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/ciArrayKlass.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciInstanceKlass.hpp"
29 #include "ci/ciSymbols.hpp"
30 #include "ci/ciUtilities.inline.hpp"
31 #include "classfile/vmIntrinsics.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compileLog.hpp"
34 #include "gc/shared/barrierSet.hpp"
35 #include "gc/shared/c2/barrierSetC2.hpp"
36 #include "jfr/support/jfrIntrinsics.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/accessDecorators.hpp"
39 #include "oops/klass.inline.hpp"
40 #include "oops/layoutKind.hpp"
41 #include "oops/objArrayKlass.hpp"
42 #include "opto/addnode.hpp"
43 #include "opto/arraycopynode.hpp"
44 #include "opto/c2compiler.hpp"
45 #include "opto/castnode.hpp"
46 #include "opto/cfgnode.hpp"
47 #include "opto/convertnode.hpp"
48 #include "opto/countbitsnode.hpp"
49 #include "opto/graphKit.hpp"
50 #include "opto/idealKit.hpp"
51 #include "opto/inlinetypenode.hpp"
52 #include "opto/library_call.hpp"
53 #include "opto/mathexactnode.hpp"
54 #include "opto/mulnode.hpp"
55 #include "opto/narrowptrnode.hpp"
56 #include "opto/opaquenode.hpp"
57 #include "opto/opcodes.hpp"
58 #include "opto/parse.hpp"
59 #include "opto/rootnode.hpp"
60 #include "opto/runtime.hpp"
61 #include "opto/subnode.hpp"
62 #include "opto/type.hpp"
63 #include "opto/vectornode.hpp"
64 #include "prims/jvmtiExport.hpp"
65 #include "prims/jvmtiThreadState.hpp"
66 #include "prims/unsafe.hpp"
67 #include "runtime/globals.hpp"
68 #include "runtime/jniHandles.inline.hpp"
69 #include "runtime/mountUnmountDisabler.hpp"
70 #include "runtime/objectMonitor.hpp"
71 #include "runtime/sharedRuntime.hpp"
72 #include "runtime/stubRoutines.hpp"
73 #include "utilities/globalDefinitions.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/powerOfTwo.hpp"
76
77 //---------------------------make_vm_intrinsic----------------------------
78 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
79 vmIntrinsicID id = m->intrinsic_id();
80 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
81
82 if (!m->is_loaded()) {
83 // Do not attempt to inline unloaded methods.
84 return nullptr;
85 }
86
87 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
88 bool is_available = false;
89
90 {
91 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
92 // the compiler must transition to '_thread_in_vm' state because both
93 // methods access VM-internal data.
94 VM_ENTRY_MARK;
95 methodHandle mh(THREAD, m->get_Method());
96 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
97 if (is_available && is_virtual) {
98 is_available = vmIntrinsics::does_virtual_dispatch(id);
99 }
100 }
101
102 if (is_available) {
103 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
104 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
105 return new LibraryIntrinsic(m, is_virtual,
106 vmIntrinsics::predicates_needed(id),
107 vmIntrinsics::does_virtual_dispatch(id),
108 id);
109 } else {
110 return nullptr;
111 }
112 }
113
114 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
115 LibraryCallKit kit(jvms, this);
116 Compile* C = kit.C;
117 int nodes = C->unique();
118 #ifndef PRODUCT
119 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
120 char buf[1000];
121 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
122 tty->print_cr("Intrinsic %s", str);
123 }
124 #endif
125 ciMethod* callee = kit.callee();
126 const int bci = kit.bci();
127 #ifdef ASSERT
128 Node* ctrl = kit.control();
129 #endif
130 // Try to inline the intrinsic.
131 if (callee->check_intrinsic_candidate() &&
132 kit.try_to_inline(_last_predicate)) {
133 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
134 : "(intrinsic)";
135 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
136 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
137 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
138 if (C->log()) {
139 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
140 vmIntrinsics::name_at(intrinsic_id()),
141 (is_virtual() ? " virtual='1'" : ""),
142 C->unique() - nodes);
143 }
144 // Push the result from the inlined method onto the stack.
145 kit.push_result();
146 return kit.transfer_exceptions_into_jvms();
147 }
148
149 // The intrinsic bailed out
150 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
151 assert(jvms->map() == kit.map(), "Out of sync JVM state");
152 if (jvms->has_method()) {
153 // Not a root compile.
154 const char* msg;
155 if (callee->intrinsic_candidate()) {
156 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
157 } else {
158 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
159 : "failed to inline (intrinsic), method not annotated";
160 }
161 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
162 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
163 } else {
164 // Root compile
165 ResourceMark rm;
166 stringStream msg_stream;
167 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
168 vmIntrinsics::name_at(intrinsic_id()),
169 is_virtual() ? " (virtual)" : "", bci);
170 const char *msg = msg_stream.freeze();
171 log_debug(jit, inlining)("%s", msg);
172 if (C->print_intrinsics() || C->print_inlining()) {
173 tty->print("%s", msg);
174 }
175 }
176 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
177
178 return nullptr;
179 }
180
181 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
182 LibraryCallKit kit(jvms, this);
183 Compile* C = kit.C;
184 int nodes = C->unique();
185 _last_predicate = predicate;
186 #ifndef PRODUCT
187 assert(is_predicated() && predicate < predicates_count(), "sanity");
188 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
189 char buf[1000];
190 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
191 tty->print_cr("Predicate for intrinsic %s", str);
192 }
193 #endif
194 ciMethod* callee = kit.callee();
195 const int bci = kit.bci();
196
197 Node* slow_ctl = kit.try_to_predicate(predicate);
198 if (!kit.failing()) {
199 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
200 : "(intrinsic, predicate)";
201 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
202 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
203
204 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
205 if (C->log()) {
206 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
207 vmIntrinsics::name_at(intrinsic_id()),
208 (is_virtual() ? " virtual='1'" : ""),
209 C->unique() - nodes);
210 }
211 return slow_ctl; // Could be null if the check folds.
212 }
213
214 // The intrinsic bailed out
215 if (jvms->has_method()) {
216 // Not a root compile.
217 const char* msg = "failed to generate predicate for intrinsic";
218 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
219 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
220 } else {
221 // Root compile
222 ResourceMark rm;
223 stringStream msg_stream;
224 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
225 vmIntrinsics::name_at(intrinsic_id()),
226 is_virtual() ? " (virtual)" : "", bci);
227 const char *msg = msg_stream.freeze();
228 log_debug(jit, inlining)("%s", msg);
229 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
230 }
231 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
232 return nullptr;
233 }
234
235 bool LibraryCallKit::try_to_inline(int predicate) {
236 // Handle symbolic names for otherwise undistinguished boolean switches:
237 const bool is_store = true;
238 const bool is_compress = true;
239 const bool is_static = true;
240 const bool is_volatile = true;
241
242 if (!jvms()->has_method()) {
243 // Root JVMState has a null method.
244 assert(map()->memory()->Opcode() == Op_Parm, "");
245 // Insert the memory aliasing node
246 set_all_memory(reset_memory());
247 }
248 assert(merged_memory(), "");
249
250 switch (intrinsic_id()) {
251 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
252 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
253 case vmIntrinsics::_getClass: return inline_native_getClass();
254
255 case vmIntrinsics::_ceil:
256 case vmIntrinsics::_floor:
257 case vmIntrinsics::_rint:
258 case vmIntrinsics::_dsin:
259 case vmIntrinsics::_dcos:
260 case vmIntrinsics::_dtan:
261 case vmIntrinsics::_dsinh:
262 case vmIntrinsics::_dtanh:
263 case vmIntrinsics::_dcbrt:
264 case vmIntrinsics::_dabs:
265 case vmIntrinsics::_fabs:
266 case vmIntrinsics::_iabs:
267 case vmIntrinsics::_labs:
268 case vmIntrinsics::_datan2:
269 case vmIntrinsics::_dsqrt:
270 case vmIntrinsics::_dsqrt_strict:
271 case vmIntrinsics::_dexp:
272 case vmIntrinsics::_dlog:
273 case vmIntrinsics::_dlog10:
274 case vmIntrinsics::_dpow:
275 case vmIntrinsics::_dcopySign:
276 case vmIntrinsics::_fcopySign:
277 case vmIntrinsics::_dsignum:
278 case vmIntrinsics::_roundF:
279 case vmIntrinsics::_roundD:
280 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
281
282 case vmIntrinsics::_notify:
283 case vmIntrinsics::_notifyAll:
284 return inline_notify(intrinsic_id());
285
286 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
287 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
288 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
289 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
290 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
291 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
292 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
293 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
294 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
295 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
296 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
297 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
298 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
299 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
300
301 case vmIntrinsics::_arraycopy: return inline_arraycopy();
302
303 case vmIntrinsics::_arraySort: return inline_array_sort();
304 case vmIntrinsics::_arrayPartition: return inline_array_partition();
305
306 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
307 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
308 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
309 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
310
311 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
312 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
313 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
314 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
315 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
316 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
317 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
318 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
319
320 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
321
322 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
323
324 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
325 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
326 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
327 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
328
329 case vmIntrinsics::_compressStringC:
330 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
331 case vmIntrinsics::_inflateStringC:
332 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
333
334 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
335 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
336 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
337 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
338 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
339 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
340 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
341 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
342 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
343 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
344 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
345
346 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
347 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
348 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
349 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
350 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
351 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
352 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
353 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
354 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
355
356 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
357 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
358 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
359 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
360 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
361 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
362 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
363 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
364 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
365
366 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
367 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
368 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
369 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
370 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
371 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
372 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
373 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
374 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
375
376 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
377 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
378 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
379 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
380
381 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
382 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
383 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
384 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
385
386 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
387 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
388 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
389 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
390 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
391 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
392 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
393 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
394 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
395
396 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
397 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
398 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
399 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
400 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
401 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
402 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
403 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
404 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
405
406 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
407 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
408 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
409 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
410 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
411 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
412 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
413 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
414 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
415
416 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
417 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
418 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
419 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
420 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
421 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
422 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
423 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
424 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
425
426 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
427 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
428
429 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
431 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
432 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
433 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
434
435 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
443 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
444 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
445 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
446 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
447 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
448 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
449 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
450 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
451 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
452 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
453 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
454 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
