chance to eliminate them: during macro expansion,
6096 // we have to set their control (CastPP nodes are eliminated).
6097 load_object_klass(src), load_object_klass(dest),
6098 load_array_length(src), load_array_length(dest));
6099
6100 ac->set_arraycopy(validated);
6101
6102 Node* n = _gvn.transform(ac);
6103 if (n == ac) {
6104 ac->connect_outputs(this);
6105 } else {
6106 assert(validated, "shouldn't transform if all arguments not validated");
6107 set_all_memory(n);
6108 }
6109 clear_upper_avx();
6110
6111
6112 return true;
6113 }
6114
6115
6116 // Helper function which determines if an arraycopy immediately follows
6117 // an allocation, with no intervening tests or other escapes for the object.
6118 AllocateArrayNode*
6119 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6120 if (stopped()) return nullptr; // no fast path
6121 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6122
6123 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6124 if (alloc == nullptr) return nullptr;
6125
6126 Node* rawmem = memory(Compile::AliasIdxRaw);
6127 // Is the allocation's memory state untouched?
6128 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6129 // Bail out if there have been raw-memory effects since the allocation.
6130 // (Example: There might have been a call or safepoint.)
6131 return nullptr;
6132 }
6133 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6134 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6135 return nullptr;
6136 }
6137
6138 // There must be no unexpected observers of this allocation.
6139 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6140 Node* obs = ptr->fast_out(i);
6141 if (obs != this->map()) {
6142 return nullptr;
6143 }
6144 }
6145
6146 // This arraycopy must unconditionally follow the allocation of the ptr.
6147 Node* alloc_ctl = ptr->in(0);
6148 Node* ctl = control();
6149 while (ctl != alloc_ctl) {
6150 // There may be guards which feed into the slow_region.
6151 // Any other control flow means that we might not get a chance
6152 // to finish initializing the allocated object.
6153 // Various low-level checks bottom out in uncommon traps. These
6154 // are considered safe since we've already checked above that
6155 // there is no unexpected observer of this allocation.
6156 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6157 assert(ctl->in(0)->is_If(), "must be If");
6158 ctl = ctl->in(0)->in(0);
6159 } else {
6160 return nullptr;
6161 }
6162 }
6163
6164 // If we get this far, we have an allocation which immediately
6165 // precedes the arraycopy, and we can take over zeroing the new object.
6166 // The arraycopy will finish the initialization, and provide
6167 // a new control state to which we will anchor the destination pointer.
6168
6169 return alloc;
6170 }
6171
6172 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6173 if (node->is_IfProj()) {
6174 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6175 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6176 Node* obs = other_proj->fast_out(j);
6177 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6178 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6179 return obs->as_CallStaticJava();
6180 }
6181 }
6182 }
6183 return nullptr;
6184 }
6185
6186 //-------------inline_encodeISOArray-----------------------------------
6187 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6188 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6189 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6190 // encode char[] to byte[] in ISO_8859_1 or ASCII
6191 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6192 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6193 // no receiver since it is static method
6194 Node *src = argument(0);
6195 Node *src_offset = argument(1);
6196 Node *dst = argument(2);
6197 Node *dst_offset = argument(3);
6198 Node *length = argument(4);
6199
6200 // Cast source & target arrays to not-null
6201 if (VerifyIntrinsicChecks) {
6202 src = must_be_not_null(src, true);
6203 dst = must_be_not_null(dst, true);
6204 if (stopped()) {
6205 return true;
6206 }
6207 }
6208
6209 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6210 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6211 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6212 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6213 // failed array check
6214 return false;
6215 }
6216
6217 // Figure out the size and type of the elements we will be copying.
6218 BasicType src_elem = src_type->elem()->array_element_basic_type();
6219 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6220 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6221 return false;
6222 }
6223
6224 // Check source & target bounds
6225 if (VerifyIntrinsicChecks) {
6226 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6227 generate_string_range_check(dst, dst_offset, length, false, true);
6228 if (stopped()) {
6229 return true;
6230 }
6231 }
6232
6233 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6234 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6235 // 'src_start' points to src array + scaled offset
6236 // 'dst_start' points to dst array + scaled offset
6237
6238 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6239 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6240 enc = _gvn.transform(enc);
6241 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6242 set_memory(res_mem, mtype);
6243 set_result(enc);
6244 clear_upper_avx();
6245
6246 return true;
6247 }
6248
6249 //-------------inline_multiplyToLen-----------------------------------
6250 bool LibraryCallKit::inline_multiplyToLen() {
6251 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6252
6253 address stubAddr = StubRoutines::multiplyToLen();
6254 if (stubAddr == nullptr) {
6255 return false; // Intrinsic's stub is not implemented on this platform
6256 }
6257 const char* stubName = "multiplyToLen";
6258
6259 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6260
6261 // no receiver because it is a static method
6262 Node* x = argument(0);
6263 Node* xlen = argument(1);
6264 Node* y = argument(2);
6265 Node* ylen = argument(3);
6266 Node* z = argument(4);
6267
6268 x = must_be_not_null(x, true);
6269 y = must_be_not_null(y, true);
6270
6271 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6272 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6273 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6274 y_type == nullptr || y_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 y_elem = y_type->elem()->array_element_basic_type();
6281 if (x_elem != T_INT || y_elem != T_INT) {
6282 return false;
6283 }
6284
6285 Node* x_start = array_element_address(x, intcon(0), x_elem);
6286 Node* y_start = array_element_address(y, intcon(0), y_elem);
6287 // 'x_start' points to x array + scaled xlen
6288 // 'y_start' points to y array + scaled ylen
6289
6290 Node* z_start = array_element_address(z, intcon(0), T_INT);
6291
6292 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6293 OptoRuntime::multiplyToLen_Type(),
6294 stubAddr, stubName, TypePtr::BOTTOM,
6295 x_start, xlen, y_start, ylen, z_start);
6296
6297 C->set_has_split_ifs(true); // Has chance for split-if optimization
6298 set_result(z);
6299 return true;
6300 }
6301
6302 //-------------inline_squareToLen------------------------------------
6303 bool LibraryCallKit::inline_squareToLen() {
6304 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6305
6306 address stubAddr = StubRoutines::squareToLen();
6307 if (stubAddr == nullptr) {
6308 return false; // Intrinsic's stub is not implemented on this platform
6309 }
6310 const char* stubName = "squareToLen";
6311
6312 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6313
6314 Node* x = argument(0);
6315 Node* len = argument(1);
6316 Node* z = argument(2);
6317 Node* zlen = argument(3);
6318
6319 x = must_be_not_null(x, true);
6320 z = must_be_not_null(z, true);
6321
6322 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6323 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6324 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6325 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6326 // failed array check
6327 return false;
6328 }
6329
6330 BasicType x_elem = x_type->elem()->array_element_basic_type();
6331 BasicType z_elem = z_type->elem()->array_element_basic_type();
6332 if (x_elem != T_INT || z_elem != T_INT) {
6333 return false;
6334 }
6335
6336
6337 Node* x_start = array_element_address(x, intcon(0), x_elem);
6338 Node* z_start = array_element_address(z, intcon(0), z_elem);
6339
6340 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6341 OptoRuntime::squareToLen_Type(),
6342 stubAddr, stubName, TypePtr::BOTTOM,
6343 x_start, len, z_start, zlen);
6344
6345 set_result(z);
6346 return true;
6347 }
6348
6349 //-------------inline_mulAdd------------------------------------------
6350 bool LibraryCallKit::inline_mulAdd() {
6351 assert(UseMulAddIntrinsic, "not implemented on this platform");
6352
6353 address stubAddr = StubRoutines::mulAdd();
6354 if (stubAddr == nullptr) {
6355 return false; // Intrinsic's stub is not implemented on this platform
6356 }
6357 const char* stubName = "mulAdd";
6358
6359 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6360
6361 Node* out = argument(0);
6362 Node* in = argument(1);
6363 Node* offset = argument(2);
6364 Node* len = argument(3);
6365 Node* k = argument(4);
6366
6367 in = must_be_not_null(in, true);
6368 out = must_be_not_null(out, true);
6369
6370 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6371 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6372 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6373 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6374 // failed array check
6375 return false;
6376 }
6377
6378 BasicType out_elem = out_type->elem()->array_element_basic_type();
6379 BasicType in_elem = in_type->elem()->array_element_basic_type();
6380 if (out_elem != T_INT || in_elem != T_INT) {
6381 return false;
6382 }
6383
6384 Node* outlen = load_array_length(out);
6385 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6386 Node* out_start = array_element_address(out, intcon(0), out_elem);
6387 Node* in_start = array_element_address(in, intcon(0), in_elem);
6388
6389 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6390 OptoRuntime::mulAdd_Type(),
6391 stubAddr, stubName, TypePtr::BOTTOM,
6392 out_start,in_start, new_offset, len, k);
6393 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6394 set_result(result);
6395 return true;
6396 }
6397
6398 //-------------inline_montgomeryMultiply-----------------------------------
6399 bool LibraryCallKit::inline_montgomeryMultiply() {
6400 address stubAddr = StubRoutines::montgomeryMultiply();
6401 if (stubAddr == nullptr) {
6402 return false; // Intrinsic's stub is not implemented on this platform
6403 }
6404
6405 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6406 const char* stubName = "montgomery_multiply";
6407
6408 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6409
6410 Node* a = argument(0);
6411 Node* b = argument(1);
6412 Node* n = argument(2);
6413 Node* len = argument(3);
6414 Node* inv = argument(4);
6415 Node* m = argument(6);
6416
6417 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6418 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6419 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6420 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6421 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6422 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6423 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6424 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6425 // failed array check
6426 return false;
6427 }
6428
6429 BasicType a_elem = a_type->elem()->array_element_basic_type();
6430 BasicType b_elem = b_type->elem()->array_element_basic_type();
6431 BasicType n_elem = n_type->elem()->array_element_basic_type();
6432 BasicType m_elem = m_type->elem()->array_element_basic_type();
6433 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6434 return false;
6435 }
6436
6437 // Make the call
6438 {
6439 Node* a_start = array_element_address(a, intcon(0), a_elem);
6440 Node* b_start = array_element_address(b, intcon(0), b_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::montgomeryMultiply_Type(),
6446 stubAddr, stubName, TypePtr::BOTTOM,
6447 a_start, b_start, n_start, len, inv, top(),
6448 m_start);
6449 set_result(m);
6450 }
6451
6452 return true;
6453 }
6454
6455 bool LibraryCallKit::inline_montgomerySquare() {
6456 address stubAddr = StubRoutines::montgomerySquare();
6457 if (stubAddr == nullptr) {
6458 return false; // Intrinsic's stub is not implemented on this platform
6459 }
6460
6461 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6462 const char* stubName = "montgomery_square";
6463
6464 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6465
6466 Node* a = argument(0);
6467 Node* n = argument(1);
6468 Node* len = argument(2);
6469 Node* inv = argument(3);
6470 Node* m = argument(5);
6471
6472 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6473 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6474 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6475 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6476 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6477 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6478 // failed array check
6479 return false;
6480 }
6481
6482 BasicType a_elem = a_type->elem()->array_element_basic_type();
6483 BasicType n_elem = n_type->elem()->array_element_basic_type();
6484 BasicType m_elem = m_type->elem()->array_element_basic_type();
6485 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6486 return false;