455
456 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
462 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
463 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
464 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
465 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
466 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
467 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
468 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
469 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
470 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
471
472 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
473 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
474 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
475 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
476
477 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
479 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
480 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
481 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
482
483 case vmIntrinsics::_loadFence:
484 case vmIntrinsics::_storeFence:
485 case vmIntrinsics::_storeStoreFence:
486 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
487
488 case vmIntrinsics::_arrayInstanceBaseOffset: return inline_arrayInstanceBaseOffset();
489 case vmIntrinsics::_arrayInstanceIndexScale: return inline_arrayInstanceIndexScale();
490 case vmIntrinsics::_arrayLayout: return inline_arrayLayout();
491 case vmIntrinsics::_getFieldMap: return inline_getFieldMap();
492
493 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
494
495 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
496 case vmIntrinsics::_currentThread: return inline_native_currentThread();
497 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
498
499 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
500 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
501
502 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
503 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
504
505 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
506 "endFirstTransition", true);
507 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
508 "startFinalTransition", true);
509 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
510 "startTransition", false);
511 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
512 "endTransition", false);
513 #if INCLUDE_JVMTI
514 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
515 #endif
516
517 #ifdef JFR_HAVE_INTRINSICS
518 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
519 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
520 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
521 #endif
522 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
523 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
524 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
525 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
526 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
527 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
528 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
529 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
530 case vmIntrinsics::_getLength: return inline_native_getLength();
531 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
532 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
533 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
534 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
535 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
536 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
537 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
538
539 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
540 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
541 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
542 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
543 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
544 case vmIntrinsics::_isFlatArray: return inline_getArrayProperties(IsFlat);
545 case vmIntrinsics::_isNullRestrictedArray: return inline_getArrayProperties(IsNullRestricted);
546 case vmIntrinsics::_isAtomicArray: return inline_getArrayProperties(IsAtomic);
547
548 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
549
550 case vmIntrinsics::_isInstance:
551 case vmIntrinsics::_isHidden:
552 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
553
554 case vmIntrinsics::_floatToRawIntBits:
555 case vmIntrinsics::_floatToIntBits:
556 case vmIntrinsics::_intBitsToFloat:
557 case vmIntrinsics::_doubleToRawLongBits:
558 case vmIntrinsics::_doubleToLongBits:
559 case vmIntrinsics::_longBitsToDouble:
560 case vmIntrinsics::_floatToFloat16:
561 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
562 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
563 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
564 case vmIntrinsics::_floatIsFinite:
565 case vmIntrinsics::_floatIsInfinite:
566 case vmIntrinsics::_doubleIsFinite:
567 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
568
569 case vmIntrinsics::_numberOfLeadingZeros_i:
570 case vmIntrinsics::_numberOfLeadingZeros_l:
571 case vmIntrinsics::_numberOfTrailingZeros_i:
572 case vmIntrinsics::_numberOfTrailingZeros_l:
573 case vmIntrinsics::_bitCount_i:
574 case vmIntrinsics::_bitCount_l:
575 case vmIntrinsics::_reverse_i:
576 case vmIntrinsics::_reverse_l:
577 case vmIntrinsics::_reverseBytes_i:
578 case vmIntrinsics::_reverseBytes_l:
579 case vmIntrinsics::_reverseBytes_s:
580 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
581
582 case vmIntrinsics::_compress_i:
583 case vmIntrinsics::_compress_l:
584 case vmIntrinsics::_expand_i:
585 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
586
587 case vmIntrinsics::_compareUnsigned_i:
588 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
589
590 case vmIntrinsics::_divideUnsigned_i:
591 case vmIntrinsics::_divideUnsigned_l:
592 case vmIntrinsics::_remainderUnsigned_i:
593 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
594
595 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
596
597 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
598 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
599 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
600 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
601 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
602
603 case vmIntrinsics::_Class_cast: return inline_Class_cast();
604
605 case vmIntrinsics::_aescrypt_encryptBlock:
606 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
607
608 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
609 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
610 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
611
612 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
613 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
614 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
615
616 case vmIntrinsics::_counterMode_AESCrypt:
617 return inline_counterMode_AESCrypt(intrinsic_id());
618
619 case vmIntrinsics::_galoisCounterMode_AESCrypt:
620 return inline_galoisCounterMode_AESCrypt();
621
622 case vmIntrinsics::_md5_implCompress:
623 case vmIntrinsics::_sha_implCompress:
624 case vmIntrinsics::_sha2_implCompress:
625 case vmIntrinsics::_sha5_implCompress:
626 case vmIntrinsics::_sha3_implCompress:
627 return inline_digestBase_implCompress(intrinsic_id());
628 case vmIntrinsics::_double_keccak:
629 return inline_double_keccak();
630
631 case vmIntrinsics::_digestBase_implCompressMB:
632 return inline_digestBase_implCompressMB(predicate);
633
634 case vmIntrinsics::_multiplyToLen:
635 return inline_multiplyToLen();
636
637 case vmIntrinsics::_squareToLen:
638 return inline_squareToLen();
639
640 case vmIntrinsics::_mulAdd:
641 return inline_mulAdd();
642
643 case vmIntrinsics::_montgomeryMultiply:
644 return inline_montgomeryMultiply();
645 case vmIntrinsics::_montgomerySquare:
646 return inline_montgomerySquare();
647
648 case vmIntrinsics::_bigIntegerRightShiftWorker:
649 return inline_bigIntegerShift(true);
650 case vmIntrinsics::_bigIntegerLeftShiftWorker:
651 return inline_bigIntegerShift(false);
652
653 case vmIntrinsics::_vectorizedMismatch:
654 return inline_vectorizedMismatch();
655
656 case vmIntrinsics::_ghash_processBlocks:
657 return inline_ghash_processBlocks();
658 case vmIntrinsics::_chacha20Block:
659 return inline_chacha20Block();
660 case vmIntrinsics::_kyberNtt:
661 return inline_kyberNtt();
662 case vmIntrinsics::_kyberInverseNtt:
663 return inline_kyberInverseNtt();
664 case vmIntrinsics::_kyberNttMult:
665 return inline_kyberNttMult();
666 case vmIntrinsics::_kyberAddPoly_2:
667 return inline_kyberAddPoly_2();
668 case vmIntrinsics::_kyberAddPoly_3:
669 return inline_kyberAddPoly_3();
670 case vmIntrinsics::_kyber12To16:
671 return inline_kyber12To16();
672 case vmIntrinsics::_kyberBarrettReduce:
673 return inline_kyberBarrettReduce();
674 case vmIntrinsics::_dilithiumAlmostNtt:
675 return inline_dilithiumAlmostNtt();
676 case vmIntrinsics::_dilithiumAlmostInverseNtt:
677 return inline_dilithiumAlmostInverseNtt();
678 case vmIntrinsics::_dilithiumNttMult:
679 return inline_dilithiumNttMult();
680 case vmIntrinsics::_dilithiumMontMulByConstant:
681 return inline_dilithiumMontMulByConstant();
682 case vmIntrinsics::_dilithiumDecomposePoly:
683 return inline_dilithiumDecomposePoly();
684 case vmIntrinsics::_base64_encodeBlock:
685 return inline_base64_encodeBlock();
686 case vmIntrinsics::_base64_decodeBlock:
687 return inline_base64_decodeBlock();
688 case vmIntrinsics::_poly1305_processBlocks:
689 return inline_poly1305_processBlocks();
690 case vmIntrinsics::_intpoly_montgomeryMult_P256:
691 return inline_intpoly_montgomeryMult_P256();
692 case vmIntrinsics::_intpoly_assign:
693 return inline_intpoly_assign();
694 case vmIntrinsics::_encodeISOArray:
695 case vmIntrinsics::_encodeByteISOArray:
696 return inline_encodeISOArray(false);
697 case vmIntrinsics::_encodeAsciiArray:
698 return inline_encodeISOArray(true);
699
700 case vmIntrinsics::_updateCRC32:
701 return inline_updateCRC32();
702 case vmIntrinsics::_updateBytesCRC32:
703 return inline_updateBytesCRC32();
704 case vmIntrinsics::_updateByteBufferCRC32:
705 return inline_updateByteBufferCRC32();
706
707 case vmIntrinsics::_updateBytesCRC32C:
708 return inline_updateBytesCRC32C();
709 case vmIntrinsics::_updateDirectByteBufferCRC32C:
710 return inline_updateDirectByteBufferCRC32C();
711
712 case vmIntrinsics::_updateBytesAdler32:
713 return inline_updateBytesAdler32();
714 case vmIntrinsics::_updateByteBufferAdler32:
715 return inline_updateByteBufferAdler32();
716
717 case vmIntrinsics::_profileBoolean:
718 return inline_profileBoolean();
719 case vmIntrinsics::_isCompileConstant:
720 return inline_isCompileConstant();
721
722 case vmIntrinsics::_countPositives:
723 return inline_countPositives();
724
725 case vmIntrinsics::_fmaD:
726 case vmIntrinsics::_fmaF:
727 return inline_fma(intrinsic_id());
728
729 case vmIntrinsics::_isDigit:
730 case vmIntrinsics::_isLowerCase:
731 case vmIntrinsics::_isUpperCase:
732 case vmIntrinsics::_isWhitespace:
733 return inline_character_compare(intrinsic_id());
734
735 case vmIntrinsics::_min:
736 case vmIntrinsics::_max:
737 case vmIntrinsics::_min_strict:
738 case vmIntrinsics::_max_strict:
739 case vmIntrinsics::_minL:
740 case vmIntrinsics::_maxL:
741 case vmIntrinsics::_minF:
742 case vmIntrinsics::_maxF:
743 case vmIntrinsics::_minD:
744 case vmIntrinsics::_maxD:
745 case vmIntrinsics::_minF_strict:
746 case vmIntrinsics::_maxF_strict:
747 case vmIntrinsics::_minD_strict:
748 case vmIntrinsics::_maxD_strict:
749 return inline_min_max(intrinsic_id());
750
751 case vmIntrinsics::_VectorUnaryOp:
752 return inline_vector_nary_operation(1);
753 case vmIntrinsics::_VectorBinaryOp:
754 return inline_vector_nary_operation(2);
755 case vmIntrinsics::_VectorUnaryLibOp:
756 return inline_vector_call(1);
757 case vmIntrinsics::_VectorBinaryLibOp:
758 return inline_vector_call(2);
759 case vmIntrinsics::_VectorTernaryOp:
760 return inline_vector_nary_operation(3);
761 case vmIntrinsics::_VectorFromBitsCoerced:
762 return inline_vector_frombits_coerced();
763 case vmIntrinsics::_VectorMaskOp:
764 return inline_vector_mask_operation();
765 case vmIntrinsics::_VectorLoadOp:
766 return inline_vector_mem_operation(/*is_store=*/false);
767 case vmIntrinsics::_VectorLoadMaskedOp:
768 return inline_vector_mem_masked_operation(/*is_store*/false);
769 case vmIntrinsics::_VectorStoreOp:
770 return inline_vector_mem_operation(/*is_store=*/true);
771 case vmIntrinsics::_VectorStoreMaskedOp:
772 return inline_vector_mem_masked_operation(/*is_store=*/true);
773 case vmIntrinsics::_VectorGatherOp:
774 return inline_vector_gather_scatter(/*is_scatter*/ false);
775 case vmIntrinsics::_VectorScatterOp:
776 return inline_vector_gather_scatter(/*is_scatter*/ true);
777 case vmIntrinsics::_VectorReductionCoerced:
778 return inline_vector_reduction();
779 case vmIntrinsics::_VectorTest:
780 return inline_vector_test();
781 case vmIntrinsics::_VectorBlend:
782 return inline_vector_blend();
783 case vmIntrinsics::_VectorRearrange:
784 return inline_vector_rearrange();
785 case vmIntrinsics::_VectorSelectFrom:
786 return inline_vector_select_from();
787 case vmIntrinsics::_VectorCompare:
788 return inline_vector_compare();
789 case vmIntrinsics::_VectorBroadcastInt:
790 return inline_vector_broadcast_int();
791 case vmIntrinsics::_VectorConvert:
792 return inline_vector_convert();
793 case vmIntrinsics::_VectorInsert:
794 return inline_vector_insert();
795 case vmIntrinsics::_VectorExtract:
796 return inline_vector_extract();
797 case vmIntrinsics::_VectorCompressExpand:
798 return inline_vector_compress_expand();
799 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
800 return inline_vector_select_from_two_vectors();
801 case vmIntrinsics::_IndexVector:
802 return inline_index_vector();
803 case vmIntrinsics::_IndexPartiallyInUpperRange:
804 return inline_index_partially_in_upper_range();
805
806 case vmIntrinsics::_getObjectSize:
807 return inline_getObjectSize();
808
809 case vmIntrinsics::_blackhole:
810 return inline_blackhole();
811
812 default:
813 // If you get here, it may be that someone has added a new intrinsic
814 // to the list in vmIntrinsics.hpp without implementing it here.
815 #ifndef PRODUCT
816 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
817 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
818 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
819 }
820 #endif
821 return false;
822 }
823 }
824
825 Node* LibraryCallKit::try_to_predicate(int predicate) {
826 if (!jvms()->has_method()) {
827 // Root JVMState has a null method.
828 assert(map()->memory()->Opcode() == Op_Parm, "");
829 // Insert the memory aliasing node
830 set_all_memory(reset_memory());
831 }
832 assert(merged_memory(), "");
833
834 switch (intrinsic_id()) {
835 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
836 return inline_cipherBlockChaining_AESCrypt_predicate(false);
837 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
838 return inline_cipherBlockChaining_AESCrypt_predicate(true);
839 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
840 return inline_electronicCodeBook_AESCrypt_predicate(false);
841 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
842 return inline_electronicCodeBook_AESCrypt_predicate(true);
843 case vmIntrinsics::_counterMode_AESCrypt:
844 return inline_counterMode_AESCrypt_predicate();
845 case vmIntrinsics::_digestBase_implCompressMB:
846 return inline_digestBase_implCompressMB_predicate(predicate);
847 case vmIntrinsics::_galoisCounterMode_AESCrypt:
848 return inline_galoisCounterMode_AESCrypt_predicate();
849
850 default:
851 // If you get here, it may be that someone has added a new intrinsic
852 // to the list in vmIntrinsics.hpp without implementing it here.
853 #ifndef PRODUCT
854 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
855 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
856 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
857 }
858 #endif
859 Node* slow_ctl = control();
860 set_control(top()); // No fast path intrinsic
861 return slow_ctl;
862 }
863 }
864
865 //------------------------------set_result-------------------------------
866 // Helper function for finishing intrinsics.
867 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
868 record_for_igvn(region);
869 set_control(_gvn.transform(region));
870 set_result( _gvn.transform(value));
871 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
872 }
873
874 //------------------------------generate_guard---------------------------
875 // Helper function for generating guarded fast-slow graph structures.
876 // The given 'test', if true, guards a slow path. If the test fails
877 // then a fast path can be taken. (We generally hope it fails.)
878 // In all cases, GraphKit::control() is updated to the fast path.
879 // The returned value represents the control for the slow path.
880 // The return value is never 'top'; it is either a valid control
881 // or null if it is obvious that the slow path can never be taken.
882 // Also, if region and the slow control are not null, the slow edge
883 // is appended to the region.
884 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
885 if (stopped()) {
886 // Already short circuited.
887 return nullptr;
888 }
889
890 // Build an if node and its projections.
891 // If test is true we take the slow path, which we assume is uncommon.
892 if (_gvn.type(test) == TypeInt::ZERO) {
893 // The slow branch is never taken. No need to build this guard.
894 return nullptr;
895 }
896
897 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
898
899 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
900 if (if_slow == top()) {
901 // The slow branch is never taken. No need to build this guard.
902 return nullptr;
903 }
904
905 if (region != nullptr)
906 region->add_req(if_slow);
907
908 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
909 set_control(if_fast);
910
911 return if_slow;
912 }
913
914 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
915 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
916 }
917 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
918 return generate_guard(test, region, PROB_FAIR);
919 }
920
921 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
922 Node* *pos_index) {
923 if (stopped())
924 return nullptr; // already stopped
925 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
926 return nullptr; // index is already adequately typed
927 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
928 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
929 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
930 if (is_neg != nullptr && pos_index != nullptr) {
931 // Emulate effect of Parse::adjust_map_after_if.
932 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
933 (*pos_index) = _gvn.transform(ccast);
934 }
935 return is_neg;
936 }
937
938 // Make sure that 'position' is a valid limit index, in [0..length].
939 // There are two equivalent plans for checking this:
940 // A. (offset + copyLength) unsigned<= arrayLength
941 // B. offset <= (arrayLength - copyLength)
942 // We require that all of the values above, except for the sum and
943 // difference, are already known to be non-negative.
944 // Plan A is robust in the face of overflow, if offset and copyLength
945 // are both hugely positive.
946 //
947 // Plan B is less direct and intuitive, but it does not overflow at
948 // all, since the difference of two non-negatives is always
949 // representable. Whenever Java methods must perform the equivalent
950 // check they generally use Plan B instead of Plan A.
951 // For the moment we use Plan A.
952 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
953 Node* subseq_length,
954 Node* array_length,
955 RegionNode* region) {
956 if (stopped())
957 return nullptr; // already stopped
958 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
959 if (zero_offset && subseq_length->eqv_uncast(array_length))
960 return nullptr; // common case of whole-array copy
961 Node* last = subseq_length;
962 if (!zero_offset) // last += offset
963 last = _gvn.transform(new AddINode(last, offset));
964 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
965 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
966 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
967 return is_over;
968 }
969
970 // Emit range checks for the given String.value byte array
971 void LibraryCallKit::generate_string_range_check(Node* array,
972 Node* offset,
973 Node* count,
974 bool char_count,
975 bool halt_on_oob) {
976 if (stopped()) {
977 return; // already stopped
978 }
979 RegionNode* bailout = new RegionNode(1);
980 record_for_igvn(bailout);
981 if (char_count) {
982 // Convert char count to byte count
983 count = _gvn.transform(new LShiftINode(count, intcon(1)));
984 }
985
986 // Offset and count must not be negative
987 generate_negative_guard(offset, bailout);
988 generate_negative_guard(count, bailout);
989 // Offset + count must not exceed length of array
990 generate_limit_guard(offset, count, load_array_length(array), bailout);
991
992 if (bailout->req() > 1) {
993 if (halt_on_oob) {
994 bailout = _gvn.transform(bailout)->as_Region();
995 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
996 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
997 C->root()->add_req(halt);
998 } else {
999 PreserveJVMState pjvms(this);
1000 set_control(_gvn.transform(bailout));
1001 uncommon_trap(Deoptimization::Reason_intrinsic,
1002 Deoptimization::Action_maybe_recompile);
1003 }
1004 }
1005 }
1006
1007 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1008 bool is_immutable) {
1009 ciKlass* thread_klass = env()->Thread_klass();
1010 const Type* thread_type
1011 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1012
1013 Node* thread = _gvn.transform(new ThreadLocalNode());
1014 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
1015 tls_output = thread;
1016
1017 Node* thread_obj_handle
1018 = (is_immutable
1019 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1020 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1021 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1022 thread_obj_handle = _gvn.transform(thread_obj_handle);
1023
1024 DecoratorSet decorators = IN_NATIVE;
1025 if (is_immutable) {
1026 decorators |= C2_IMMUTABLE_MEMORY;
1027 }
1028 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1029 }
1030
1031 //--------------------------generate_current_thread--------------------
1032 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1033 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1034 /*is_immutable*/false);
1035 }
1036
1037 //--------------------------generate_virtual_thread--------------------
1038 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1039 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1040 !C->method()->changes_current_thread());
1041 }
1042
1043 //------------------------------make_string_method_node------------------------
1044 // Helper method for String intrinsic functions. This version is called with
1045 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1046 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1047 // containing the lengths of str1 and str2.
1048 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1049 Node* result = nullptr;
1050 switch (opcode) {
1051 case Op_StrIndexOf:
1052 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1053 str1_start, cnt1, str2_start, cnt2, ae);
1054 break;
1055 case Op_StrComp:
1056 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1057 str1_start, cnt1, str2_start, cnt2, ae);
1058 break;
1059 case Op_StrEquals:
1060 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1061 // Use the constant length if there is one because optimized match rule may exist.
1062 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1063 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1064 break;
1065 default:
1066 ShouldNotReachHere();
1067 return nullptr;
1068 }
1069
1070 // All these intrinsics have checks.
1071 C->set_has_split_ifs(true); // Has chance for split-if optimization
1072 clear_upper_avx();
1073
1074 return _gvn.transform(result);
1075 }
1076
1077 //------------------------------inline_string_compareTo------------------------
1078 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1079 Node* arg1 = argument(0);
1080 Node* arg2 = argument(1);
1081
1082 arg1 = must_be_not_null(arg1, true);
1083 arg2 = must_be_not_null(arg2, true);
1084
1085 // Get start addr and length of first argument
1086 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1087 Node* arg1_cnt = load_array_length(arg1);
1088
1089 // Get start addr and length of second argument
1090 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1091 Node* arg2_cnt = load_array_length(arg2);
1092
1093 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1094 set_result(result);
1095 return true;
1096 }
1097
1098 //------------------------------inline_string_equals------------------------
1099 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1100 Node* arg1 = argument(0);
1101 Node* arg2 = argument(1);
1102
1103 // paths (plus control) merge
1104 RegionNode* region = new RegionNode(3);
1105 Node* phi = new PhiNode(region, TypeInt::BOOL);
1106
1107 if (!stopped()) {
1108
1109 arg1 = must_be_not_null(arg1, true);
1110 arg2 = must_be_not_null(arg2, true);
1111
1112 // Get start addr and length of first argument
1113 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1114 Node* arg1_cnt = load_array_length(arg1);
1115
1116 // Get start addr and length of second argument
1117 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1118 Node* arg2_cnt = load_array_length(arg2);
1119
1120 // Check for arg1_cnt != arg2_cnt
1121 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1122 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1123 Node* if_ne = generate_slow_guard(bol, nullptr);
1124 if (if_ne != nullptr) {
1125 phi->init_req(2, intcon(0));
1126 region->init_req(2, if_ne);
1127 }
1128
1129 // Check for count == 0 is done by assembler code for StrEquals.
1130
1131 if (!stopped()) {
1132 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1133 phi->init_req(1, equals);
1134 region->init_req(1, control());
1135 }
1136 }
1137
1138 // post merge
1139 set_control(_gvn.transform(region));
1140 record_for_igvn(region);
1141
1142 set_result(_gvn.transform(phi));
1143 return true;
1144 }
1145
1146 //------------------------------inline_array_equals----------------------------
1147 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1148 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1149 Node* arg1 = argument(0);
1150 Node* arg2 = argument(1);
1151
1152 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1153 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1154 clear_upper_avx();
1155
1156 return true;
1157 }
1158
1159
1160 //------------------------------inline_countPositives------------------------------
1161 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1162 bool LibraryCallKit::inline_countPositives() {
1163 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1164 return false;
1165 }
1166
1167 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1168 // no receiver since it is static method
1169 Node* ba = argument(0);
1170 Node* offset = argument(1);
1171 Node* len = argument(2);
1172
1173 if (VerifyIntrinsicChecks) {
1174 ba = must_be_not_null(ba, true);
1175 generate_string_range_check(ba, offset, len, false, true);
1176 if (stopped()) {
1177 return true;
1178 }
1179 }
1180
1181 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1182 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1183 set_result(_gvn.transform(result));
1184 clear_upper_avx();
1185 return true;
1186 }
1187
1188 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1189 Node* index = argument(0);
1190 Node* length = bt == T_INT ? argument(1) : argument(2);
1191 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1192 return false;
1193 }
1194
1195 // check that length is positive
1196 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1197 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1198
1199 {
1200 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1201 uncommon_trap(Deoptimization::Reason_intrinsic,
1202 Deoptimization::Action_make_not_entrant);
1203 }
1204
1205 if (stopped()) {
1206 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1207 return true;
1208 }
1209
1210 // length is now known positive, add a cast node to make this explicit
1211 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1212 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1213 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1214 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1215 casted_length = _gvn.transform(casted_length);
1216 replace_in_map(length, casted_length);
1217 length = casted_length;
1218
1219 // Use an unsigned comparison for the range check itself
1220 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1221 BoolTest::mask btest = BoolTest::lt;
1222 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1223 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1224 _gvn.set_type(rc, rc->Value(&_gvn));
1225 if (!rc_bool->is_Con()) {
1226 record_for_igvn(rc);
1227 }
1228 set_control(_gvn.transform(new IfTrueNode(rc)));
1229 {
1230 PreserveJVMState pjvms(this);
1231 set_control(_gvn.transform(new IfFalseNode(rc)));
1232 uncommon_trap(Deoptimization::Reason_range_check,
1233 Deoptimization::Action_make_not_entrant);
1234 }
1235
1236 if (stopped()) {
1237 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1238 return true;
1239 }
1240
1241 // index is now known to be >= 0 and < length, cast it
1242 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1243 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1244 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1245 result = _gvn.transform(result);
1246 set_result(result);
1247 replace_in_map(index, result);
1248 return true;
1249 }
1250
1251 //------------------------------inline_string_indexOf------------------------
1252 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1253 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1254 return false;
1255 }
1256 Node* src = argument(0);
1257 Node* tgt = argument(1);
1258
1259 // Make the merge point
1260 RegionNode* result_rgn = new RegionNode(4);
1261 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1262
1263 src = must_be_not_null(src, true);
1264 tgt = must_be_not_null(tgt, true);
1265
1266 // Get start addr and length of source string
1267 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1268 Node* src_count = load_array_length(src);
1269
1270 // Get start addr and length of substring
1271 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1272 Node* tgt_count = load_array_length(tgt);
1273
1274 Node* result = nullptr;
1275 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1276
1277 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1278 // Divide src size by 2 if String is UTF16 encoded
1279 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1280 }
1281 if (ae == StrIntrinsicNode::UU) {
1282 // Divide substring size by 2 if String is UTF16 encoded
1283 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1284 }
1285
1286 if (call_opt_stub) {
1287 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1288 StubRoutines::_string_indexof_array[ae],
1289 "stringIndexOf", TypePtr::BOTTOM, src_start,
1290 src_count, tgt_start, tgt_count);
1291 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1292 } else {
1293 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1294 result_rgn, result_phi, ae);
1295 }
1296 if (result != nullptr) {
1297 result_phi->init_req(3, result);
1298 result_rgn->init_req(3, control());
1299 }
1300 set_control(_gvn.transform(result_rgn));
1301 record_for_igvn(result_rgn);
1302 set_result(_gvn.transform(result_phi));
1303
1304 return true;
1305 }
1306
1307 //-----------------------------inline_string_indexOfI-----------------------
1308 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1309 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1310 return false;
1311 }
1312 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1313 return false;
1314 }
1315
1316 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1317 Node* src = argument(0); // byte[]
1318 Node* src_count = argument(1); // char count
1319 Node* tgt = argument(2); // byte[]
1320 Node* tgt_count = argument(3); // char count
1321 Node* from_index = argument(4); // char index
1322
1323 src = must_be_not_null(src, true);
1324 tgt = must_be_not_null(tgt, true);
1325
1326 // Multiply byte array index by 2 if String is UTF16 encoded
1327 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1328 src_count = _gvn.transform(new SubINode(src_count, from_index));
1329 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1330 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1331
1332 // Range checks
1333 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1334 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1335 if (stopped()) {
1336 return true;
1337 }
1338
1339 RegionNode* region = new RegionNode(5);
1340 Node* phi = new PhiNode(region, TypeInt::INT);
1341 Node* result = nullptr;
1342
1343 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1344
1345 if (call_opt_stub) {
1346 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1347 StubRoutines::_string_indexof_array[ae],
1348 "stringIndexOf", TypePtr::BOTTOM, src_start,
1349 src_count, tgt_start, tgt_count);
1350 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1351 } else {
1352 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1353 region, phi, ae);
1354 }
1355 if (result != nullptr) {
1356 // The result is index relative to from_index if substring was found, -1 otherwise.