6487 }
6488
6489 // Make the call
6490 {
6491 Node* a_start = array_element_address(a, intcon(0), a_elem);
6492 Node* n_start = array_element_address(n, intcon(0), n_elem);
6493 Node* m_start = array_element_address(m, intcon(0), m_elem);
6494
6495 Node* call = make_runtime_call(RC_LEAF,
6496 OptoRuntime::montgomerySquare_Type(),
6497 stubAddr, stubName, TypePtr::BOTTOM,
6498 a_start, n_start, len, inv, top(),
6499 m_start);
6500 set_result(m);
6501 }
6502
6503 return true;
6504 }
6505
6506 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6507 address stubAddr = nullptr;
6508 const char* stubName = nullptr;
6509
6510 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6511 if (stubAddr == nullptr) {
6512 return false; // Intrinsic's stub is not implemented on this platform
6513 }
6514
6515 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6516
6517 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6518
6519 Node* newArr = argument(0);
6520 Node* oldArr = argument(1);
6521 Node* newIdx = argument(2);
6522 Node* shiftCount = argument(3);
6523 Node* numIter = argument(4);
6524
6525 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6526 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6527 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6528 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6529 return false;
6530 }
6531
6532 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6533 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6534 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6535 return false;
6536 }
6537
6538 // Make the call
6539 {
6540 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6541 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6542
6543 Node* call = make_runtime_call(RC_LEAF,
6544 OptoRuntime::bigIntegerShift_Type(),
6545 stubAddr,
6546 stubName,
6547 TypePtr::BOTTOM,
6548 newArr_start,
6549 oldArr_start,
6550 newIdx,
6551 shiftCount,
6552 numIter);
6553 }
6554
6555 return true;
6556 }
6557
6558 //-------------inline_vectorizedMismatch------------------------------
6559 bool LibraryCallKit::inline_vectorizedMismatch() {
6560 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6561
6562 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6563 Node* obja = argument(0); // Object
6564 Node* aoffset = argument(1); // long
6565 Node* objb = argument(3); // Object
6566 Node* boffset = argument(4); // long
6567 Node* length = argument(6); // int
6568 Node* scale = argument(7); // int
6569
6570 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6571 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6572 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6573 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6574 scale == top()) {
6575 return false; // failed input validation
6576 }
6577
6578 Node* obja_adr = make_unsafe_address(obja, aoffset);
6579 Node* objb_adr = make_unsafe_address(objb, boffset);
6580
6581 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6582 //
6583 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6584 // if (length <= inline_limit) {
6585 // inline_path:
6586 // vmask = VectorMaskGen length
6587 // vload1 = LoadVectorMasked obja, vmask
6588 // vload2 = LoadVectorMasked objb, vmask
6589 // result1 = VectorCmpMasked vload1, vload2, vmask
6590 // } else {
6591 // call_stub_path:
6592 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6593 // }
6594 // exit_block:
6595 // return Phi(result1, result2);
6596 //
6597 enum { inline_path = 1, // input is small enough to process it all at once
6598 stub_path = 2, // input is too large; call into the VM
6599 PATH_LIMIT = 3
6600 };
6601
6602 Node* exit_block = new RegionNode(PATH_LIMIT);
6603 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6604 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6605
6606 Node* call_stub_path = control();
6607
6608 BasicType elem_bt = T_ILLEGAL;
6609
6610 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6611 if (scale_t->is_con()) {
6612 switch (scale_t->get_con()) {
6613 case 0: elem_bt = T_BYTE; break;
6614 case 1: elem_bt = T_SHORT; break;
6615 case 2: elem_bt = T_INT; break;
6616 case 3: elem_bt = T_LONG; break;
6617
6618 default: elem_bt = T_ILLEGAL; break; // not supported
6619 }
6620 }
6621
6622 int inline_limit = 0;
6623 bool do_partial_inline = false;
6624
6625 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6626 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6627 do_partial_inline = inline_limit >= 16;
6628 }
6629
6630 if (do_partial_inline) {
6631 assert(elem_bt != T_ILLEGAL, "sanity");
6632
6633 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6634 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6635 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6636
6637 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6638 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6639 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6640
6641 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6642
6643 if (!stopped()) {
6644 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6645
6646 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6647 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6648 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6649 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6650
6651 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6652 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6653 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6654 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6655
6656 exit_block->init_req(inline_path, control());
6657 memory_phi->init_req(inline_path, map()->memory());
6658 result_phi->init_req(inline_path, result);
6659
6660 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6661 clear_upper_avx();
6662 }
6663 }
6664 }
6665
6666 if (call_stub_path != nullptr) {
6667 set_control(call_stub_path);
6668
6669 Node* call = make_runtime_call(RC_LEAF,
6670 OptoRuntime::vectorizedMismatch_Type(),
6671 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6672 obja_adr, objb_adr, length, scale);
6673
6674 exit_block->init_req(stub_path, control());
6675 memory_phi->init_req(stub_path, map()->memory());
6676 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6677 }
6678
6679 exit_block = _gvn.transform(exit_block);
6680 memory_phi = _gvn.transform(memory_phi);
6681 result_phi = _gvn.transform(result_phi);
6682
6683 record_for_igvn(exit_block);
6684 record_for_igvn(memory_phi);
6685 record_for_igvn(result_phi);
6686
6687 set_control(exit_block);
6688 set_all_memory(memory_phi);
6689 set_result(result_phi);
6690
6691 return true;
6692 }
6693
6694 //------------------------------inline_vectorizedHashcode----------------------------
6695 bool LibraryCallKit::inline_vectorizedHashCode() {
6696 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6697
6698 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6699 Node* array = argument(0);
6700 Node* offset = argument(1);
6701 Node* length = argument(2);
6702 Node* initialValue = argument(3);
6703 Node* basic_type = argument(4);
6704
6705 if (basic_type == top()) {
6706 return false; // failed input validation
6707 }
6708
6709 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6710 if (!basic_type_t->is_con()) {
6711 return false; // Only intrinsify if mode argument is constant
6712 }
6713
6714 array = must_be_not_null(array, true);
6715
6716 BasicType bt = (BasicType)basic_type_t->get_con();
6717
6718 // Resolve address of first element
6719 Node* array_start = array_element_address(array, offset, bt);
6720
6721 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6722 array_start, length, initialValue, basic_type)));
6723 clear_upper_avx();
6724
6725 return true;
6726 }
6727
6728 /**
6729 * Calculate CRC32 for byte.
6730 * int java.util.zip.CRC32.update(int crc, int b)
6731 */
6732 bool LibraryCallKit::inline_updateCRC32() {
6733 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6734 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6735 // no receiver since it is static method
6736 Node* crc = argument(0); // type: int
6737 Node* b = argument(1); // type: int
6738
6739 /*
6740 * int c = ~ crc;
6741 * b = timesXtoThe32[(b ^ c) & 0xFF];
6742 * b = b ^ (c >>> 8);
6743 * crc = ~b;
6744 */
6745
6746 Node* M1 = intcon(-1);
6747 crc = _gvn.transform(new XorINode(crc, M1));
6748 Node* result = _gvn.transform(new XorINode(crc, b));
6749 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6750
6751 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6752 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6753 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6754 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6755
6756 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6757 result = _gvn.transform(new XorINode(crc, result));
6758 result = _gvn.transform(new XorINode(result, M1));
6759 set_result(result);
6760 return true;
6761 }
6762
6763 /**
6764 * Calculate CRC32 for byte[] array.
6765 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6766 */
6767 bool LibraryCallKit::inline_updateBytesCRC32() {
6768 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6769 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6770 // no receiver since it is static method
6771 Node* crc = argument(0); // type: int
6772 Node* src = argument(1); // type: oop
6773 Node* offset = argument(2); // type: int
6774 Node* length = argument(3); // type: int
6775
6776 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6777 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6778 // failed array check
6779 return false;
6780 }
6781
6782 // Figure out the size and type of the elements we will be copying.
6783 BasicType src_elem = src_type->elem()->array_element_basic_type();
6784 if (src_elem != T_BYTE) {
6785 return false;
6786 }
6787
6788 // 'src_start' points to src array + scaled offset
6789 src = must_be_not_null(src, true);
6790 Node* src_start = array_element_address(src, offset, src_elem);
6791
6792 // We assume that range check is done by caller.
6793 // TODO: generate range check (offset+length < src.length) in debug VM.
6794
6795 // Call the stub.
6796 address stubAddr = StubRoutines::updateBytesCRC32();
6797 const char *stubName = "updateBytesCRC32";
6798
6799 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6800 stubAddr, stubName, TypePtr::BOTTOM,
6801 crc, src_start, length);
6802 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6803 set_result(result);
6804 return true;
6805 }
6806
6807 /**
6808 * Calculate CRC32 for ByteBuffer.
6809 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6810 */
6811 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6812 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6813 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6814 // no receiver since it is static method
6815 Node* crc = argument(0); // type: int
6816 Node* src = argument(1); // type: long
6817 Node* offset = argument(3); // type: int
6818 Node* length = argument(4); // type: int
6819
6820 src = ConvL2X(src); // adjust Java long to machine word
6821 Node* base = _gvn.transform(new CastX2PNode(src));
6822 offset = ConvI2X(offset);
6823
6824 // 'src_start' points to src array + scaled offset
6825 Node* src_start = basic_plus_adr(top(), base, offset);
6826
6827 // Call the stub.
6828 address stubAddr = StubRoutines::updateBytesCRC32();
6829 const char *stubName = "updateBytesCRC32";
6830
6831 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6832 stubAddr, stubName, TypePtr::BOTTOM,
6833 crc, src_start, length);
6834 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6835 set_result(result);
6836 return true;
6837 }
6838
6839 //------------------------------get_table_from_crc32c_class-----------------------
6840 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6841 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6842 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6843
6844 return table;
6845 }
6846
6847 //------------------------------inline_updateBytesCRC32C-----------------------
6848 //
6849 // Calculate CRC32C for byte[] array.
6850 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6851 //
6852 bool LibraryCallKit::inline_updateBytesCRC32C() {
6853 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6854 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6855 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6856 // no receiver since it is a static method
6857 Node* crc = argument(0); // type: int
6858 Node* src = argument(1); // type: oop
6859 Node* offset = argument(2); // type: int
6860 Node* end = argument(3); // type: int
6861
6862 Node* length = _gvn.transform(new SubINode(end, offset));
6863
6864 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6865 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6866 // failed array check
6867 return false;
6868 }
6869
6870 // Figure out the size and type of the elements we will be copying.
6871 BasicType src_elem = src_type->elem()->array_element_basic_type();
6872 if (src_elem != T_BYTE) {
6873 return false;
6874 }
6875
6876 // 'src_start' points to src array + scaled offset
6877 src = must_be_not_null(src, true);
6878 Node* src_start = array_element_address(src, offset, src_elem);
6879
6880 // static final int[] byteTable in class CRC32C
6881 Node* table = get_table_from_crc32c_class(callee()->holder());
6882 table = must_be_not_null(table, true);
6883 Node* table_start = array_element_address(table, intcon(0), T_INT);
6884
6885 // We assume that range check is done by caller.
6886 // TODO: generate range check (offset+length < src.length) in debug VM.
6887
6888 // Call the stub.