1357 // Generate code which will fold into cmove.
1358 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1359 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1360
1361 Node* if_lt = generate_slow_guard(bol, nullptr);
1362 if (if_lt != nullptr) {
1363 // result == -1
1364 phi->init_req(3, result);
1365 region->init_req(3, if_lt);
1366 }
1367 if (!stopped()) {
1368 result = _gvn.transform(new AddINode(result, from_index));
1369 phi->init_req(4, result);
1370 region->init_req(4, control());
1371 }
1372 }
1373
1374 set_control(_gvn.transform(region));
1375 record_for_igvn(region);
1376 set_result(_gvn.transform(phi));
1377 clear_upper_avx();
1378
1379 return true;
1380 }
1381
1382 // Create StrIndexOfNode with fast path checks
1383 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1384 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1385 // Check for substr count > string count
1386 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1387 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1388 Node* if_gt = generate_slow_guard(bol, nullptr);
1389 if (if_gt != nullptr) {
1390 phi->init_req(1, intcon(-1));
1391 region->init_req(1, if_gt);
1392 }
1393 if (!stopped()) {
1394 // Check for substr count == 0
1395 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1396 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1397 Node* if_zero = generate_slow_guard(bol, nullptr);
1398 if (if_zero != nullptr) {
1399 phi->init_req(2, intcon(0));
1400 region->init_req(2, if_zero);
1401 }
1402 }
1403 if (!stopped()) {
1404 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1405 }
1406 return nullptr;
1407 }
1408
1409 //-----------------------------inline_string_indexOfChar-----------------------
1410 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1411 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1412 return false;
1413 }
1414 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1415 return false;
1416 }
1417 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1418 Node* src = argument(0); // byte[]
1419 Node* int_ch = argument(1);
1420 Node* from_index = argument(2);
1421 Node* max = argument(3);
1422
1423 src = must_be_not_null(src, true);
1424
1425 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1426 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1427 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1428
1429 // Range checks
1430 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1431
1432 // Check for int_ch >= 0
1433 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1434 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1435 {
1436 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1437 uncommon_trap(Deoptimization::Reason_intrinsic,
1438 Deoptimization::Action_maybe_recompile);
1439 }
1440 if (stopped()) {
1441 return true;
1442 }
1443
1444 RegionNode* region = new RegionNode(3);
1445 Node* phi = new PhiNode(region, TypeInt::INT);
1446
1447 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1448 C->set_has_split_ifs(true); // Has chance for split-if optimization
1449 _gvn.transform(result);
1450
1451 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1452 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1453
1454 Node* if_lt = generate_slow_guard(bol, nullptr);
1455 if (if_lt != nullptr) {
1456 // result == -1
1457 phi->init_req(2, result);
1458 region->init_req(2, if_lt);
1459 }
1460 if (!stopped()) {
1461 result = _gvn.transform(new AddINode(result, from_index));
1462 phi->init_req(1, result);
1463 region->init_req(1, control());
1464 }
1465 set_control(_gvn.transform(region));
1466 record_for_igvn(region);
1467 set_result(_gvn.transform(phi));
1468 clear_upper_avx();
1469
1470 return true;
1471 }
1472 //---------------------------inline_string_copy---------------------
1473 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1474 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1475 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1476 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1477 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1478 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1479 bool LibraryCallKit::inline_string_copy(bool compress) {
1480 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1481 return false;
1482 }
1483 int nargs = 5; // 2 oops, 3 ints
1484 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1485
1486 Node* src = argument(0);
1487 Node* src_offset = argument(1);
1488 Node* dst = argument(2);
1489 Node* dst_offset = argument(3);
1490 Node* length = argument(4);
1491
1492 // Check for allocation before we add nodes that would confuse
1493 // tightly_coupled_allocation()
1494 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1495
1496 // Figure out the size and type of the elements we will be copying.
1497 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1498 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1499 if (src_type == nullptr || dst_type == nullptr) {
1500 return false;
1501 }
1502 BasicType src_elem = src_type->elem()->array_element_basic_type();
1503 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1504 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1505 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1506 "Unsupported array types for inline_string_copy");
1507
1508 src = must_be_not_null(src, true);
1509 dst = must_be_not_null(dst, true);
1510
1511 // Convert char[] offsets to byte[] offsets
1512 bool convert_src = (compress && src_elem == T_BYTE);
1513 bool convert_dst = (!compress && dst_elem == T_BYTE);
1514 if (convert_src) {
1515 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1516 } else if (convert_dst) {
1517 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1518 }
1519
1520 // Range checks
1521 generate_string_range_check(src, src_offset, length, convert_src);
1522 generate_string_range_check(dst, dst_offset, length, convert_dst);
1523 if (stopped()) {
1524 return true;
1525 }
1526
1527 Node* src_start = array_element_address(src, src_offset, src_elem);
1528 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1529 // 'src_start' points to src array + scaled offset
1530 // 'dst_start' points to dst array + scaled offset
1531 Node* count = nullptr;
1532 if (compress) {
1533 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1534 } else {
1535 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1536 }
1537
1538 if (alloc != nullptr) {
1539 if (alloc->maybe_set_complete(&_gvn)) {
1540 // "You break it, you buy it."
1541 InitializeNode* init = alloc->initialization();
1542 assert(init->is_complete(), "we just did this");
1543 init->set_complete_with_arraycopy();
1544 assert(dst->is_CheckCastPP(), "sanity");
1545 assert(dst->in(0)->in(0) == init, "dest pinned");
1546 }
1547 // Do not let stores that initialize this object be reordered with
1548 // a subsequent store that would make this object accessible by
1549 // other threads.
1550 // Record what AllocateNode this StoreStore protects so that
1551 // escape analysis can go from the MemBarStoreStoreNode to the
1552 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1553 // based on the escape status of the AllocateNode.
1554 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1555 }
1556 if (compress) {
1557 set_result(_gvn.transform(count));
1558 }
1559 clear_upper_avx();
1560
1561 return true;
1562 }
1563
1564 #ifdef _LP64
1565 #define XTOP ,top() /*additional argument*/
1566 #else //_LP64
1567 #define XTOP /*no additional argument*/
1568 #endif //_LP64
1569
1570 //------------------------inline_string_toBytesU--------------------------
1571 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1572 bool LibraryCallKit::inline_string_toBytesU() {
1573 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1574 return false;
1575 }
1576 // Get the arguments.
1577 Node* value = argument(0);
1578 Node* offset = argument(1);
1579 Node* length = argument(2);
1580
1581 Node* newcopy = nullptr;
1582
1583 // Set the original stack and the reexecute bit for the interpreter to reexecute
1584 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1585 { PreserveReexecuteState preexecs(this);
1586 jvms()->set_should_reexecute(true);
1587
1588 // Check if a null path was taken unconditionally.
1589 value = null_check(value);
1590
1591 RegionNode* bailout = new RegionNode(1);
1592 record_for_igvn(bailout);
1593
1594 // Range checks
1595 generate_negative_guard(offset, bailout);
1596 generate_negative_guard(length, bailout);
1597 generate_limit_guard(offset, length, load_array_length(value), bailout);
1598 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1599 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1600
1601 if (bailout->req() > 1) {
1602 PreserveJVMState pjvms(this);
1603 set_control(_gvn.transform(bailout));
1604 uncommon_trap(Deoptimization::Reason_intrinsic,
1605 Deoptimization::Action_maybe_recompile);
1606 }
1607 if (stopped()) {
1608 return true;
1609 }
1610
1611 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1612 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1613 newcopy = new_array(klass_node, size, 0); // no arguments to push
1614 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1615 guarantee(alloc != nullptr, "created above");
1616
1617 // Calculate starting addresses.
1618 Node* src_start = array_element_address(value, offset, T_CHAR);
1619 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1620
1621 // Check if dst array address is aligned to HeapWordSize
1622 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1623 // If true, then check if src array address is aligned to HeapWordSize
1624 if (aligned) {
1625 const TypeInt* toffset = gvn().type(offset)->is_int();
1626 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1627 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1628 }
1629
1630 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1631 const char* copyfunc_name = "arraycopy";
1632 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1633 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1634 OptoRuntime::fast_arraycopy_Type(),
1635 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1636 src_start, dst_start, ConvI2X(length) XTOP);
1637 // Do not let reads from the cloned object float above the arraycopy.
1638 if (alloc->maybe_set_complete(&_gvn)) {
1639 // "You break it, you buy it."
1640 InitializeNode* init = alloc->initialization();
1641 assert(init->is_complete(), "we just did this");
1642 init->set_complete_with_arraycopy();
1643 assert(newcopy->is_CheckCastPP(), "sanity");
1644 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1645 }
1646 // Do not let stores that initialize this object be reordered with
1647 // a subsequent store that would make this object accessible by
1648 // other threads.
1649 // Record what AllocateNode this StoreStore protects so that
1650 // escape analysis can go from the MemBarStoreStoreNode to the
1651 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1652 // based on the escape status of the AllocateNode.
1653 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1654 } // original reexecute is set back here
1655
1656 C->set_has_split_ifs(true); // Has chance for split-if optimization
1657 if (!stopped()) {
1658 set_result(newcopy);
1659 }
1660 clear_upper_avx();
1661
1662 return true;
1663 }
1664
1665 //------------------------inline_string_getCharsU--------------------------
1666 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1667 bool LibraryCallKit::inline_string_getCharsU() {
1668 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1669 return false;
1670 }
1671
1672 // Get the arguments.
1673 Node* src = argument(0);
1674 Node* src_begin = argument(1);
1675 Node* src_end = argument(2); // exclusive offset (i < src_end)
1676 Node* dst = argument(3);
1677 Node* dst_begin = argument(4);
1678
1679 // Check for allocation before we add nodes that would confuse
1680 // tightly_coupled_allocation()
1681 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1682
1683 // Check if a null path was taken unconditionally.
1684 src = null_check(src);
1685 dst = null_check(dst);
1686 if (stopped()) {
1687 return true;
1688 }
1689
1690 // Get length and convert char[] offset to byte[] offset
1691 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1692 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1693
1694 // Range checks
1695 generate_string_range_check(src, src_begin, length, true);
1696 generate_string_range_check(dst, dst_begin, length, false);
1697 if (stopped()) {
1698 return true;
1699 }
1700
1701 if (!stopped()) {
1702 // Calculate starting addresses.
1703 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1704 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1705
1706 // Check if array addresses are aligned to HeapWordSize
1707 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1708 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1709 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1710 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1711
1712 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1713 const char* copyfunc_name = "arraycopy";
1714 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1715 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1716 OptoRuntime::fast_arraycopy_Type(),
1717 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1718 src_start, dst_start, ConvI2X(length) XTOP);
1719 // Do not let reads from the cloned object float above the arraycopy.
1720 if (alloc != nullptr) {
1721 if (alloc->maybe_set_complete(&_gvn)) {
1722 // "You break it, you buy it."
1723 InitializeNode* init = alloc->initialization();
1724 assert(init->is_complete(), "we just did this");
1725 init->set_complete_with_arraycopy();
1726 assert(dst->is_CheckCastPP(), "sanity");
1727 assert(dst->in(0)->in(0) == init, "dest pinned");
1728 }
1729 // Do not let stores that initialize this object be reordered with
1730 // a subsequent store that would make this object accessible by
1731 // other threads.
1732 // Record what AllocateNode this StoreStore protects so that
1733 // escape analysis can go from the MemBarStoreStoreNode to the
1734 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1735 // based on the escape status of the AllocateNode.
1736 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1737 } else {
1738 insert_mem_bar(Op_MemBarCPUOrder);
1739 }
1740 }
1741
1742 C->set_has_split_ifs(true); // Has chance for split-if optimization
1743 return true;
1744 }
1745
1746 //----------------------inline_string_char_access----------------------------
1747 // Store/Load char to/from byte[] array.
1748 // static void StringUTF16.putChar(byte[] val, int index, int c)
1749 // static char StringUTF16.getChar(byte[] val, int index)
1750 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1751 Node* value = argument(0);
1752 Node* index = argument(1);
1753 Node* ch = is_store ? argument(2) : nullptr;
1754
1755 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1756 // correctly requires matched array shapes.