6889 address stubAddr = StubRoutines::updateBytesCRC32C();
6890 const char *stubName = "updateBytesCRC32C";
6891
6892 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6893 stubAddr, stubName, TypePtr::BOTTOM,
6894 crc, src_start, length, table_start);
6895 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6896 set_result(result);
6897 return true;
6898 }
6899
6900 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6901 //
6902 // Calculate CRC32C for DirectByteBuffer.
6903 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6904 //
6905 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6906 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6907 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6908 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6909 // no receiver since it is a static method
6910 Node* crc = argument(0); // type: int
6911 Node* src = argument(1); // type: long
6912 Node* offset = argument(3); // type: int
6913 Node* end = argument(4); // type: int
6914
6915 Node* length = _gvn.transform(new SubINode(end, offset));
6916
6917 src = ConvL2X(src); // adjust Java long to machine word
6918 Node* base = _gvn.transform(new CastX2PNode(src));
6919 offset = ConvI2X(offset);
6920
6921 // 'src_start' points to src array + scaled offset
6922 Node* src_start = basic_plus_adr(top(), base, offset);
6923
6924 // static final int[] byteTable in class CRC32C
6925 Node* table = get_table_from_crc32c_class(callee()->holder());
6926 table = must_be_not_null(table, true);
6927 Node* table_start = array_element_address(table, intcon(0), T_INT);
6928
6929 // Call the stub.
6930 address stubAddr = StubRoutines::updateBytesCRC32C();
6931 const char *stubName = "updateBytesCRC32C";
6932
6933 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6934 stubAddr, stubName, TypePtr::BOTTOM,
6935 crc, src_start, length, table_start);
6936 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6937 set_result(result);
6938 return true;
6939 }
6940
6941 //------------------------------inline_updateBytesAdler32----------------------
6942 //
6943 // Calculate Adler32 checksum for byte[] array.
6944 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6945 //
6946 bool LibraryCallKit::inline_updateBytesAdler32() {
6947 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6948 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6949 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6950 // no receiver since it is static method
6951 Node* crc = argument(0); // type: int
6952 Node* src = argument(1); // type: oop
6953 Node* offset = argument(2); // type: int
6954 Node* length = argument(3); // type: int
6955
6956 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6957 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6958 // failed array check
6959 return false;
6960 }
6961
6962 // Figure out the size and type of the elements we will be copying.
6963 BasicType src_elem = src_type->elem()->array_element_basic_type();
6964 if (src_elem != T_BYTE) {
6965 return false;
6966 }
6967
6968 // 'src_start' points to src array + scaled offset
6969 Node* src_start = array_element_address(src, offset, src_elem);
6970
6971 // We assume that range check is done by caller.
6972 // TODO: generate range check (offset+length < src.length) in debug VM.
6973
6974 // Call the stub.
6975 address stubAddr = StubRoutines::updateBytesAdler32();
6976 const char *stubName = "updateBytesAdler32";
6977
6978 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6979 stubAddr, stubName, TypePtr::BOTTOM,
6980 crc, src_start, length);
6981 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6982 set_result(result);
6983 return true;
6984 }
6985
6986 //------------------------------inline_updateByteBufferAdler32---------------
6987 //
6988 // Calculate Adler32 checksum for DirectByteBuffer.
6989 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6990 //
6991 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6992 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6993 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6994 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6995 // no receiver since it is static method
6996 Node* crc = argument(0); // type: int
6997 Node* src = argument(1); // type: long
6998 Node* offset = argument(3); // type: int
6999 Node* length = argument(4); // type: int
7000
7001 src = ConvL2X(src); // adjust Java long to machine word
7002 Node* base = _gvn.transform(new CastX2PNode(src));
7003 offset = ConvI2X(offset);
7004
7005 // 'src_start' points to src array + scaled offset
7006 Node* src_start = basic_plus_adr(top(), base, offset);
7007
7008 // Call the stub.
7009 address stubAddr = StubRoutines::updateBytesAdler32();
7010 const char *stubName = "updateBytesAdler32";
7011
7012 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7013 stubAddr, stubName, TypePtr::BOTTOM,
7014 crc, src_start, length);
7015
7016 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7017 set_result(result);
7018 return true;
7019 }
7020
7021 //----------------------------inline_reference_get0----------------------------
7022 // public T java.lang.ref.Reference.get();
7023 bool LibraryCallKit::inline_reference_get0() {
7024 const int referent_offset = java_lang_ref_Reference::referent_offset();
7025
7026 // Get the argument:
7027 Node* reference_obj = null_check_receiver();
7028 if (stopped()) return true;
7029
7030 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7031 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7032 decorators, /*is_static*/ false, nullptr);
7033 if (result == nullptr) return false;
7034
7035 // Add memory barrier to prevent commoning reads from this field
7036 // across safepoint since GC can change its value.
7037 insert_mem_bar(Op_MemBarCPUOrder);
7038
7039 set_result(result);
7040 return true;
7041 }
7042
7043 //----------------------------inline_reference_refersTo0----------------------------
7044 // bool java.lang.ref.Reference.refersTo0();
7045 // bool java.lang.ref.PhantomReference.refersTo0();
7046 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7047 // Get arguments:
7048 Node* reference_obj = null_check_receiver();
7049 Node* other_obj = argument(1);
7050 if (stopped()) return true;
7051
7052 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7053 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7054 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7055 decorators, /*is_static*/ false, nullptr);
7056 if (referent == nullptr) return false;
7057
7058 // Add memory barrier to prevent commoning reads from this field
7059 // across safepoint since GC can change its value.
7060 insert_mem_bar(Op_MemBarCPUOrder);
7061
7062 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7063 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7064 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7065
7066 RegionNode* region = new RegionNode(3);
7067 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7068
7069 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7070 region->init_req(1, if_true);
7071 phi->init_req(1, intcon(1));
7072
7073 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7074 region->init_req(2, if_false);
7075 phi->init_req(2, intcon(0));
7076
7077 set_control(_gvn.transform(region));
7078 record_for_igvn(region);
7079 set_result(_gvn.transform(phi));
7080 return true;
7081 }
7082
7083 //----------------------------inline_reference_clear0----------------------------
7084 // void java.lang.ref.Reference.clear0();
7085 // void java.lang.ref.PhantomReference.clear0();
7086 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7087 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7088
7089 // Get arguments
7090 Node* reference_obj = null_check_receiver();
7091 if (stopped()) return true;
7092
7093 // Common access parameters
7094 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7095 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7096 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7097 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7098 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7099
7100 Node* referent = access_load_at(reference_obj,
7101 referent_field_addr,
7102 referent_field_addr_type,
7103 val_type,
7104 T_OBJECT,
7105 decorators);
7106
7107 IdealKit ideal(this);
7108 #define __ ideal.
7109 __ if_then(referent, BoolTest::ne, null());
7110 sync_kit(ideal);
7111 access_store_at(reference_obj,
7112 referent_field_addr,
7113 referent_field_addr_type,
7114 null(),
7115 val_type,
7116 T_OBJECT,
7117 decorators);
7118 __ sync_kit(this);
7119 __ end_if();
7120 final_sync(ideal);
7121 #undef __
7122
7123 return true;
7124 }
7125
7126 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7127 DecoratorSet decorators, bool is_static,
7128 ciInstanceKlass* fromKls) {
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 fromKls = tinst->instance_klass();
7134 } else {
7135 assert(is_static, "only for static field access");
7136 }
7137 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7138 ciSymbol::make(fieldTypeString),
7139 is_static);
7140
7141 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7142 if (field == nullptr) return (Node *) nullptr;
7143
7144 if (is_static) {
7145 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7146 fromObj = makecon(tip);
7147 }
7148
7149 // Next code copied from Parse::do_get_xxx():
7150
7151 // Compute address and memory type.
7152 int offset = field->offset_in_bytes();
7153 bool is_vol = field->is_volatile();
7154 ciType* field_klass = field->type();
7155 assert(field_klass->is_loaded(), "should be loaded");
7156 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7157 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7158 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7159 "slice of address and input slice don't match");
7160 BasicType bt = field->layout_type();
7161
7162 // Build the resultant type of the load
7163 const Type *type;
7164 if (bt == T_OBJECT) {
7165 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7166 } else {
7167 type = Type::get_const_basic_type(bt);
7168 }
7169
7170 if (is_vol) {
7171 decorators |= MO_SEQ_CST;
7172 }
7173
7174 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7175 }
7176
7177 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7178 bool is_exact /* true */, bool is_static /* false */,
7179 ciInstanceKlass * fromKls /* nullptr */) {
7180 if (fromKls == nullptr) {
7181 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7182 assert(tinst != nullptr, "obj is null");
7183 assert(tinst->is_loaded(), "obj is not loaded");
7184 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7185 fromKls = tinst->instance_klass();
7186 }
7187 else {
7188 assert(is_static, "only for static field access");
7189 }
7190 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7191 ciSymbol::make(fieldTypeString),
7192 is_static);
7193
7194 assert(field != nullptr, "undefined field");
7195 assert(!field->is_volatile(), "not defined for volatile fields");
7196
7197 if (is_static) {
7198 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7199 fromObj = makecon(tip);
7200 }
7201
7202 // Next code copied from Parse::do_get_xxx():
7203
7204 // Compute address and memory type.
7205 int offset = field->offset_in_bytes();
7206 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7207
7208 return adr;
7209 }
7210
7211 //------------------------------inline_aescrypt_Block-----------------------
7212 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7213 address stubAddr = nullptr;
7214 const char *stubName;
7215 bool is_decrypt = false;
7216 assert(UseAES, "need AES instruction support");
7217
7218 switch(id) {
7219 case vmIntrinsics::_aescrypt_encryptBlock:
7220 stubAddr = StubRoutines::aescrypt_encryptBlock();
7221 stubName = "aescrypt_encryptBlock";
7222 break;
7223 case vmIntrinsics::_aescrypt_decryptBlock:
7224 stubAddr = StubRoutines::aescrypt_decryptBlock();
7225 stubName = "aescrypt_decryptBlock";
7226 is_decrypt = true;
7227 break;
7228 default:
7229 break;
7230 }
7231 if (stubAddr == nullptr) return false;
7232
7233 Node* aescrypt_object = argument(0);
7234 Node* src = argument(1);
7235 Node* src_offset = argument(2);
7236 Node* dest = argument(3);
7237 Node* dest_offset = argument(4);
7238
7239 src = must_be_not_null(src, true);
7240 dest = must_be_not_null(dest, true);
7241
7242 // (1) src and dest are arrays.
7243 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7244 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7245 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7246 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7247
7248 // for the quick and dirty code we will skip all the checks.
7249 // we are just trying to get the call to be generated.
7250 Node* src_start = src;
7251 Node* dest_start = dest;
7252 if (src_offset != nullptr || dest_offset != nullptr) {
7253 assert(src_offset != nullptr && dest_offset != nullptr, "");
7254 src_start = array_element_address(src, src_offset, T_BYTE);
7255 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7256 }
7257
7258 // now need to get the start of its expanded key array
7259 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7260 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7261 if (k_start == nullptr) return false;
7262
7263 // Call the stub.
7264 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7265 stubAddr, stubName, TypePtr::BOTTOM,
7266 src_start, dest_start, k_start);
7267
7268 return true;
7269 }
7270
7271 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7272 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7273 address stubAddr = nullptr;
7274 const char *stubName = nullptr;
7275 bool is_decrypt = false;
7276 assert(UseAES, "need AES instruction support");
7277
7278 switch(id) {
7279 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7280 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7281 stubName = "cipherBlockChaining_encryptAESCrypt";
7282 break;
7283 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7284 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7285 stubName = "cipherBlockChaining_decryptAESCrypt";
7286 is_decrypt = true;
7287 break;
7288 default:
7289 break;
7290 }
7291 if (stubAddr == nullptr) return false;
7292
7293 Node* cipherBlockChaining_object = argument(0);
7294 Node* src = argument(1);
7295 Node* src_offset = argument(2);
7296 Node* len = argument(3);
7297 Node* dest = argument(4);
7298 Node* dest_offset = argument(5);
7299
7300 src = must_be_not_null(src, false);
7301 dest = must_be_not_null(dest, false);
7302
7303 // (1) src and dest are arrays.
7304 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7305 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7306 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7307 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7308
7309 // checks are the responsibility of the caller
7310 Node* src_start = src;
7311 Node* dest_start = dest;
7312 if (src_offset != nullptr || dest_offset != nullptr) {
7313 assert(src_offset != nullptr && dest_offset != nullptr, "");
7314 src_start = array_element_address(src, src_offset, T_BYTE);
7315 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7316 }
7317
7318 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7319 // (because of the predicated logic executed earlier).
7320 // so we cast it here safely.
7321 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7322
7323 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7324 if (embeddedCipherObj == nullptr) return false;
7325
7326 // cast it to what we know it will be at runtime
7327 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7328 assert(tinst != nullptr, "CBC obj is null");
7329 assert(tinst->is_loaded(), "CBC obj is not loaded");
7330 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7331 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7332
7333 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7334 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7335 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7336 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7337 aescrypt_object = _gvn.transform(aescrypt_object);
7338
7339 // we need to get the start of the aescrypt_object's expanded key array
7340 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7341 if (k_start == nullptr) return false;
7342
7343 // similarly, get the start address of the r vector
7344 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7345 if (objRvec == nullptr) return false;
7346 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7347
7348 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7349 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7350 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7351 stubAddr, stubName, TypePtr::BOTTOM,
7352 src_start, dest_start, k_start, r_start, len);
7353
7354 // return cipher length (int)
7355 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7356 set_result(retvalue);
7357 return true;
7358 }
7359
7360 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7361 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7362 address stubAddr = nullptr;
7363 const char *stubName = nullptr;
7364 bool is_decrypt = false;
7365 assert(UseAES, "need AES instruction support");
7366
7367 switch (id) {
7368 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7369 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7370 stubName = "electronicCodeBook_encryptAESCrypt";
7371 break;
7372 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7373 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7374 stubName = "electronicCodeBook_decryptAESCrypt";
7375 is_decrypt = true;
7376 break;
7377 default:
7378 break;
7379 }
7380
7381 if (stubAddr == nullptr) return false;
7382
7383 Node* electronicCodeBook_object = argument(0);
7384 Node* src = argument(1);
7385 Node* src_offset = argument(2);
7386 Node* len = argument(3);
7387 Node* dest = argument(4);
7388 Node* dest_offset = argument(5);
7389
7390 // (1) src and dest are arrays.
7391 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7392 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7393 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7394 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7395
7396 // checks are the responsibility of the caller
7397 Node* src_start = src;
7398 Node* dest_start = dest;
7399 if (src_offset != nullptr || dest_offset != nullptr) {
7400 assert(src_offset != nullptr && dest_offset != nullptr, "");
7401 src_start = array_element_address(src, src_offset, T_BYTE);
7402 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7403 }
7404
7405 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7406 // (because of the predicated logic executed earlier).
7407 // so we cast it here safely.
7408 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7409
7410 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7411 if (embeddedCipherObj == nullptr) return false;
7412
7413 // cast it to what we know it will be at runtime
7414 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7415 assert(tinst != nullptr, "ECB obj is null");
7416 assert(tinst->is_loaded(), "ECB obj is not loaded");
7417 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7418 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7419
7420 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7421 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7422 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7423 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7424 aescrypt_object = _gvn.transform(aescrypt_object);
7425
7426 // we need to get the start of the aescrypt_object's expanded key array
7427 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7428 if (k_start == nullptr) return false;
7429
7430 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7431 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7432 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7433 stubAddr, stubName, TypePtr::BOTTOM,
7434 src_start, dest_start, k_start, len);
7435
7436 // return cipher length (int)
7437 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7438 set_result(retvalue);
7439 return true;
7440 }
7441
7442 //------------------------------inline_counterMode_AESCrypt-----------------------
7443 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7444 assert(UseAES, "need AES instruction support");
7445 if (!UseAESCTRIntrinsics) return false;
7446
7447 address stubAddr = nullptr;
7448 const char *stubName = nullptr;
7449 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7450 stubAddr = StubRoutines::counterMode_AESCrypt();
7451 stubName = "counterMode_AESCrypt";
7452 }
7453 if (stubAddr == nullptr) return false;
7454
7455 Node* counterMode_object = argument(0);
7456 Node* src = argument(1);
7457 Node* src_offset = argument(2);
7458 Node* len = argument(3);
7459 Node* dest = argument(4);
7460 Node* dest_offset = argument(5);
7461
7462 // (1) src and dest are arrays.
7463 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7464 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7465 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7466 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7467
7468 // checks are the responsibility of the caller
7469 Node* src_start = src;
7470 Node* dest_start = dest;
7471 if (src_offset != nullptr || dest_offset != nullptr) {
7472 assert(src_offset != nullptr && dest_offset != nullptr, "");
7473 src_start = array_element_address(src, src_offset, T_BYTE);
7474 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7475 }
7476
7477 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7478 // (because of the predicated logic executed earlier).
7479 // so we cast it here safely.
7480 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7481 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7482 if (embeddedCipherObj == nullptr) return false;
7483 // cast it to what we know it will be at runtime
7484 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7485 assert(tinst != nullptr, "CTR obj is null");
7486 assert(tinst->is_loaded(), "CTR obj is not loaded");
7487 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7488 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7489 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7490 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7491 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7492 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7493 aescrypt_object = _gvn.transform(aescrypt_object);
7494 // we need to get the start of the aescrypt_object's expanded key array
7495 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7496 if (k_start == nullptr) return false;
7497 // similarly, get the start address of the r vector
7498 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7499 if (obj_counter == nullptr) return false;
7500 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7501
7502 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7503 if (saved_encCounter == nullptr) return false;
7504 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7505 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7506
7507 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7508 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7509 OptoRuntime::counterMode_aescrypt_Type(),
7510 stubAddr, stubName, TypePtr::BOTTOM,
7511 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7512
7513 // return cipher length (int)
7514 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7515 set_result(retvalue);
7516 return true;
7517 }
7518
7519 //------------------------------get_key_start_from_aescrypt_object-----------------------
7520 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7521 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7522 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7523 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7524 // The following platform specific stubs of encryption and decryption use the same round keys.
7525 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7526 bool use_decryption_key = false;
7527 #else
7528 bool use_decryption_key = is_decrypt;
7529 #endif
7530 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7531 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7532 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7533
7534 // now have the array, need to get the start address of the selected key array
7535 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7536 return k_start;
7537 }
7538
7539 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7540 // Return node representing slow path of predicate check.
7541 // the pseudo code we want to emulate with this predicate is:
7542 // for encryption:
7543 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7544 // for decryption:
7545 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7546 // note cipher==plain is more conservative than the original java code but that's OK
7547 //
7548 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7549 // The receiver was checked for null already.
7550 Node* objCBC = argument(0);
7551
7552 Node* src = argument(1);
7553 Node* dest = argument(4);
7554
7555 // Load embeddedCipher field of CipherBlockChaining object.
7556 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7557
7558 // get AESCrypt klass for instanceOf check
7559 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7560 // will have same classloader as CipherBlockChaining object
7561 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7562 assert(tinst != nullptr, "CBCobj is null");
7563 assert(tinst->is_loaded(), "CBCobj is not loaded");
7564
7565 // we want to do an instanceof comparison against the AESCrypt class
7566 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7567 if (!klass_AESCrypt->is_loaded()) {
7568 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7569 Node* ctrl = control();
7570 set_control(top()); // no regular fast path
7571 return ctrl;
7572 }
7573
7574 src = must_be_not_null(src, true);
7575 dest = must_be_not_null(dest, true);
7576
7577 // Resolve oops to stable for CmpP below.
7578 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7579
7580 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7581 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7582 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7583
7584 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7585
7586 // for encryption, we are done
7587 if (!decrypting)
7588 return instof_false; // even if it is null
7589
7590 // for decryption, we need to add a further check to avoid
7591 // taking the intrinsic path when cipher and plain are the same
7592 // see the original java code for why.
7593 RegionNode* region = new RegionNode(3);
7594 region->init_req(1, instof_false);
7595
7596 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7597 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7598 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7599 region->init_req(2, src_dest_conjoint);
7600
7601 record_for_igvn(region);
7602 return _gvn.transform(region);
7603 }
7604
7605 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7606 // Return node representing slow path of predicate check.
7607 // the pseudo code we want to emulate with this predicate is:
7608 // for encryption:
7609 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7610 // for decryption:
7611 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7612 // note cipher==plain is more conservative than the original java code but that's OK
7613 //
7614 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7615 // The receiver was checked for null already.
7616 Node* objECB = argument(0);
7617
7618 // Load embeddedCipher field of ElectronicCodeBook object.
7619 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7620
7621 // get AESCrypt klass for instanceOf check
7622 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7623 // will have same classloader as ElectronicCodeBook object
7624 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7625 assert(tinst != nullptr, "ECBobj is null");
7626 assert(tinst->is_loaded(), "ECBobj is not loaded");
7627
7628 // we want to do an instanceof comparison against the AESCrypt class
7629 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7630 if (!klass_AESCrypt->is_loaded()) {
7631 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7632 Node* ctrl = control();
7633 set_control(top()); // no regular fast path
7634 return ctrl;
7635 }
7636 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7637
7638 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7639 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7640 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7641
7642 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7643
7644 // for encryption, we are done
7645 if (!decrypting)
7646 return instof_false; // even if it is null
7647
7648 // for decryption, we need to add a further check to avoid
7649 // taking the intrinsic path when cipher and plain are the same
7650 // see the original java code for why.
7651 RegionNode* region = new RegionNode(3);
7652 region->init_req(1, instof_false);
7653 Node* src = argument(1);
7654 Node* dest = argument(4);
7655 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7656 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7657 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7658 region->init_req(2, src_dest_conjoint);
7659
7660 record_for_igvn(region);
7661 return _gvn.transform(region);
7662 }
7663
7664 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7665 // Return node representing slow path of predicate check.
7666 // the pseudo code we want to emulate with this predicate is:
7667 // for encryption:
7668 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7669 // for decryption:
7670 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7671 // note cipher==plain is more conservative than the original java code but that's OK
7672 //
7673
7674 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7675 // The receiver was checked for null already.
7676 Node* objCTR = argument(0);
7677
7678 // Load embeddedCipher field of CipherBlockChaining object.