1757 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1758 "sanity: byte[] and char[] bases agree");
1759 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1760 "sanity: byte[] and char[] scales agree");
1761
1762 // Bail when getChar over constants is requested: constant folding would
1763 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1764 // Java method would constant fold nicely instead.
1765 if (!is_store && value->is_Con() && index->is_Con()) {
1766 return false;
1767 }
1768
1769 // Save state and restore on bailout
1770 SavedState old_state(this);
1771
1772 value = must_be_not_null(value, true);
1773
1774 Node* adr = array_element_address(value, index, T_CHAR);
1775 if (adr->is_top()) {
1776 return false;
1777 }
1778 old_state.discard();
1779 if (is_store) {
1780 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1781 } else {
1782 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);
1783 set_result(ch);
1784 }
1785 return true;
1786 }
1787
1788
1789 //------------------------------inline_math-----------------------------------
1790 // public static double Math.abs(double)
1791 // public static double Math.sqrt(double)
1792 // public static double Math.log(double)
1793 // public static double Math.log10(double)
1794 // public static double Math.round(double)
1795 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1796 Node* arg = argument(0);
1797 Node* n = nullptr;
1798 switch (id) {
1799 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1800 case vmIntrinsics::_dsqrt:
1801 case vmIntrinsics::_dsqrt_strict:
1802 n = new SqrtDNode(C, control(), arg); break;
1803 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1804 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1805 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1806 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1807 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1808 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1809 default: fatal_unexpected_iid(id); break;
1810 }
1811 set_result(_gvn.transform(n));
1812 return true;
1813 }
1814
1815 //------------------------------inline_math-----------------------------------
1816 // public static float Math.abs(float)
1817 // public static int Math.abs(int)
1818 // public static long Math.abs(long)
1819 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1820 Node* arg = argument(0);
1821 Node* n = nullptr;
1822 switch (id) {
1823 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1824 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1825 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1826 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1827 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1828 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1829 default: fatal_unexpected_iid(id); break;
1830 }
1831 set_result(_gvn.transform(n));
1832 return true;
1833 }
1834
1835 //------------------------------runtime_math-----------------------------
1836 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1837 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1838 "must be (DD)D or (D)D type");
1839
1840 // Inputs
1841 Node* a = argument(0);
1842 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1843
1844 const TypePtr* no_memory_effects = nullptr;
1845 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1846 no_memory_effects,
1847 a, top(), b, b ? top() : nullptr);
1848 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1849 #ifdef ASSERT
1850 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1851 assert(value_top == top(), "second value must be top");
1852 #endif
1853
1854 set_result(value);
1855 return true;
1856 }
1857
1858 //------------------------------inline_math_pow-----------------------------
1859 bool LibraryCallKit::inline_math_pow() {
1860 Node* exp = argument(2);
1861 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1862 if (d != nullptr) {
1863 if (d->getd() == 2.0) {
1864 // Special case: pow(x, 2.0) => x * x
1865 Node* base = argument(0);
1866 set_result(_gvn.transform(new MulDNode(base, base)));
1867 return true;
1868 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1869 // Special case: pow(x, 0.5) => sqrt(x)
1870 Node* base = argument(0);
1871 Node* zero = _gvn.zerocon(T_DOUBLE);
1872
1873 RegionNode* region = new RegionNode(3);
1874 Node* phi = new PhiNode(region, Type::DOUBLE);
1875
1876 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1877 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1878 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1879 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1880 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1881
1882 Node* if_pow = generate_slow_guard(test, nullptr);
1883 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1884 phi->init_req(1, value_sqrt);
1885 region->init_req(1, control());
1886
1887 if (if_pow != nullptr) {
1888 set_control(if_pow);
1889 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1890 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1891 const TypePtr* no_memory_effects = nullptr;
1892 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1893 no_memory_effects, base, top(), exp, top());
1894 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1895 #ifdef ASSERT
1896 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1897 assert(value_top == top(), "second value must be top");
1898 #endif
1899 phi->init_req(2, value_pow);
1900 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1901 }
1902
1903 C->set_has_split_ifs(true); // Has chance for split-if optimization
1904 set_control(_gvn.transform(region));
1905 record_for_igvn(region);
1906 set_result(_gvn.transform(phi));
1907
1908 return true;
1909 }
1910 }
1911
1912 return StubRoutines::dpow() != nullptr ?
1913 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1914 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1915 }
1916
1917 //------------------------------inline_math_native-----------------------------
1918 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1919 switch (id) {
1920 case vmIntrinsics::_dsin:
1921 return StubRoutines::dsin() != nullptr ?
1922 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1923 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1924 case vmIntrinsics::_dcos:
1925 return StubRoutines::dcos() != nullptr ?
1926 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1927 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1928 case vmIntrinsics::_dtan:
1929 return StubRoutines::dtan() != nullptr ?
1930 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1931 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1932 case vmIntrinsics::_dsinh:
1933 return StubRoutines::dsinh() != nullptr ?
1934 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1935 case vmIntrinsics::_dtanh:
1936 return StubRoutines::dtanh() != nullptr ?
1937 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1938 case vmIntrinsics::_dcbrt:
1939 return StubRoutines::dcbrt() != nullptr ?
1940 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1941 case vmIntrinsics::_dexp:
1942 return StubRoutines::dexp() != nullptr ?
1943 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1944 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1945 case vmIntrinsics::_dlog:
1946 return StubRoutines::dlog() != nullptr ?
1947 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1948 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1949 case vmIntrinsics::_dlog10:
1950 return StubRoutines::dlog10() != nullptr ?
1951 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1952 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1953
1954 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1955 case vmIntrinsics::_ceil:
1956 case vmIntrinsics::_floor:
1957 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1958
1959 case vmIntrinsics::_dsqrt:
1960 case vmIntrinsics::_dsqrt_strict:
1961 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1962 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1963 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1964 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1965 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1966
1967 case vmIntrinsics::_dpow: return inline_math_pow();
1968 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1969 case vmIntrinsics::_fcopySign: return inline_math(id);
1970 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1971 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1972 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1973
1974 // These intrinsics are not yet correctly implemented
1975 case vmIntrinsics::_datan2:
1976 return false;
1977
1978 default:
1979 fatal_unexpected_iid(id);
1980 return false;
1981 }
1982 }
1983
1984 //----------------------------inline_notify-----------------------------------*
1985 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1986 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1987 address func;
1988 if (id == vmIntrinsics::_notify) {
1989 func = OptoRuntime::monitor_notify_Java();
1990 } else {
1991 func = OptoRuntime::monitor_notifyAll_Java();
1992 }
1993 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1994 make_slow_call_ex(call, env()->Throwable_klass(), false);
1995 return true;
1996 }
1997
1998
1999 //----------------------------inline_min_max-----------------------------------
2000 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
2001 Node* a = nullptr;
2002 Node* b = nullptr;
2003 Node* n = nullptr;
2004 switch (id) {
2005 case vmIntrinsics::_min:
2006 case vmIntrinsics::_max:
2007 case vmIntrinsics::_minF:
2008 case vmIntrinsics::_maxF:
2009 case vmIntrinsics::_minF_strict:
2010 case vmIntrinsics::_maxF_strict:
2011 case vmIntrinsics::_min_strict:
2012 case vmIntrinsics::_max_strict:
2013 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
2014 a = argument(0);
2015 b = argument(1);
2016 break;
2017 case vmIntrinsics::_minD:
2018 case vmIntrinsics::_maxD:
2019 case vmIntrinsics::_minD_strict:
2020 case vmIntrinsics::_maxD_strict:
2021 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2022 a = argument(0);
2023 b = argument(2);
2024 break;
2025 case vmIntrinsics::_minL:
2026 case vmIntrinsics::_maxL:
2027 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2028 a = argument(0);
2029 b = argument(2);
2030 break;
2031 default:
2032 fatal_unexpected_iid(id);
2033 break;
2034 }
2035
2036 switch (id) {
2037 case vmIntrinsics::_min:
2038 case vmIntrinsics::_min_strict:
2039 n = new MinINode(a, b);
2040 break;
2041 case vmIntrinsics::_max:
2042 case vmIntrinsics::_max_strict:
2043 n = new MaxINode(a, b);
2044 break;
2045 case vmIntrinsics::_minF:
2046 case vmIntrinsics::_minF_strict:
2047 n = new MinFNode(a, b);
2048 break;
2049 case vmIntrinsics::_maxF:
2050 case vmIntrinsics::_maxF_strict:
2051 n = new MaxFNode(a, b);
2052 break;
2053 case vmIntrinsics::_minD:
2054 case vmIntrinsics::_minD_strict:
2055 n = new MinDNode(a, b);
2056 break;
2057 case vmIntrinsics::_maxD:
2058 case vmIntrinsics::_maxD_strict:
2059 n = new MaxDNode(a, b);
2060 break;
2061 case vmIntrinsics::_minL:
2062 n = new MinLNode(_gvn.C, a, b);
2063 break;
2064 case vmIntrinsics::_maxL:
2065 n = new MaxLNode(_gvn.C, a, b);
2066 break;
2067 default:
2068 fatal_unexpected_iid(id);
2069 break;
2070 }
2071
2072 set_result(_gvn.transform(n));
2073 return true;
2074 }
2075
2076 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2077 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2078 env()->ArithmeticException_instance())) {
2079 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2080 // so let's bail out intrinsic rather than risking deopting again.
2081 return false;
2082 }
2083
2084 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2085 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2086 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2087 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2088
2089 {
2090 PreserveJVMState pjvms(this);
2091 PreserveReexecuteState preexecs(this);
2092 jvms()->set_should_reexecute(true);
2093
2094 set_control(slow_path);
2095 set_i_o(i_o());
2096
2097 builtin_throw(Deoptimization::Reason_intrinsic,
2098 env()->ArithmeticException_instance(),
2099 /*allow_too_many_traps*/ false);
2100 }
2101
2102 set_control(fast_path);
2103 set_result(math);
2104 return true;
2105 }
2106
2107 template <typename OverflowOp>
2108 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2109 typedef typename OverflowOp::MathOp MathOp;
2110
2111 MathOp* mathOp = new MathOp(arg1, arg2);
2112 Node* operation = _gvn.transform( mathOp );
2113 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2114 return inline_math_mathExact(operation, ofcheck);
2115 }
2116
2117 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2118 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2119 }
2120
2121 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2122 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2123 }
2124
2125 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2126 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2127 }
2128
2129 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2130 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2131 }
2132
2133 bool LibraryCallKit::inline_math_negateExactI() {
2134 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2135 }
2136
2137 bool LibraryCallKit::inline_math_negateExactL() {
2138 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2139 }
2140
2141 bool LibraryCallKit::inline_math_multiplyExactI() {
2142 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2143 }
2144
2145 bool LibraryCallKit::inline_math_multiplyExactL() {
2146 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2147 }
2148
2149 bool LibraryCallKit::inline_math_multiplyHigh() {
2150 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2151 return true;
2152 }
2153
2154 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2155 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2156 return true;
2157 }
2158
2159 inline int
2160 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2161 const TypePtr* base_type = TypePtr::NULL_PTR;
2162 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2163 if (base_type == nullptr) {
2164 // Unknown type.
2165 return Type::AnyPtr;
2166 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2167 // Since this is a null+long form, we have to switch to a rawptr.
2168 base = _gvn.transform(new CastX2PNode(offset));
2169 offset = MakeConX(0);
2170 return Type::RawPtr;
2171 } else if (base_type->base() == Type::RawPtr) {
2172 return Type::RawPtr;
2173 } else if (base_type->isa_oopptr()) {
2174 // Base is never null => always a heap address.
2175 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2176 return Type::OopPtr;
2177 }
2178 // Offset is small => always a heap address.
2179 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2180 if (offset_type != nullptr &&
2181 base_type->offset() == 0 && // (should always be?)
2182 offset_type->_lo >= 0 &&
2183 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2184 return Type::OopPtr;
2185 } else if (type == T_OBJECT) {
2186 // off heap access to an oop doesn't make any sense. Has to be on
2187 // heap.
2188 return Type::OopPtr;
2189 }
2190 // Otherwise, it might either be oop+off or null+addr.