7679 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7680
7681 // get AESCrypt klass for instanceOf check
7682 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7683 // will have same classloader as CipherBlockChaining object
7684 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7685 assert(tinst != nullptr, "CTRobj is null");
7686 assert(tinst->is_loaded(), "CTRobj is not loaded");
7687
7688 // we want to do an instanceof comparison against the AESCrypt class
7689 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7690 if (!klass_AESCrypt->is_loaded()) {
7691 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7692 Node* ctrl = control();
7693 set_control(top()); // no regular fast path
7694 return ctrl;
7695 }
7696
7697 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7698 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7699 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7700 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7701 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7702
7703 return instof_false; // even if it is null
7704 }
7705
7706 //------------------------------inline_ghash_processBlocks
7707 bool LibraryCallKit::inline_ghash_processBlocks() {
7708 address stubAddr;
7709 const char *stubName;
7710 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7711
7712 stubAddr = StubRoutines::ghash_processBlocks();
7713 stubName = "ghash_processBlocks";
7714
7715 Node* data = argument(0);
7716 Node* offset = argument(1);
7717 Node* len = argument(2);
7718 Node* state = argument(3);
7719 Node* subkeyH = argument(4);
7720
7721 state = must_be_not_null(state, true);
7722 subkeyH = must_be_not_null(subkeyH, true);
7723 data = must_be_not_null(data, true);
7724
7725 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7726 assert(state_start, "state is null");
7727 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7728 assert(subkeyH_start, "subkeyH is null");
7729 Node* data_start = array_element_address(data, offset, T_BYTE);
7730 assert(data_start, "data is null");
7731
7732 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7733 OptoRuntime::ghash_processBlocks_Type(),
7734 stubAddr, stubName, TypePtr::BOTTOM,
7735 state_start, subkeyH_start, data_start, len);
7736 return true;
7737 }
7738
7739 //------------------------------inline_chacha20Block
7740 bool LibraryCallKit::inline_chacha20Block() {
7741 address stubAddr;
7742 const char *stubName;
7743 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7744
7745 stubAddr = StubRoutines::chacha20Block();
7746 stubName = "chacha20Block";
7747
7748 Node* state = argument(0);
7749 Node* result = argument(1);
7750
7751 state = must_be_not_null(state, true);
7752 result = must_be_not_null(result, true);
7753
7754 Node* state_start = array_element_address(state, intcon(0), T_INT);
7755 assert(state_start, "state is null");
7756 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7757 assert(result_start, "result is null");
7758
7759 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7760 OptoRuntime::chacha20Block_Type(),
7761 stubAddr, stubName, TypePtr::BOTTOM,
7762 state_start, result_start);
7763 // return key stream length (int)
7764 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7765 set_result(retvalue);
7766 return true;
7767 }
7768
7769 //------------------------------inline_kyberNtt
7770 bool LibraryCallKit::inline_kyberNtt() {
7771 address stubAddr;
7772 const char *stubName;
7773 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7774 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7775
7776 stubAddr = StubRoutines::kyberNtt();
7777 stubName = "kyberNtt";
7778 if (!stubAddr) return false;
7779
7780 Node* coeffs = argument(0);
7781 Node* ntt_zetas = argument(1);
7782
7783 coeffs = must_be_not_null(coeffs, true);
7784 ntt_zetas = must_be_not_null(ntt_zetas, true);
7785
7786 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7787 assert(coeffs_start, "coeffs is null");
7788 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7789 assert(ntt_zetas_start, "ntt_zetas is null");
7790 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7791 OptoRuntime::kyberNtt_Type(),
7792 stubAddr, stubName, TypePtr::BOTTOM,
7793 coeffs_start, ntt_zetas_start);
7794 // return an int
7795 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7796 set_result(retvalue);
7797 return true;
7798 }
7799
7800 //------------------------------inline_kyberInverseNtt
7801 bool LibraryCallKit::inline_kyberInverseNtt() {
7802 address stubAddr;
7803 const char *stubName;
7804 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7805 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7806
7807 stubAddr = StubRoutines::kyberInverseNtt();
7808 stubName = "kyberInverseNtt";
7809 if (!stubAddr) return false;
7810
7811 Node* coeffs = argument(0);
7812 Node* zetas = argument(1);
7813
7814 coeffs = must_be_not_null(coeffs, true);
7815 zetas = must_be_not_null(zetas, true);
7816
7817 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7818 assert(coeffs_start, "coeffs is null");
7819 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7820 assert(zetas_start, "inverseNtt_zetas is null");
7821 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7822 OptoRuntime::kyberInverseNtt_Type(),
7823 stubAddr, stubName, TypePtr::BOTTOM,
7824 coeffs_start, zetas_start);
7825
7826 // return an int
7827 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7828 set_result(retvalue);
7829 return true;
7830 }
7831
7832 //------------------------------inline_kyberNttMult
7833 bool LibraryCallKit::inline_kyberNttMult() {
7834 address stubAddr;
7835 const char *stubName;
7836 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7837 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7838
7839 stubAddr = StubRoutines::kyberNttMult();
7840 stubName = "kyberNttMult";
7841 if (!stubAddr) return false;
7842
7843 Node* result = argument(0);
7844 Node* ntta = argument(1);
7845 Node* nttb = argument(2);
7846 Node* zetas = argument(3);
7847
7848 result = must_be_not_null(result, true);
7849 ntta = must_be_not_null(ntta, true);
7850 nttb = must_be_not_null(nttb, true);
7851 zetas = must_be_not_null(zetas, true);
7852
7853 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7854 assert(result_start, "result is null");
7855 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7856 assert(ntta_start, "ntta is null");
7857 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7858 assert(nttb_start, "nttb is null");
7859 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7860 assert(zetas_start, "nttMult_zetas is null");
7861 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7862 OptoRuntime::kyberNttMult_Type(),
7863 stubAddr, stubName, TypePtr::BOTTOM,
7864 result_start, ntta_start, nttb_start,
7865 zetas_start);
7866
7867 // return an int
7868 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7869 set_result(retvalue);
7870
7871 return true;
7872 }
7873
7874 //------------------------------inline_kyberAddPoly_2
7875 bool LibraryCallKit::inline_kyberAddPoly_2() {
7876 address stubAddr;
7877 const char *stubName;
7878 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7879 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7880
7881 stubAddr = StubRoutines::kyberAddPoly_2();
7882 stubName = "kyberAddPoly_2";
7883 if (!stubAddr) return false;
7884
7885 Node* result = argument(0);
7886 Node* a = argument(1);
7887 Node* b = argument(2);
7888
7889 result = must_be_not_null(result, true);
7890 a = must_be_not_null(a, true);
7891 b = must_be_not_null(b, true);
7892
7893 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7894 assert(result_start, "result is null");
7895 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7896 assert(a_start, "a is null");
7897 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7898 assert(b_start, "b is null");
7899 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7900 OptoRuntime::kyberAddPoly_2_Type(),
7901 stubAddr, stubName, TypePtr::BOTTOM,
7902 result_start, a_start, b_start);
7903 // return an int
7904 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7905 set_result(retvalue);
7906 return true;
7907 }
7908
7909 //------------------------------inline_kyberAddPoly_3
7910 bool LibraryCallKit::inline_kyberAddPoly_3() {
7911 address stubAddr;
7912 const char *stubName;
7913 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7914 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7915
7916 stubAddr = StubRoutines::kyberAddPoly_3();
7917 stubName = "kyberAddPoly_3";
7918 if (!stubAddr) return false;
7919
7920 Node* result = argument(0);
7921 Node* a = argument(1);
7922 Node* b = argument(2);
7923 Node* c = argument(3);
7924
7925 result = must_be_not_null(result, true);
7926 a = must_be_not_null(a, true);
7927 b = must_be_not_null(b, true);
7928 c = must_be_not_null(c, true);
7929
7930 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7931 assert(result_start, "result is null");
7932 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7933 assert(a_start, "a is null");
7934 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7935 assert(b_start, "b is null");
7936 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7937 assert(c_start, "c is null");
7938 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7939 OptoRuntime::kyberAddPoly_3_Type(),
7940 stubAddr, stubName, TypePtr::BOTTOM,
7941 result_start, a_start, b_start, c_start);
7942 // return an int
7943 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7944 set_result(retvalue);
7945 return true;
7946 }
7947
7948 //------------------------------inline_kyber12To16
7949 bool LibraryCallKit::inline_kyber12To16() {
7950 address stubAddr;
7951 const char *stubName;
7952 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7953 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7954
7955 stubAddr = StubRoutines::kyber12To16();
7956 stubName = "kyber12To16";
7957 if (!stubAddr) return false;
7958
7959 Node* condensed = argument(0);
7960 Node* condensedOffs = argument(1);
7961 Node* parsed = argument(2);
7962 Node* parsedLength = argument(3);
7963
7964 condensed = must_be_not_null(condensed, true);
7965 parsed = must_be_not_null(parsed, true);
7966
7967 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7968 assert(condensed_start, "condensed is null");
7969 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7970 assert(parsed_start, "parsed is null");
7971 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7972 OptoRuntime::kyber12To16_Type(),
7973 stubAddr, stubName, TypePtr::BOTTOM,
7974 condensed_start, condensedOffs, parsed_start, parsedLength);
7975 // return an int
7976 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7977 set_result(retvalue);
7978 return true;
7979
7980 }
7981
7982 //------------------------------inline_kyberBarrettReduce
7983 bool LibraryCallKit::inline_kyberBarrettReduce() {
7984 address stubAddr;
7985 const char *stubName;
7986 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7987 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7988
7989 stubAddr = StubRoutines::kyberBarrettReduce();
7990 stubName = "kyberBarrettReduce";
7991 if (!stubAddr) return false;
7992
7993 Node* coeffs = argument(0);
7994
7995 coeffs = must_be_not_null(coeffs, true);
7996
7997 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7998 assert(coeffs_start, "coeffs is null");
7999 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8000 OptoRuntime::kyberBarrettReduce_Type(),
8001 stubAddr, stubName, TypePtr::BOTTOM,
8002 coeffs_start);
8003 // return an int
8004 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8005 set_result(retvalue);
8006 return true;
8007 }
8008
8009 //------------------------------inline_dilithiumAlmostNtt
8010 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8011 address stubAddr;
8012 const char *stubName;
8013 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8014 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8015
8016 stubAddr = StubRoutines::dilithiumAlmostNtt();
8017 stubName = "dilithiumAlmostNtt";
8018 if (!stubAddr) return false;
8019
8020 Node* coeffs = argument(0);
8021 Node* ntt_zetas = argument(1);
8022
8023 coeffs = must_be_not_null(coeffs, true);
8024 ntt_zetas = must_be_not_null(ntt_zetas, true);
8025
8026 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8027 assert(coeffs_start, "coeffs is null");
8028 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8029 assert(ntt_zetas_start, "ntt_zetas is null");
8030 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8031 OptoRuntime::dilithiumAlmostNtt_Type(),
8032 stubAddr, stubName, TypePtr::BOTTOM,
8033 coeffs_start, ntt_zetas_start);
8034 // return an int
8035 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8036 set_result(retvalue);
8037 return true;
8038 }
8039
8040 //------------------------------inline_dilithiumAlmostInverseNtt
8041 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8042 address stubAddr;
8043 const char *stubName;
8044 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8045 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8046
8047 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8048 stubName = "dilithiumAlmostInverseNtt";
8049 if (!stubAddr) return false;
8050
8051 Node* coeffs = argument(0);
8052 Node* zetas = argument(1);
8053
8054 coeffs = must_be_not_null(coeffs, true);
8055 zetas = must_be_not_null(zetas, true);
8056