2191 return Type::AnyPtr;
2192 } else {
2193 // No information:
2194 return Type::AnyPtr;
2195 }
2196 }
2197
2198 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2199 Node* uncasted_base = base;
2200 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2201 if (kind == Type::RawPtr) {
2202 return basic_plus_adr(top(), uncasted_base, offset);
2203 } else if (kind == Type::AnyPtr) {
2204 assert(base == uncasted_base, "unexpected base change");
2205 if (can_cast) {
2206 if (!_gvn.type(base)->speculative_maybe_null() &&
2207 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2208 // According to profiling, this access is always on
2209 // heap. Casting the base to not null and thus avoiding membars
2210 // around the access should allow better optimizations
2211 Node* null_ctl = top();
2212 base = null_check_oop(base, &null_ctl, true, true, true);
2213 assert(null_ctl->is_top(), "no null control here");
2214 return basic_plus_adr(base, offset);
2215 } else if (_gvn.type(base)->speculative_always_null() &&
2216 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2217 // According to profiling, this access is always off
2218 // heap.
2219 base = null_assert(base);
2220 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2221 offset = MakeConX(0);
2222 return basic_plus_adr(top(), raw_base, offset);
2223 }
2224 }
2225 // We don't know if it's an on heap or off heap access. Fall back
2226 // to raw memory access.
2227 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2228 return basic_plus_adr(top(), raw, offset);
2229 } else {
2230 assert(base == uncasted_base, "unexpected base change");
2231 // We know it's an on heap access so base can't be null
2232 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2233 base = must_be_not_null(base, true);
2234 }
2235 return basic_plus_adr(base, offset);
2236 }
2237 }
2238
2239 //--------------------------inline_number_methods-----------------------------
2240 // inline int Integer.numberOfLeadingZeros(int)
2241 // inline int Long.numberOfLeadingZeros(long)
2242 //
2243 // inline int Integer.numberOfTrailingZeros(int)
2244 // inline int Long.numberOfTrailingZeros(long)
2245 //
2246 // inline int Integer.bitCount(int)
2247 // inline int Long.bitCount(long)
2248 //
2249 // inline char Character.reverseBytes(char)
2250 // inline short Short.reverseBytes(short)
2251 // inline int Integer.reverseBytes(int)
2252 // inline long Long.reverseBytes(long)
2253 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2254 Node* arg = argument(0);
2255 Node* n = nullptr;
2256 switch (id) {
2257 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2258 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2259 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2260 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2261 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2262 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2263 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2264 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2265 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2266 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2267 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2268 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2269 default: fatal_unexpected_iid(id); break;
2270 }
2271 set_result(_gvn.transform(n));
2272 return true;
2273 }
2274
2275 //--------------------------inline_bitshuffle_methods-----------------------------
2276 // inline int Integer.compress(int, int)
2277 // inline int Integer.expand(int, int)
2278 // inline long Long.compress(long, long)
2279 // inline long Long.expand(long, long)
2280 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2281 Node* n = nullptr;
2282 switch (id) {
2283 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2284 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2285 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2286 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2287 default: fatal_unexpected_iid(id); break;
2288 }
2289 set_result(_gvn.transform(n));
2290 return true;
2291 }
2292
2293 //--------------------------inline_number_methods-----------------------------
2294 // inline int Integer.compareUnsigned(int, int)
2295 // inline int Long.compareUnsigned(long, long)
2296 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2297 Node* arg1 = argument(0);
2298 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2299 Node* n = nullptr;
2300 switch (id) {
2301 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2302 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2303 default: fatal_unexpected_iid(id); break;
2304 }
2305 set_result(_gvn.transform(n));
2306 return true;
2307 }
2308
2309 //--------------------------inline_unsigned_divmod_methods-----------------------------
2310 // inline int Integer.divideUnsigned(int, int)
2311 // inline int Integer.remainderUnsigned(int, int)
2312 // inline long Long.divideUnsigned(long, long)
2313 // inline long Long.remainderUnsigned(long, long)
2314 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2315 Node* n = nullptr;
2316 switch (id) {
2317 case vmIntrinsics::_divideUnsigned_i: {
2318 zero_check_int(argument(1));
2319 // Compile-time detect of null-exception
2320 if (stopped()) {
2321 return true; // keep the graph constructed so far
2322 }
2323 n = new UDivINode(control(), argument(0), argument(1));
2324 break;
2325 }
2326 case vmIntrinsics::_divideUnsigned_l: {
2327 zero_check_long(argument(2));
2328 // Compile-time detect of null-exception
2329 if (stopped()) {
2330 return true; // keep the graph constructed so far
2331 }
2332 n = new UDivLNode(control(), argument(0), argument(2));
2333 break;
2334 }
2335 case vmIntrinsics::_remainderUnsigned_i: {
2336 zero_check_int(argument(1));
2337 // Compile-time detect of null-exception
2338 if (stopped()) {
2339 return true; // keep the graph constructed so far
2340 }
2341 n = new UModINode(control(), argument(0), argument(1));
2342 break;
2343 }
2344 case vmIntrinsics::_remainderUnsigned_l: {
2345 zero_check_long(argument(2));
2346 // Compile-time detect of null-exception
2347 if (stopped()) {
2348 return true; // keep the graph constructed so far
2349 }
2350 n = new UModLNode(control(), argument(0), argument(2));
2351 break;
2352 }
2353 default: fatal_unexpected_iid(id); break;
2354 }
2355 set_result(_gvn.transform(n));
2356 return true;
2357 }
2358
2359 //----------------------------inline_unsafe_access----------------------------
2360
2361 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2362 // Attempt to infer a sharper value type from the offset and base type.
2363 ciKlass* sharpened_klass = nullptr;
2364 bool null_free = false;
2365
2366 // See if it is an instance field, with an object type.
2367 if (alias_type->field() != nullptr) {
2368 if (alias_type->field()->type()->is_klass()) {
2369 sharpened_klass = alias_type->field()->type()->as_klass();
2370 null_free = alias_type->field()->is_null_free();
2371 }
2372 }
2373
2374 const TypeOopPtr* result = nullptr;
2375 // See if it is a narrow oop array.
2376 if (adr_type->isa_aryptr()) {
2377 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2378 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2379 null_free = adr_type->is_aryptr()->is_null_free();
2380 if (elem_type != nullptr && elem_type->is_loaded()) {
2381 // Sharpen the value type.
2382 result = elem_type;
2383 }
2384 }
2385 }
2386
2387 // The sharpened class might be unloaded if there is no class loader
2388 // contraint in place.
2389 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2390 // Sharpen the value type.
2391 result = TypeOopPtr::make_from_klass(sharpened_klass);
2392 if (null_free) {
2393 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2394 }
2395 }
2396 if (result != nullptr) {
2397 #ifndef PRODUCT
2398 if (C->print_intrinsics() || C->print_inlining()) {
2399 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2400 tty->print(" sharpened value: "); result->dump(); tty->cr();
2401 }
2402 #endif
2403 }
2404 return result;
2405 }
2406
2407 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2408 switch (kind) {
2409 case Relaxed:
2410 return MO_UNORDERED;
2411 case Opaque:
2412 return MO_RELAXED;
2413 case Acquire:
2414 return MO_ACQUIRE;
2415 case Release:
2416 return MO_RELEASE;
2417 case Volatile:
2418 return MO_SEQ_CST;
2419 default:
2420 ShouldNotReachHere();
2421 return 0;
2422 }
2423 }
2424
2425 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2426 _kit(kit),
2427 _sp(kit->sp()),
2428 _jvms(kit->jvms()),
2429 _map(kit->clone_map()),
2430 _discarded(false)
2431 {
2432 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2433 Node* out = kit->control()->fast_out(i);
2434 if (out->is_CFG()) {
2435 _ctrl_succ.push(out);
2436 }
2437 }
2438 }
2439
2440 LibraryCallKit::SavedState::~SavedState() {
2441 if (_discarded) {
2442 _kit->destruct_map_clone(_map);
2443 return;
2444 }
2445 _kit->jvms()->set_map(_map);
2446 _kit->jvms()->set_sp(_sp);
2447 _map->set_jvms(_kit->jvms());
2448 _kit->set_map(_map);
2449 _kit->set_sp(_sp);
2450 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2451 Node* out = _kit->control()->fast_out(i);
2452 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2453 _kit->_gvn.hash_delete(out);
2454 out->set_req(0, _kit->C->top());
2455 _kit->C->record_for_igvn(out);
2456 --i; --imax;
2457 _kit->_gvn.hash_find_insert(out);
2458 }
2459 }
2460 }
2461
2462 void LibraryCallKit::SavedState::discard() {
2463 _discarded = true;
2464 }
2465
2466 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2467 if (callee()->is_static()) return false; // caller must have the capability!
2468 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2469 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2470 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2471 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2472
2473 if (is_reference_type(type)) {
2474 decorators |= ON_UNKNOWN_OOP_REF;
2475 }
2476
2477 if (unaligned) {
2478 decorators |= C2_UNALIGNED;
2479 }
2480
2481 #ifndef PRODUCT
2482 {
2483 ResourceMark rm;
2484 // Check the signatures.
2485 ciSignature* sig = callee()->signature();
2486 #ifdef ASSERT
2487 if (!is_store) {
2488 // Object getReference(Object base, int/long offset), etc.
2489 BasicType rtype = sig->return_type()->basic_type();
2490 assert(rtype == type, "getter must return the expected value");
2491 assert(sig->count() == 2, "oop getter has 2 arguments");
2492 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2493 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2494 } else {
2495 // void putReference(Object base, int/long offset, Object x), etc.
2496 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2497 assert(sig->count() == 3, "oop putter has 3 arguments");
2498 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2499 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2500 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2501 assert(vtype == type, "putter must accept the expected value");
2502 }
2503 #endif // ASSERT
2504 }
2505 #endif //PRODUCT
2506
2507 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2508
2509 Node* receiver = argument(0); // type: oop
2510
2511 // Build address expression.
2512 Node* heap_base_oop = top();
2513
2514 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2515 Node* base = argument(1); // type: oop
2516 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2517 Node* offset = argument(2); // type: long
2518 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2519 // to be plain byte offsets, which are also the same as those accepted
2520 // by oopDesc::field_addr.
2521 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2522 "fieldOffset must be byte-scaled");
2523
2524 if (base->is_InlineType()) {
2525 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2526 InlineTypeNode* vt = base->as_InlineType();
2527 if (offset->is_Con()) {
2528 long off = find_long_con(offset, 0);
2529 ciInlineKlass* vk = vt->type()->inline_klass();
2530 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2531 return false;
2532 }
2533
2534 ciField* field = vk->get_non_flat_field_by_offset(off);
2535 if (field != nullptr) {
2536 BasicType bt = type2field[field->type()->basic_type()];
2537 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2538 bt = T_OBJECT;
2539 }
2540 if (bt == type && !field->is_flat()) {
2541 Node* value = vt->field_value_by_offset(off, false);
2542 if (value->is_InlineType()) {
2543 value = value->as_InlineType()->adjust_scalarization_depth(this);
2544 }
2545 set_result(value);
2546 return true;
2547 }
2548 }
2549 }
2550 {
2551 // Re-execute the unsafe access if allocation triggers deoptimization.
2552 PreserveReexecuteState preexecs(this);
2553 jvms()->set_should_reexecute(true);
2554 vt = vt->buffer(this);
2555 }
2556 base = vt->get_oop();
2557 }
2558
2559 // 32-bit machines ignore the high half!
2560 offset = ConvL2X(offset);
2561
2562 // Save state and restore on bailout
2563 SavedState old_state(this);
2564
2565 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2566 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2567
2568 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2569 if (type != T_OBJECT) {
2570 decorators |= IN_NATIVE; // off-heap primitive access
2571 } else {
2572 return false; // off-heap oop accesses are not supported
2573 }
2574 } else {
2575 heap_base_oop = base; // on-heap or mixed access
2576 }
2577
2578 // Can base be null? Otherwise, always on-heap access.
2579 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2580
2581 if (!can_access_non_heap) {
2582 decorators |= IN_HEAP;
2583 }
2584
2585 Node* val = is_store ? argument(4) : nullptr;
2586
2587 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2588 if (adr_type == TypePtr::NULL_PTR) {
2589 return false; // off-heap access with zero address
2590 }
2591
2592 // Try to categorize the address.
2593 Compile::AliasType* alias_type = C->alias_type(adr_type);
2594 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2595
2596 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2597 alias_type->adr_type() == TypeAryPtr::RANGE) {
2598 return false; // not supported
2599 }
2600
2601 bool mismatched = false;
2602 BasicType bt = T_ILLEGAL;
2603 ciField* field = nullptr;
2604 if (adr_type->isa_instptr()) {
2605 const TypeInstPtr* instptr = adr_type->is_instptr();
2606 ciInstanceKlass* k = instptr->instance_klass();
2607 int off = instptr->offset();
2608 if (instptr->const_oop() != nullptr &&
2609 k == ciEnv::current()->Class_klass() &&
2610 instptr->offset() >= (k->size_helper() * wordSize)) {
2611 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2612 field = k->get_field_by_offset(off, true);
2613 } else {
2614 field = k->get_non_flat_field_by_offset(off);
2615 }
2616 if (field != nullptr) {
2617 bt = type2field[field->type()->basic_type()];
2618 }
2619 if (bt != alias_type->basic_type()) {
2620 // Type mismatch. Is it an access to a nested flat field?