8057 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8058 assert(coeffs_start, "coeffs is null");
8059 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8060 assert(zetas_start, "inverseNtt_zetas is null");
8061 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8062 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8063 stubAddr, stubName, TypePtr::BOTTOM,
8064 coeffs_start, zetas_start);
8065 // return an int
8066 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8067 set_result(retvalue);
8068 return true;
8069 }
8070
8071 //------------------------------inline_dilithiumNttMult
8072 bool LibraryCallKit::inline_dilithiumNttMult() {
8073 address stubAddr;
8074 const char *stubName;
8075 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8076 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8077
8078 stubAddr = StubRoutines::dilithiumNttMult();
8079 stubName = "dilithiumNttMult";
8080 if (!stubAddr) return false;
8081
8082 Node* result = argument(0);
8083 Node* ntta = argument(1);
8084 Node* nttb = argument(2);
8085 Node* zetas = argument(3);
8086
8087 result = must_be_not_null(result, true);
8088 ntta = must_be_not_null(ntta, true);
8089 nttb = must_be_not_null(nttb, true);
8090 zetas = must_be_not_null(zetas, true);
8091
8092 Node* result_start = array_element_address(result, intcon(0), T_INT);
8093 assert(result_start, "result is null");
8094 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8095 assert(ntta_start, "ntta is null");
8096 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8097 assert(nttb_start, "nttb is null");
8098 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8099 OptoRuntime::dilithiumNttMult_Type(),
8100 stubAddr, stubName, TypePtr::BOTTOM,
8101 result_start, ntta_start, nttb_start);
8102
8103 // return an int
8104 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8105 set_result(retvalue);
8106
8107 return true;
8108 }
8109
8110 //------------------------------inline_dilithiumMontMulByConstant
8111 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8112 address stubAddr;
8113 const char *stubName;
8114 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8115 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8116
8117 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8118 stubName = "dilithiumMontMulByConstant";
8119 if (!stubAddr) return false;
8120
8121 Node* coeffs = argument(0);
8122 Node* constant = argument(1);
8123
8124 coeffs = must_be_not_null(coeffs, true);
8125
8126 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8127 assert(coeffs_start, "coeffs is null");
8128 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8129 OptoRuntime::dilithiumMontMulByConstant_Type(),
8130 stubAddr, stubName, TypePtr::BOTTOM,
8131 coeffs_start, constant);
8132
8133 // return an int
8134 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8135 set_result(retvalue);
8136 return true;
8137 }
8138
8139
8140 //------------------------------inline_dilithiumDecomposePoly
8141 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8142 address stubAddr;
8143 const char *stubName;
8144 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8145 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8146
8147 stubAddr = StubRoutines::dilithiumDecomposePoly();
8148 stubName = "dilithiumDecomposePoly";
8149 if (!stubAddr) return false;
8150
8151 Node* input = argument(0);
8152 Node* lowPart = argument(1);
8153 Node* highPart = argument(2);
8154 Node* twoGamma2 = argument(3);
8155 Node* multiplier = argument(4);
8156
8157 input = must_be_not_null(input, true);
8158 lowPart = must_be_not_null(lowPart, true);
8159 highPart = must_be_not_null(highPart, true);
8160
8161 Node* input_start = array_element_address(input, intcon(0), T_INT);
8162 assert(input_start, "input is null");
8163 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8164 assert(lowPart_start, "lowPart is null");
8165 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8166 assert(highPart_start, "highPart is null");
8167
8168 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8169 OptoRuntime::dilithiumDecomposePoly_Type(),
8170 stubAddr, stubName, TypePtr::BOTTOM,
8171 input_start, lowPart_start, highPart_start,
8172 twoGamma2, multiplier);
8173
8174 // return an int
8175 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8176 set_result(retvalue);
8177 return true;
8178 }
8179
8180 bool LibraryCallKit::inline_base64_encodeBlock() {
8181 address stubAddr;
8182 const char *stubName;
8183 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8184 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8185 stubAddr = StubRoutines::base64_encodeBlock();
8186 stubName = "encodeBlock";
8187
8188 if (!stubAddr) return false;
8189 Node* base64obj = argument(0);
8190 Node* src = argument(1);
8191 Node* offset = argument(2);
8192 Node* len = argument(3);
8193 Node* dest = argument(4);
8194 Node* dp = argument(5);
8195 Node* isURL = argument(6);
8196
8197 src = must_be_not_null(src, true);
8198 dest = must_be_not_null(dest, true);
8199
8200 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8201 assert(src_start, "source array is null");
8202 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8203 assert(dest_start, "destination array is null");
8204
8205 Node* base64 = make_runtime_call(RC_LEAF,
8206 OptoRuntime::base64_encodeBlock_Type(),
8207 stubAddr, stubName, TypePtr::BOTTOM,
8208 src_start, offset, len, dest_start, dp, isURL);
8209 return true;
8210 }
8211
8212 bool LibraryCallKit::inline_base64_decodeBlock() {
8213 address stubAddr;
8214 const char *stubName;
8215 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8216 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8217 stubAddr = StubRoutines::base64_decodeBlock();
8218 stubName = "decodeBlock";
8219
8220 if (!stubAddr) return false;
8221 Node* base64obj = argument(0);
8222 Node* src = argument(1);
8223 Node* src_offset = argument(2);
8224 Node* len = argument(3);
8225 Node* dest = argument(4);
8226 Node* dest_offset = argument(5);
8227 Node* isURL = argument(6);
8228 Node* isMIME = argument(7);
8229
8230 src = must_be_not_null(src, true);
8231 dest = must_be_not_null(dest, true);
8232
8233 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8234 assert(src_start, "source array is null");
8235 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8236 assert(dest_start, "destination array is null");
8237
8238 Node* call = make_runtime_call(RC_LEAF,
8239 OptoRuntime::base64_decodeBlock_Type(),
8240 stubAddr, stubName, TypePtr::BOTTOM,
8241 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8242 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8243 set_result(result);
8244 return true;
8245 }
8246
8247 bool LibraryCallKit::inline_poly1305_processBlocks() {
8248 address stubAddr;
8249 const char *stubName;
8250 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8251 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8252 stubAddr = StubRoutines::poly1305_processBlocks();
8253 stubName = "poly1305_processBlocks";
8254
8255 if (!stubAddr) return false;
8256 null_check_receiver(); // null-check receiver
8257 if (stopped()) return true;
8258
8259 Node* input = argument(1);
8260 Node* input_offset = argument(2);
8261 Node* len = argument(3);
8262 Node* alimbs = argument(4);
8263 Node* rlimbs = argument(5);
8264
8265 input = must_be_not_null(input, true);
8266 alimbs = must_be_not_null(alimbs, true);
8267 rlimbs = must_be_not_null(rlimbs, true);
8268
8269 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8270 assert(input_start, "input array is null");
8271 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8272 assert(acc_start, "acc array is null");
8273 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8274 assert(r_start, "r array is null");
8275
8276 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8277 OptoRuntime::poly1305_processBlocks_Type(),
8278 stubAddr, stubName, TypePtr::BOTTOM,
8279 input_start, len, acc_start, r_start);
8280 return true;
8281 }
8282
8283 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8284 address stubAddr;
8285 const char *stubName;
8286 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8287 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8288 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8289 stubName = "intpoly_montgomeryMult_P256";
8290
8291 if (!stubAddr) return false;
8292 null_check_receiver(); // null-check receiver
8293 if (stopped()) return true;
8294
8295 Node* a = argument(1);
8296 Node* b = argument(2);
8297 Node* r = argument(3);
8298
8299 a = must_be_not_null(a, true);
8300 b = must_be_not_null(b, true);
8301 r = must_be_not_null(r, true);
8302
8303 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8304 assert(a_start, "a array is null");
8305 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8306 assert(b_start, "b array is null");
8307 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8308 assert(r_start, "r array is null");
8309
8310 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8311 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8312 stubAddr, stubName, TypePtr::BOTTOM,
8313 a_start, b_start, r_start);
8314 return true;
8315 }
8316
8317 bool LibraryCallKit::inline_intpoly_assign() {
8318 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8319 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8320 const char *stubName = "intpoly_assign";
8321 address stubAddr = StubRoutines::intpoly_assign();
8322 if (!stubAddr) return false;
8323
8324 Node* set = argument(0);
8325 Node* a = argument(1);
8326 Node* b = argument(2);
8327 Node* arr_length = load_array_length(a);
8328
8329 a = must_be_not_null(a, true);
8330 b = must_be_not_null(b, true);
8331
8332 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8333 assert(a_start, "a array is null");
8334 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8335 assert(b_start, "b array is null");
8336
8337 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8338 OptoRuntime::intpoly_assign_Type(),
8339 stubAddr, stubName, TypePtr::BOTTOM,
8340 set, a_start, b_start, arr_length);
8341 return true;
8342 }
8343
8344 //------------------------------inline_digestBase_implCompress-----------------------
8345 //
8346 // Calculate MD5 for single-block byte[] array.
8347 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8348 //
8349 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8350 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8351 //
8352 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8353 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8354 //
8355 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8356 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8357 //
8358 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8359 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8360 //
8361 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8362 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8363
8364 Node* digestBase_obj = argument(0);
8365 Node* src = argument(1); // type oop
8366 Node* ofs = argument(2); // type int
8367
8368 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8369 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8370 // failed array check
8371 return false;
8372 }
8373 // Figure out the size and type of the elements we will be copying.
8374 BasicType src_elem = src_type->elem()->array_element_basic_type();
8375 if (src_elem != T_BYTE) {
8376 return false;
8377 }
8378 // 'src_start' points to src array + offset
8379 src = must_be_not_null(src, true);
8380 Node* src_start = array_element_address(src, ofs, src_elem);
8381 Node* state = nullptr;
8382 Node* block_size = nullptr;
8383 address stubAddr;
8384 const char *stubName;
8385
8386 switch(id) {
8387 case vmIntrinsics::_md5_implCompress:
8388 assert(UseMD5Intrinsics, "need MD5 instruction support");
8389 state = get_state_from_digest_object(digestBase_obj, T_INT);
8390 stubAddr = StubRoutines::md5_implCompress();
8391 stubName = "md5_implCompress";
8392 break;
8393 case vmIntrinsics::_sha_implCompress:
8394 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8395 state = get_state_from_digest_object(digestBase_obj, T_INT);
8396 stubAddr = StubRoutines::sha1_implCompress();
8397 stubName = "sha1_implCompress";
8398 break;
8399 case vmIntrinsics::_sha2_implCompress:
8400 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8401 state = get_state_from_digest_object(digestBase_obj, T_INT);
8402 stubAddr = StubRoutines::sha256_implCompress();
8403 stubName = "sha256_implCompress";
8404 break;
8405 case vmIntrinsics::_sha5_implCompress:
8406 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8407 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8408 stubAddr = StubRoutines::sha512_implCompress();
8409 stubName = "sha512_implCompress";
8410 break;
8411 case vmIntrinsics::_sha3_implCompress:
8412 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8413 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8414 stubAddr = StubRoutines::sha3_implCompress();
8415 stubName = "sha3_implCompress";
8416 block_size = get_block_size_from_digest_object(digestBase_obj);
8417 if (block_size == nullptr) return false;
8418 break;
8419 default:
8420 fatal_unexpected_iid(id);
8421 return false;
8422 }
8423 if (state == nullptr) return false;
8424
8425 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8426 if (stubAddr == nullptr) return false;
8427
8428 // Call the stub.