2621 field = k->get_field_by_offset(off, false);
2622 if (field != nullptr) {
2623 bt = type2field[field->type()->basic_type()];
2624 }
2625 }
2626 assert(bt == alias_type->basic_type(), "should match");
2627 } else {
2628 bt = alias_type->basic_type();
2629 }
2630
2631 if (bt != T_ILLEGAL) {
2632 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2633 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2634 // Alias type doesn't differentiate between byte[] and boolean[]).
2635 // Use address type to get the element type.
2636 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2637 }
2638 if (is_reference_type(bt, true)) {
2639 // accessing an array field with getReference is not a mismatch
2640 bt = T_OBJECT;
2641 }
2642 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2643 // Don't intrinsify mismatched object accesses
2644 return false;
2645 }
2646 mismatched = (bt != type);
2647 } else if (alias_type->adr_type()->isa_oopptr()) {
2648 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2649 }
2650
2651 old_state.discard();
2652 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2653
2654 if (mismatched) {
2655 decorators |= C2_MISMATCHED;
2656 }
2657
2658 // First guess at the value type.
2659 const Type *value_type = Type::get_const_basic_type(type);
2660
2661 // Figure out the memory ordering.
2662 decorators |= mo_decorator_for_access_kind(kind);
2663
2664 if (!is_store) {
2665 if (type == T_OBJECT) {
2666 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2667 if (tjp != nullptr) {
2668 value_type = tjp;
2669 }
2670 }
2671 }
2672
2673 receiver = null_check(receiver);
2674 if (stopped()) {
2675 return true;
2676 }
2677 // Heap pointers get a null-check from the interpreter,
2678 // as a courtesy. However, this is not guaranteed by Unsafe,
2679 // and it is not possible to fully distinguish unintended nulls
2680 // from intended ones in this API.
2681
2682 if (!is_store) {
2683 Node* p = nullptr;
2684 // Try to constant fold a load from a constant field
2685
2686 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2687 // final or stable field
2688 p = make_constant_from_field(field, heap_base_oop);
2689 }
2690
2691 if (p == nullptr) { // Could not constant fold the load
2692 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2693 const TypeOopPtr* ptr = value_type->make_oopptr();
2694 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2695 // Load a non-flattened inline type from memory
2696 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2697 }
2698 // Normalize the value returned by getBoolean in the following cases
2699 if (type == T_BOOLEAN &&
2700 (mismatched ||
2701 heap_base_oop == top() || // - heap_base_oop is null or
2702 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2703 // and the unsafe access is made to large offset
2704 // (i.e., larger than the maximum offset necessary for any
2705 // field access)
2706 ) {
2707 IdealKit ideal = IdealKit(this);
2708 #define __ ideal.
2709 IdealVariable normalized_result(ideal);
2710 __ declarations_done();
2711 __ set(normalized_result, p);
2712 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2713 __ set(normalized_result, ideal.ConI(1));
2714 ideal.end_if();
2715 final_sync(ideal);
2716 p = __ value(normalized_result);
2717 #undef __
2718 }
2719 }
2720 if (type == T_ADDRESS) {
2721 p = gvn().transform(new CastP2XNode(nullptr, p));
2722 p = ConvX2UL(p);
2723 }
2724 // The load node has the control of the preceding MemBarCPUOrder. All
2725 // following nodes will have the control of the MemBarCPUOrder inserted at
2726 // the end of this method. So, pushing the load onto the stack at a later
2727 // point is fine.
2728 set_result(p);
2729 } else {
2730 if (bt == T_ADDRESS) {
2731 // Repackage the long as a pointer.
2732 val = ConvL2X(val);
2733 val = gvn().transform(new CastX2PNode(val));
2734 }
2735 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2736 }
2737
2738 return true;
2739 }
2740
2741 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2742 #ifdef ASSERT
2743 {
2744 ResourceMark rm;
2745 // Check the signatures.
2746 ciSignature* sig = callee()->signature();
2747 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2748 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2749 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2750 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2751 if (is_store) {
2752 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2753 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2754 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2755 } else {
2756 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2757 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2758 }
2759 }
2760 #endif // ASSERT
2761
2762 assert(kind == Relaxed, "Only plain accesses for now");
2763 if (callee()->is_static()) {
2764 // caller must have the capability!
2765 return false;
2766 }
2767 C->set_has_unsafe_access(true);
2768
2769 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2770 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2771 // parameter valueType is not a constant
2772 return false;
2773 }
2774 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2775 if (!mirror_type->is_inlinetype()) {
2776 // Dead code
2777 return false;
2778 }
2779 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2780
2781 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2782 if (layout_type == nullptr || !layout_type->is_con()) {
2783 // parameter layoutKind is not a constant
2784 return false;
2785 }
2786 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2787 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2788 "invalid layoutKind %d", layout_type->get_con());
2789 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2790 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2791 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2792 "unexpected layoutKind %d", layout_type->get_con());
2793
2794 null_check(argument(0));
2795 if (stopped()) {
2796 return true;
2797 }
2798
2799 Node* base = must_be_not_null(argument(1), true);
2800 Node* offset = argument(2);
2801 const Type* base_type = _gvn.type(base);
2802
2803 Node* ptr;
2804 bool immutable_memory = false;
2805 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2806 if (base_type->isa_instptr()) {
2807 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2808 if (offset_type == nullptr || !offset_type->is_con()) {
2809 // Offset into a non-array should be a constant
2810 decorators |= C2_MISMATCHED;
2811 } else {
2812 int offset_con = checked_cast<int>(offset_type->get_con());
2813 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2814 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2815 if (field == nullptr) {
2816 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2817 decorators |= C2_MISMATCHED;
2818 } else {
2819 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2820 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2821 immutable_memory = field->is_strict() && field->is_final();
2822
2823 if (base->is_InlineType()) {
2824 assert(!is_store, "Cannot store into a non-larval value object");
2825 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2826 return true;
2827 }
2828 }
2829 }
2830
2831 if (base->is_InlineType()) {
2832 assert(!is_store, "Cannot store into a non-larval value object");
2833 base = base->as_InlineType()->buffer(this, true);
2834 }
2835 ptr = basic_plus_adr(base, ConvL2X(offset));
2836 } else if (base_type->isa_aryptr()) {
2837 decorators |= IS_ARRAY;
2838 if (layout == LayoutKind::REFERENCE) {
2839 if (!base_type->is_aryptr()->is_not_flat()) {
2840 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2841 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2842 replace_in_map(base, new_base);
2843 base = new_base;
2844 }
2845 ptr = basic_plus_adr(base, ConvL2X(offset));
2846 } else {
2847 if (UseArrayFlattening) {
2848 // Flat array must have an exact type
2849 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2850 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2851 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2852 replace_in_map(base, new_base);
2853 base = new_base;
2854 ptr = basic_plus_adr(base, ConvL2X(offset));
2855 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2856 if (ptr_type->field_offset().get() != 0) {
2857 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2858 }
2859 } else {
2860 uncommon_trap(Deoptimization::Reason_intrinsic,
2861 Deoptimization::Action_none);
2862 return true;
2863 }
2864 }
2865 } else {
2866 decorators |= C2_MISMATCHED;
2867 ptr = basic_plus_adr(base, ConvL2X(offset));
2868 }
2869
2870 if (is_store) {
2871 Node* value = argument(6);
2872 const Type* value_type = _gvn.type(value);
2873 if (!value_type->is_inlinetypeptr()) {
2874 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2875 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2876 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2877 replace_in_map(value, new_value);
2878 value = new_value;
2879 }
2880
2881 assert(value_type->inline_klass() == value_klass, "value is of type %s while valueType is %s", value_type->inline_klass()->name()->as_utf8(), value_klass->name()->as_utf8());
2882 if (layout == LayoutKind::REFERENCE) {
2883 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2884 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2885 } else {
2886 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2887 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2888 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2889 }
2890
2891 return true;
2892 } else {
2893 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2894 InlineTypeNode* result;
2895 if (layout == LayoutKind::REFERENCE) {
2896 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2897 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2898 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2899 } else {
2900 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2901 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2902 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2903 }
2904
2905 set_result(result);
2906 return true;
2907 }
2908 }
2909
2910 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2911 Node* receiver = argument(0);
2912 Node* value = argument(1);
2913
2914 const Type* type = gvn().type(value);
2915 if (!type->is_inlinetypeptr()) {
2916 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2917 return false;
2918 }
2919
2920 null_check(receiver);
2921 if (stopped()) {
2922 return true;
2923 }
2924
2925 value = null_check(value);
2926 if (stopped()) {
2927 return true;
2928 }
2929
2930 ciInlineKlass* vk = type->inline_klass();
2931 Node* klass = makecon(TypeKlassPtr::make(vk));
2932 Node* obj = new_instance(klass);
2933 AllocateNode::Ideal_allocation(obj)->_larval = true;
2934
2935 assert(value->is_InlineType(), "must be an InlineTypeNode");
2936 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2937 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2938
2939 set_result(obj);
2940 return true;
2941 }
2942
2943 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2944 Node* receiver = argument(0);
2945 Node* buffer = argument(1);
2946
2947 const Type* type = gvn().type(buffer);
2948 if (!type->is_inlinetypeptr()) {
2949 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2950 return false;
2951 }
2952
2953 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2954 if (alloc == nullptr) {
2955 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2956 return false;
2957 }
2958
2959 null_check(receiver);
2960 if (stopped()) {
2961 return true;
2962 }
2963
2964 // Unset the larval bit in the object header
2965 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2966 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2967 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2968
2969 // We must ensure that the buffer is properly published
2970 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2971 assert(!type->maybe_null(), "result of an allocation should not be null");
2972 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
2973 return true;
2974 }
2975
2976 //----------------------------inline_unsafe_load_store----------------------------
2977 // This method serves a couple of different customers (depending on LoadStoreKind):
2978 //
2979 // LS_cmp_swap:
2980 //
2981 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2982 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2983 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2984 //
2985 // LS_cmp_swap_weak:
2986 //
2987 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2988 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2989 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2990 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2991 //
2992 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2993 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2994 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2995 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2996 //
2997 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2998 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2999 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
3000 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
3001 //
3002 // LS_cmp_exchange:
3003 //
3004 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
3005 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
3006 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
3007 //
3008 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
3009 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
3010 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
3011 //
3012 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
3013 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
3014 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
3015 //
3016 // LS_get_add:
3017 //
3018 // int getAndAddInt( Object o, long offset, int delta)
3019 // long getAndAddLong(Object o, long offset, long delta)
3020 //
3021 // LS_get_set:
3022 //
3023 // int getAndSet(Object o, long offset, int newValue)
3024 // long getAndSet(Object o, long offset, long newValue)
3025 // Object getAndSet(Object o, long offset, Object newValue)
3026 //
3027 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
3028 // This basic scheme here is the same as inline_unsafe_access, but
3029 // differs in enough details that combining them would make the code
3030 // overly confusing. (This is a true fact! I originally combined
3031 // them, but even I was confused by it!) As much code/comments as
3032 // possible are retained from inline_unsafe_access though to make
3033 // the correspondences clearer. - dl
3034
3035 if (callee()->is_static()) return false; // caller must have the capability!
3036
3037 DecoratorSet decorators = C2_UNSAFE_ACCESS;
3038 decorators |= mo_decorator_for_access_kind(access_kind);
3039
3040 #ifndef PRODUCT
3041 BasicType rtype;
3042 {
3043 ResourceMark rm;
3044 // Check the signatures.
3045 ciSignature* sig = callee()->signature();
3046 rtype = sig->return_type()->basic_type();
3047 switch(kind) {
3048 case LS_get_add:
3049 case LS_get_set: {
3050 // Check the signatures.
3051 #ifdef ASSERT
3052 assert(rtype == type, "get and set must return the expected type");
3053 assert(sig->count() == 3, "get and set has 3 arguments");
3054 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
3055 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
3056 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
3057 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
3058 #endif // ASSERT
3059 break;
3060 }
3061 case LS_cmp_swap:
3062 case LS_cmp_swap_weak: {
3063 // Check the signatures.
3064 #ifdef ASSERT
3065 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3066 assert(sig->count() == 4, "CAS has 4 arguments");
3067 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3068 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3069 #endif // ASSERT
3070 break;
3071 }
3072 case LS_cmp_exchange: {
3073 // Check the signatures.