8429 Node* call;
8430 if (block_size == nullptr) {
8431 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8432 stubAddr, stubName, TypePtr::BOTTOM,
8433 src_start, state);
8434 } else {
8435 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8436 stubAddr, stubName, TypePtr::BOTTOM,
8437 src_start, state, block_size);
8438 }
8439
8440 return true;
8441 }
8442
8443 //------------------------------inline_double_keccak
8444 bool LibraryCallKit::inline_double_keccak() {
8445 address stubAddr;
8446 const char *stubName;
8447 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8448 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8449
8450 stubAddr = StubRoutines::double_keccak();
8451 stubName = "double_keccak";
8452 if (!stubAddr) return false;
8453
8454 Node* status0 = argument(0);
8455 Node* status1 = argument(1);
8456
8457 status0 = must_be_not_null(status0, true);
8458 status1 = must_be_not_null(status1, true);
8459
8460 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8461 assert(status0_start, "status0 is null");
8462 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8463 assert(status1_start, "status1 is null");
8464 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8465 OptoRuntime::double_keccak_Type(),
8466 stubAddr, stubName, TypePtr::BOTTOM,
8467 status0_start, status1_start);
8468 // return an int
8469 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8470 set_result(retvalue);
8471 return true;
8472 }
8473
8474
8475 //------------------------------inline_digestBase_implCompressMB-----------------------
8476 //
8477 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8478 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8479 //
8480 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8481 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8482 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8483 assert((uint)predicate < 5, "sanity");
8484 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8485
8486 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8487 Node* src = argument(1); // byte[] array
8488 Node* ofs = argument(2); // type int
8489 Node* limit = argument(3); // type int
8490
8491 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8492 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8493 // failed array check
8494 return false;
8495 }
8496 // Figure out the size and type of the elements we will be copying.
8497 BasicType src_elem = src_type->elem()->array_element_basic_type();
8498 if (src_elem != T_BYTE) {
8499 return false;
8500 }
8501 // 'src_start' points to src array + offset
8502 src = must_be_not_null(src, false);
8503 Node* src_start = array_element_address(src, ofs, src_elem);
8504
8505 const char* klass_digestBase_name = nullptr;
8506 const char* stub_name = nullptr;
8507 address stub_addr = nullptr;
8508 BasicType elem_type = T_INT;
8509
8510 switch (predicate) {
8511 case 0:
8512 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8513 klass_digestBase_name = "sun/security/provider/MD5";
8514 stub_name = "md5_implCompressMB";
8515 stub_addr = StubRoutines::md5_implCompressMB();
8516 }
8517 break;
8518 case 1:
8519 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8520 klass_digestBase_name = "sun/security/provider/SHA";
8521 stub_name = "sha1_implCompressMB";
8522 stub_addr = StubRoutines::sha1_implCompressMB();
8523 }
8524 break;
8525 case 2:
8526 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8527 klass_digestBase_name = "sun/security/provider/SHA2";
8528 stub_name = "sha256_implCompressMB";
8529 stub_addr = StubRoutines::sha256_implCompressMB();
8530 }
8531 break;
8532 case 3:
8533 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8534 klass_digestBase_name = "sun/security/provider/SHA5";
8535 stub_name = "sha512_implCompressMB";
8536 stub_addr = StubRoutines::sha512_implCompressMB();
8537 elem_type = T_LONG;
8538 }
8539 break;
8540 case 4:
8541 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8542 klass_digestBase_name = "sun/security/provider/SHA3";
8543 stub_name = "sha3_implCompressMB";
8544 stub_addr = StubRoutines::sha3_implCompressMB();
8545 elem_type = T_LONG;
8546 }
8547 break;
8548 default:
8549 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8550 }
8551 if (klass_digestBase_name != nullptr) {
8552 assert(stub_addr != nullptr, "Stub is generated");
8553 if (stub_addr == nullptr) return false;
8554
8555 // get DigestBase klass to lookup for SHA klass
8556 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8557 assert(tinst != nullptr, "digestBase_obj is not instance???");
8558 assert(tinst->is_loaded(), "DigestBase is not loaded");
8559
8560 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8561 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8562 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8563 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8564 }
8565 return false;
8566 }
8567
8568 //------------------------------inline_digestBase_implCompressMB-----------------------
8569 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8570 BasicType elem_type, address stubAddr, const char *stubName,
8571 Node* src_start, Node* ofs, Node* limit) {
8572 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8573 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8574 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8575 digest_obj = _gvn.transform(digest_obj);
8576
8577 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8578 if (state == nullptr) return false;
8579
8580 Node* block_size = nullptr;
8581 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8582 block_size = get_block_size_from_digest_object(digest_obj);
8583 if (block_size == nullptr) return false;
8584 }
8585
8586 // Call the stub.
8587 Node* call;
8588 if (block_size == nullptr) {
8589 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8590 OptoRuntime::digestBase_implCompressMB_Type(false),
8591 stubAddr, stubName, TypePtr::BOTTOM,
8592 src_start, state, ofs, limit);
8593 } else {
8594 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8595 OptoRuntime::digestBase_implCompressMB_Type(true),
8596 stubAddr, stubName, TypePtr::BOTTOM,
8597 src_start, state, block_size, ofs, limit);
8598 }
8599
8600 // return ofs (int)
8601 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8602 set_result(result);
8603
8604 return true;
8605 }
8606
8607 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8608 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8609 assert(UseAES, "need AES instruction support");
8610 address stubAddr = nullptr;
8611 const char *stubName = nullptr;
8612 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8613 stubName = "galoisCounterMode_AESCrypt";
8614
8615 if (stubAddr == nullptr) return false;
8616
8617 Node* in = argument(0);
8618 Node* inOfs = argument(1);
8619 Node* len = argument(2);
8620 Node* ct = argument(3);
8621 Node* ctOfs = argument(4);
8622 Node* out = argument(5);
8623 Node* outOfs = argument(6);
8624 Node* gctr_object = argument(7);
8625 Node* ghash_object = argument(8);
8626
8627 // (1) in, ct and out are arrays.
8628 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8629 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8630 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8631 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8632 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8633 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8634
8635 // checks are the responsibility of the caller
8636 Node* in_start = in;
8637 Node* ct_start = ct;
8638 Node* out_start = out;
8639 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8640 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8641 in_start = array_element_address(in, inOfs, T_BYTE);
8642 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8643 out_start = array_element_address(out, outOfs, T_BYTE);
8644 }
8645
8646 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8647 // (because of the predicated logic executed earlier).
8648 // so we cast it here safely.
8649 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8650 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8651 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8652 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8653 Node* state = load_field_from_object(ghash_object, "state", "[J");
8654
8655 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8656 return false;
8657 }
8658 // cast it to what we know it will be at runtime
8659 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8660 assert(tinst != nullptr, "GCTR obj is null");
8661 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8662 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8663 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8664 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8665 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8666 const TypeOopPtr* xtype = aklass->as_instance_type();
8667 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8668 aescrypt_object = _gvn.transform(aescrypt_object);
8669 // we need to get the start of the aescrypt_object's expanded key array
8670 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8671 if (k_start == nullptr) return false;
8672 // similarly, get the start address of the r vector
8673 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8674 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8675 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8676
8677
8678 // Call the stub, passing params
8679 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8680 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8681 stubAddr, stubName, TypePtr::BOTTOM,
8682 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8683
8684 // return cipher length (int)
8685 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8686 set_result(retvalue);
8687
8688 return true;
8689 }
8690
8691 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8692 // Return node representing slow path of predicate check.
8693 // the pseudo code we want to emulate with this predicate is:
8694 // for encryption:
8695 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8696 // for decryption:
8697 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8698 // note cipher==plain is more conservative than the original java code but that's OK
8699 //
8700
8701 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8702 // The receiver was checked for null already.
8703 Node* objGCTR = argument(7);
8704 // Load embeddedCipher field of GCTR object.
8705 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8706 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8707
8708 // get AESCrypt klass for instanceOf check
8709 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8710 // will have same classloader as CipherBlockChaining object
8711 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8712 assert(tinst != nullptr, "GCTR obj is null");
8713 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8714
8715 // we want to do an instanceof comparison against the AESCrypt class
8716 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8717 if (!klass_AESCrypt->is_loaded()) {
8718 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8719 Node* ctrl = control();
8720 set_control(top()); // no regular fast path
8721 return ctrl;
8722 }
8723
8724 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8725 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8726 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8727 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8728 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8729
8730 return instof_false; // even if it is null
8731 }
8732
8733 //------------------------------get_state_from_digest_object-----------------------
8734 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8735 const char* state_type;
8736 switch (elem_type) {
8737 case T_BYTE: state_type = "[B"; break;
8738 case T_INT: state_type = "[I"; break;
8739 case T_LONG: state_type = "[J"; break;
8740 default: ShouldNotReachHere();
8741 }
8742 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8743 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8744 if (digest_state == nullptr) return (Node *) nullptr;
8745
8746 // now have the array, need to get the start address of the state array
8747 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8748 return state;
8749 }
8750
8751 //------------------------------get_block_size_from_sha3_object----------------------------------
8752 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8753 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8754 assert (block_size != nullptr, "sanity");
8755 return block_size;
8756 }
8757
8758 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8759 // Return node representing slow path of predicate check.
8760 // the pseudo code we want to emulate with this predicate is:
8761 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8762 //
8763 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8764 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8765 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8766 assert((uint)predicate < 5, "sanity");
8767
8768 // The receiver was checked for null already.
8769 Node* digestBaseObj = argument(0);
8770
8771 // get DigestBase klass for instanceOf check
8772 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8773 assert(tinst != nullptr, "digestBaseObj is null");
8774 assert(tinst->is_loaded(), "DigestBase is not loaded");
8775
8776 const char* klass_name = nullptr;
8777 switch (predicate) {
8778 case 0:
8779 if (UseMD5Intrinsics) {
8780 // we want to do an instanceof comparison against the MD5 class
8781 klass_name = "sun/security/provider/MD5";
8782 }
8783 break;
8784 case 1:
8785 if (UseSHA1Intrinsics) {
8786 // we want to do an instanceof comparison against the SHA class
8787 klass_name = "sun/security/provider/SHA";
8788 }
8789 break;
8790 case 2:
8791 if (UseSHA256Intrinsics) {
8792 // we want to do an instanceof comparison against the SHA2 class
8793 klass_name = "sun/security/provider/SHA2";
8794 }
8795 break;
8796 case 3:
8797 if (UseSHA512Intrinsics) {
8798 // we want to do an instanceof comparison against the SHA5 class
8799 klass_name = "sun/security/provider/SHA5";
8800 }
8801 break;
8802 case 4:
8803 if (UseSHA3Intrinsics) {
8804 // we want to do an instanceof comparison against the SHA3 class
8805 klass_name = "sun/security/provider/SHA3";
8806 }
8807 break;
8808 default:
8809 fatal("unknown SHA intrinsic predicate: %d", predicate);
8810 }
8811
8812 ciKlass* klass = nullptr;
8813 if (klass_name != nullptr) {
8814 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8815 }
8816 if ((klass == nullptr) || !klass->is_loaded()) {
8817 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8818 Node* ctrl = control();
8819 set_control(top()); // no intrinsic path
8820 return ctrl;
8821 }
8822 ciInstanceKlass* instklass = klass->as_instance_klass();
8823
8824 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8825 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8826 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8827 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8828
8829 return instof_false; // even if it is null
8830 }
8831
8832 //-------------inline_fma-----------------------------------
8833 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8834 Node *a = nullptr;
8835 Node *b = nullptr;
8836 Node *c = nullptr;
8837 Node* result = nullptr;
8838 switch (id) {
8839 case vmIntrinsics::_fmaD:
8840 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8841 // no receiver since it is static method
8842 a = argument(0);
8843 b = argument(2);
8844 c = argument(4);
8845 result = _gvn.transform(new FmaDNode(a, b, c));
8846 break;
8847 case vmIntrinsics::_fmaF:
8848 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8849 a = argument(0);
8850 b = argument(1);
8851 c = argument(2);
8852 result = _gvn.transform(new FmaFNode(a, b, c));
8853 break;
8854 default:
8855 fatal_unexpected_iid(id); break;
8856 }
8857 set_result(result);
8858 return true;
8859 }
8860
8861 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8862 // argument(0) is receiver
8863 Node* codePoint = argument(1);
8864 Node* n = nullptr;
8865
8866 switch (id) {
8867 case vmIntrinsics::_isDigit :
8868 n = new DigitNode(control(), codePoint);
8869 break;
8870 case vmIntrinsics::_isLowerCase :
8871 n = new LowerCaseNode(control(), codePoint);
8872 break;
8873 case vmIntrinsics::_isUpperCase :
8874 n = new UpperCaseNode(control(), codePoint);
8875 break;
8876 case vmIntrinsics::_isWhitespace :
8877 n = new WhitespaceNode(control(), codePoint);
8878 break;
8879 default:
8880 fatal_unexpected_iid(id);
8881 }
8882
8883 set_result(_gvn.transform(n));
8884 return true;
8885 }
8886
8887 bool LibraryCallKit::inline_profileBoolean() {
8888 Node* counts = argument(1);
8889 const TypeAryPtr* ary = nullptr;
8890 ciArray* aobj = nullptr;
8891 if (counts->is_Con()
8892 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8893 && (aobj = ary->const_oop()->as_array()) != nullptr
8894 && (aobj->length() == 2)) {
8895 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8896 jint false_cnt = aobj->element_value(0).as_int();
8897 jint true_cnt = aobj->element_value(1).as_int();
8898
8899 if (C->log() != nullptr) {
8900 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8901 false_cnt, true_cnt);
8902 }
8903
8904 if (false_cnt + true_cnt == 0) {
8905 // According to profile, never executed.
8906 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8907 Deoptimization::Action_reinterpret);
8908 return true;
8909 }
8910
8911 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8912 // is a number of each value occurrences.
8913 Node* result = argument(0);
8914 if (false_cnt == 0 || true_cnt == 0) {
8915 // According to profile, one value has been never seen.
8916 int expected_val = (false_cnt == 0) ? 1 : 0;
8917
8918 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8919 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8920
8921 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8922 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8923 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8924
8925 { // Slow path: uncommon trap for never seen value and then reexecute
8926 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8927 // the value has been seen at least once.
8928 PreserveJVMState pjvms(this);
8929 PreserveReexecuteState preexecs(this);
8930 jvms()->set_should_reexecute(true);
8931
8932 set_control(slow_path);
8933 set_i_o(i_o());
8934
8935 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8936 Deoptimization::Action_reinterpret);
8937 }
8938 // The guard for never seen value enables sharpening of the result and
8939 // returning a constant. It allows to eliminate branches on the same value
8940 // later on.
8941 set_control(fast_path);
8942 result = intcon(expected_val);
8943 }
8944 // Stop profiling.
8945 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8946 // By replacing method body with profile data (represented as ProfileBooleanNode
8947 // on IR level) we effectively disable profiling.
8948 // It enables full speed execution once optimized code is generated.
8949 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8950 C->record_for_igvn(profile);
8951 set_result(profile);
8952 return true;
8953 } else {
8954 // Continue profiling.
8955 // Profile data isn't available at the moment. So, execute method's bytecode version.
8956 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8957 // is compiled and counters aren't available since corresponding MethodHandle
8958 // isn't a compile-time constant.
8959 return false;
8960 }
8961 }
8962
8963 bool LibraryCallKit::inline_isCompileConstant() {
8964 Node* n = argument(0);
8965 set_result(n->is_Con() ? intcon(1) : intcon(0));
8966 return true;
8967 }
8968
8969 //------------------------------- inline_getObjectSize --------------------------------------
8970 //
8971 // Calculate the runtime size of the object/array.
8972 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8973 //
8974 bool LibraryCallKit::inline_getObjectSize() {
8975 Node* obj = argument(3);
8976 Node* klass_node = load_object_klass(obj);
8977
8978 jint layout_con = Klass::_lh_neutral_value;
8979 Node* layout_val = get_layout_helper(klass_node, layout_con);
8980 int layout_is_con = (layout_val == nullptr);
8981
8982 if (layout_is_con) {
8983 // Layout helper is constant, can figure out things at compile time.
8984
8985 if (Klass::layout_helper_is_instance(layout_con)) {
8986 // Instance case: layout_con contains the size itself.
8987 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8988 set_result(size);
8989 } else {
8990 // Array case: size is round(header + element_size*arraylength).
8991 // Since arraylength is different for every array instance, we have to
8992 // compute the whole thing at runtime.
8993
8994 Node* arr_length = load_array_length(obj);
8995
8996 int round_mask = MinObjAlignmentInBytes - 1;
8997 int hsize = Klass::layout_helper_header_size(layout_con);
8998 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8999
9000 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9001 round_mask = 0; // strength-reduce it if it goes away completely
9002 }
9003 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9004 Node* header_size = intcon(hsize + round_mask);
9005
9006 Node* lengthx = ConvI2X(arr_length);
9007 Node* headerx = ConvI2X(header_size);
9008
9009 Node* abody = lengthx;
9010 if (eshift != 0) {
9011 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9012 }
9013 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9014 if (round_mask != 0) {
9015 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9016 }
9017 size = ConvX2L(size);
9018 set_result(size);
9019 }
9020 } else {
9021 // Layout helper is not constant, need to test for array-ness at runtime.
9022
9023 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9024 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9025 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9026 record_for_igvn(result_reg);
9027
9028 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9029 if (array_ctl != nullptr) {
9030 // Array case: size is round(header + element_size*arraylength).
9031 // Since arraylength is different for every array instance, we have to
9032 // compute the whole thing at runtime.
9033
9034 PreserveJVMState pjvms(this);
9035 set_control(array_ctl);
9036 Node* arr_length = load_array_length(obj);
9037
9038 int round_mask = MinObjAlignmentInBytes - 1;
9039 Node* mask = intcon(round_mask);
9040
9041 Node* hss = intcon(Klass::_lh_header_size_shift);
9042 Node* hsm = intcon(Klass::_lh_header_size_mask);
9043 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9044 header_size = _gvn.transform(new AndINode(header_size, hsm));
9045 header_size = _gvn.transform(new AddINode(header_size, mask));
9046
9047 // There is no need to mask or shift this value.
9048 // The semantics of LShiftINode include an implicit mask to 0x1F.
9049 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9050 Node* elem_shift = layout_val;
9051
9052 Node* lengthx = ConvI2X(arr_length);
9053 Node* headerx = ConvI2X(header_size);
9054
9055 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9056 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9057 if (round_mask != 0) {
9058 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9059 }
9060 size = ConvX2L(size);
9061
9062 result_reg->init_req(_array_path, control());
9063 result_val->init_req(_array_path, size);
9064 }
9065
9066 if (!stopped()) {
9067 // Instance case: the layout helper gives us instance size almost directly,
9068 // but we need to mask out the _lh_instance_slow_path_bit.
9069 Node* size = ConvI2X(layout_val);
9070 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9071 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9072 size = _gvn.transform(new AndXNode(size, mask));
9073 size = ConvX2L(size);
9074
9075 result_reg->init_req(_instance_path, control());
9076 result_val->init_req(_instance_path, size);
9077 }
9078
9079 set_result(result_reg, result_val);
9080 }
9081
9082 return true;
9083 }
9084
9085 //------------------------------- inline_blackhole --------------------------------------
9086 //
9087 // Make sure all arguments to this node are alive.
9088 // This matches methods that were requested to be blackholed through compile commands.
9089 //
9090 bool LibraryCallKit::inline_blackhole() {
9091 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9092 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9093 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9094
9095 // Blackhole node pinches only the control, not memory. This allows
9096 // the blackhole to be pinned in the loop that computes blackholed
9097 // values, but have no other side effects, like breaking the optimizations
9098 // across the blackhole.
9099
9100 Node* bh = _gvn.transform(new BlackholeNode(control()));
9101 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9102
9103 // Bind call arguments as blackhole arguments to keep them alive
9104 uint nargs = callee()->arg_size();
9105 for (uint i = 0; i < nargs; i++) {
9106 bh->add_req(argument(i));
9107 }
9108
9109 return true;
9110 }
9111
9112 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9113 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9114 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9115 return nullptr; // box klass is not Float16
9116 }
9117
9118 // Null check; get notnull casted pointer
9119 Node* null_ctl = top();
9120 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9121 // If not_null_box is dead, only null-path is taken
9122 if (stopped()) {
9123 set_control(null_ctl);
9124 return nullptr;
9125 }
9126 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9127 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9128 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9129 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9130 }
9131
9132 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9133 PreserveReexecuteState preexecs(this);
9134 jvms()->set_should_reexecute(true);
9135
9136 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9137 Node* klass_node = makecon(klass_type);
9138 Node* box = new_instance(klass_node);
9139
9140 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9141 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9142
9143 Node* field_store = _gvn.transform(access_store_at(box,
9144 value_field,
9145 value_adr_type,
9146 value,
9147 TypeInt::SHORT,
9148 T_SHORT,
9149 IN_HEAP));
9150 set_memory(field_store, value_adr_type);
9151 return box;
9152 }
9153
9154 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9155 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9156 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9157 return false;
9158 }
9159
9160 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9161 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9162 return false;
9163 }
9164
9165 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9166 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9167 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9168 ciSymbols::short_signature(),
9169 false);
9170 assert(field != nullptr, "");
9171
9172 // Transformed nodes
9173 Node* fld1 = nullptr;
9174 Node* fld2 = nullptr;
9175 Node* fld3 = nullptr;
9176 switch(num_args) {
9177 case 3:
9178 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9179 if (fld3 == nullptr) {
9180 return false;
9181 }
9182 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9183 // fall-through
9184 case 2:
9185 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9186 if (fld2 == nullptr) {
9187 return false;
9188 }
9189 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9190 // fall-through
9191 case 1:
9192 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9193 if (fld1 == nullptr) {
9194 return false;
9195 }
9196 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9197 break;
9198 default: fatal("Unsupported number of arguments %d", num_args);
9199 }
9200
9201 Node* result = nullptr;
9202 switch (id) {
9203 // Unary operations
9204 case vmIntrinsics::_sqrt_float16:
9205 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9206 break;
9207 // Ternary operations
9208 case vmIntrinsics::_fma_float16:
9209 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9210 break;
9211 default:
9212 fatal_unexpected_iid(id);
9213 break;
9214 }
9215 result = _gvn.transform(new ReinterpretHF2SNode(result));
9216 set_result(box_fp16_value(float16_box_type, field, result));
9217 return true;
9218 }
9219