3074 #ifdef ASSERT
3075 assert(rtype == type, "CAS must return the expected type");
3076 assert(sig->count() == 4, "CAS has 4 arguments");
3077 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3078 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3079 #endif // ASSERT
3080 break;
3081 }
3082 default:
3083 ShouldNotReachHere();
3084 }
3085 }
3086 #endif //PRODUCT
3087
3088 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3089
3090 // Get arguments:
3091 Node* receiver = nullptr;
3092 Node* base = nullptr;
3093 Node* offset = nullptr;
3094 Node* oldval = nullptr;
3095 Node* newval = nullptr;
3096 switch(kind) {
3097 case LS_cmp_swap:
3098 case LS_cmp_swap_weak:
3099 case LS_cmp_exchange: {
3100 const bool two_slot_type = type2size[type] == 2;
3101 receiver = argument(0); // type: oop
3102 base = argument(1); // type: oop
3103 offset = argument(2); // type: long
3104 oldval = argument(4); // type: oop, int, or long
3105 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3106 break;
3107 }
3108 case LS_get_add:
3109 case LS_get_set: {
3110 receiver = argument(0); // type: oop
3111 base = argument(1); // type: oop
3112 offset = argument(2); // type: long
3113 oldval = nullptr;
3114 newval = argument(4); // type: oop, int, or long
3115 break;
3116 }
3117 default:
3118 ShouldNotReachHere();
3119 }
3120
3121 // Build field offset expression.
3122 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3123 // to be plain byte offsets, which are also the same as those accepted
3124 // by oopDesc::field_addr.
3125 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3126 // 32-bit machines ignore the high half of long offsets
3127 offset = ConvL2X(offset);
3128 // Save state and restore on bailout
3129 SavedState old_state(this);
3130 Node* adr = make_unsafe_address(base, offset,type, false);
3131 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3132
3133 Compile::AliasType* alias_type = C->alias_type(adr_type);
3134 BasicType bt = alias_type->basic_type();
3135 if (bt != T_ILLEGAL &&
3136 (is_reference_type(bt) != (type == T_OBJECT))) {
3137 // Don't intrinsify mismatched object accesses.
3138 return false;
3139 }
3140
3141 old_state.discard();
3142
3143 // For CAS, unlike inline_unsafe_access, there seems no point in
3144 // trying to refine types. Just use the coarse types here.
3145 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3146 const Type *value_type = Type::get_const_basic_type(type);
3147
3148 switch (kind) {
3149 case LS_get_set:
3150 case LS_cmp_exchange: {
3151 if (type == T_OBJECT) {
3152 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3153 if (tjp != nullptr) {
3154 value_type = tjp;
3155 }
3156 }
3157 break;
3158 }
3159 case LS_cmp_swap:
3160 case LS_cmp_swap_weak:
3161 case LS_get_add:
3162 break;
3163 default:
3164 ShouldNotReachHere();
3165 }
3166
3167 // Null check receiver.
3168 receiver = null_check(receiver);
3169 if (stopped()) {
3170 return true;
3171 }
3172
3173 int alias_idx = C->get_alias_index(adr_type);
3174
3175 if (is_reference_type(type)) {
3176 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3177
3178 if (oldval != nullptr && oldval->is_InlineType()) {
3179 // Re-execute the unsafe access if allocation triggers deoptimization.
3180 PreserveReexecuteState preexecs(this);
3181 jvms()->set_should_reexecute(true);
3182 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3183 }
3184 if (newval != nullptr && newval->is_InlineType()) {
3185 // Re-execute the unsafe access if allocation triggers deoptimization.
3186 PreserveReexecuteState preexecs(this);
3187 jvms()->set_should_reexecute(true);
3188 newval = newval->as_InlineType()->buffer(this)->get_oop();
3189 }
3190
3191 // Transformation of a value which could be null pointer (CastPP #null)
3192 // could be delayed during Parse (for example, in adjust_map_after_if()).
3193 // Execute transformation here to avoid barrier generation in such case.
3194 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3195 newval = _gvn.makecon(TypePtr::NULL_PTR);
3196
3197 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3198 // Refine the value to a null constant, when it is known to be null
3199 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3200 }
3201 }
3202
3203 Node* result = nullptr;
3204 switch (kind) {
3205 case LS_cmp_exchange: {
3206 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3207 oldval, newval, value_type, type, decorators);
3208 break;
3209 }
3210 case LS_cmp_swap_weak:
3211 decorators |= C2_WEAK_CMPXCHG;
3212 case LS_cmp_swap: {
3213 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3214 oldval, newval, value_type, type, decorators);
3215 break;
3216 }
3217 case LS_get_set: {
3218 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3219 newval, value_type, type, decorators);
3220 break;
3221 }
3222 case LS_get_add: {
3223 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3224 newval, value_type, type, decorators);
3225 break;
3226 }
3227 default:
3228 ShouldNotReachHere();
3229 }
3230
3231 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3232 set_result(result);
3233 return true;
3234 }
3235
3236 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3237 // Regardless of form, don't allow previous ld/st to move down,
3238 // then issue acquire, release, or volatile mem_bar.
3239 insert_mem_bar(Op_MemBarCPUOrder);
3240 switch(id) {
3241 case vmIntrinsics::_loadFence:
3242 insert_mem_bar(Op_LoadFence);
3243 return true;
3244 case vmIntrinsics::_storeFence:
3245 insert_mem_bar(Op_StoreFence);
3246 return true;
3247 case vmIntrinsics::_storeStoreFence:
3248 insert_mem_bar(Op_StoreStoreFence);
3249 return true;
3250 case vmIntrinsics::_fullFence:
3251 insert_mem_bar(Op_MemBarVolatile);
3252 return true;
3253 default:
3254 fatal_unexpected_iid(id);
3255 return false;
3256 }
3257 }
3258
3259 // private native int arrayInstanceBaseOffset0(Object[] array);
3260 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3261 Node* array = argument(1);
3262 Node* klass_node = load_object_klass(array);
3263
3264 jint layout_con = Klass::_lh_neutral_value;
3265 Node* layout_val = get_layout_helper(klass_node, layout_con);
3266 int layout_is_con = (layout_val == nullptr);
3267
3268 Node* header_size = nullptr;
3269 if (layout_is_con) {
3270 int hsize = Klass::layout_helper_header_size(layout_con);
3271 header_size = intcon(hsize);
3272 } else {
3273 Node* hss = intcon(Klass::_lh_header_size_shift);
3274 Node* hsm = intcon(Klass::_lh_header_size_mask);
3275 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3276 header_size = _gvn.transform(new AndINode(header_size, hsm));
3277 }
3278 set_result(header_size);
3279 return true;
3280 }
3281
3282 // private native int arrayInstanceIndexScale0(Object[] array);
3283 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3284 Node* array = argument(1);
3285 Node* klass_node = load_object_klass(array);
3286
3287 jint layout_con = Klass::_lh_neutral_value;
3288 Node* layout_val = get_layout_helper(klass_node, layout_con);
3289 int layout_is_con = (layout_val == nullptr);
3290
3291 Node* element_size = nullptr;
3292 if (layout_is_con) {
3293 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3294 int elem_size = 1 << log_element_size;
3295 element_size = intcon(elem_size);
3296 } else {
3297 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3298 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3299 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3300 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3301 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3302 }
3303 set_result(element_size);
3304 return true;
3305 }
3306
3307 // private native int arrayLayout0(Object[] array);
3308 bool LibraryCallKit::inline_arrayLayout() {
3309 RegionNode* region = new RegionNode(2);
3310 Node* phi = new PhiNode(region, TypeInt::POS);
3311
3312 Node* array = argument(1);
3313 Node* klass_node = load_object_klass(array);
3314 generate_refArray_guard(klass_node, region);
3315 if (region->req() == 3) {
3316 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3317 }
3318
3319 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3320 Node* layout_kind_addr = basic_plus_adr(klass_node, layout_kind_offset);
3321 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3322
3323 region->init_req(1, control());
3324 phi->init_req(1, layout_kind);
3325
3326 set_control(_gvn.transform(region));
3327 set_result(_gvn.transform(phi));
3328 return true;
3329 }
3330
3331 // private native int[] getFieldMap0(Class <?> c);
3332 // int offset = c._klass._acmp_maps_offset;
3333 // return (int[])c.obj_field(offset);
3334 bool LibraryCallKit::inline_getFieldMap() {
3335 if (!UseAltSubstitutabilityMethod) {
3336 return false;
3337 }
3338
3339 Node* mirror = argument(1);
3340 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3341
3342 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3343 Node* field_map_offset_addr = basic_plus_adr(klass, field_map_offset_offset);
3344 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3345 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3346
3347 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3348 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3349 // TODO 8350865 Remove this
3350 val_type = val_type->cast_to_not_flat(true)->cast_to_not_null_free(true);
3351 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3352
3353 set_result(map);
3354 return true;
3355 }
3356
3357 bool LibraryCallKit::inline_onspinwait() {
3358 insert_mem_bar(Op_OnSpinWait);
3359 return true;
3360 }
3361
3362 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3363 if (!kls->is_Con()) {
3364 return true;
3365 }
3366 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3367 if (klsptr == nullptr) {
3368 return true;
3369 }
3370 ciInstanceKlass* ik = klsptr->instance_klass();
3371 // don't need a guard for a klass that is already initialized
3372 return !ik->is_initialized();
3373 }
3374
3375 //----------------------------inline_unsafe_writeback0-------------------------
3376 // public native void Unsafe.writeback0(long address)
3377 bool LibraryCallKit::inline_unsafe_writeback0() {
3378 if (!Matcher::has_match_rule(Op_CacheWB)) {
3379 return false;
3380 }
3381 #ifndef PRODUCT
3382 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3383 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3384 ciSignature* sig = callee()->signature();
3385 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3386 #endif
3387 null_check_receiver(); // null-check, then ignore
3388 Node *addr = argument(1);
3389 addr = new CastX2PNode(addr);
3390 addr = _gvn.transform(addr);
3391 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3392 flush = _gvn.transform(flush);
3393 set_memory(flush, TypeRawPtr::BOTTOM);
3394 return true;
3395 }
3396
3397 //----------------------------inline_unsafe_writeback0-------------------------
3398 // public native void Unsafe.writeback0(long address)
3399 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3400 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3401 return false;
3402 }
3403 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3404 return false;
3405 }
3406 #ifndef PRODUCT
3407 assert(Matcher::has_match_rule(Op_CacheWB),
3408 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3409 : "found match rule for CacheWBPostSync but not CacheWB"));
3410
3411 #endif
3412 null_check_receiver(); // null-check, then ignore
3413 Node *sync;
3414 if (is_pre) {
3415 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3416 } else {
3417 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3418 }
3419 sync = _gvn.transform(sync);
3420 set_memory(sync, TypeRawPtr::BOTTOM);
3421 return true;
3422 }
3423
3424 //----------------------------inline_unsafe_allocate---------------------------
3425 // public native Object Unsafe.allocateInstance(Class<?> cls);
3426 bool LibraryCallKit::inline_unsafe_allocate() {
3427
3428 #if INCLUDE_JVMTI
3429 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3430 return false;
3431 }
3432 #endif //INCLUDE_JVMTI
3433
3434 if (callee()->is_static()) return false; // caller must have the capability!
3435
3436 null_check_receiver(); // null-check, then ignore
3437 Node* cls = null_check(argument(1));
3438 if (stopped()) return true;
3439
3440 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3441 kls = null_check(kls);
3442 if (stopped()) return true; // argument was like int.class
3443
3444 #if INCLUDE_JVMTI
3445 // Don't try to access new allocated obj in the intrinsic.
3446 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3447 // Deoptimize and allocate in interpreter instead.
3448 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3449 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3450 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3451 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3452 {
3453 BuildCutout unless(this, tst, PROB_MAX);
3454 uncommon_trap(Deoptimization::Reason_intrinsic,
3455 Deoptimization::Action_make_not_entrant);
3456 }
3457 if (stopped()) {
3458 return true;
3459 }
3460 #endif //INCLUDE_JVMTI
3461
3462 Node* test = nullptr;
3463 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3464 // Note: The argument might still be an illegal value like
3465 // Serializable.class or Object[].class. The runtime will handle it.
3466 // But we must make an explicit check for initialization.
3467 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3468 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3469 // can generate code to load it as unsigned byte.
3470 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3471 Node* bits = intcon(InstanceKlass::fully_initialized);
3472 test = _gvn.transform(new SubINode(inst, bits));
3473 // The 'test' is non-zero if we need to take a slow path.
3474 }
3475 Node* obj = nullptr;
3476 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3477 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3478 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3479 } else {
3480 obj = new_instance(kls, test);
3481 }
3482 set_result(obj);
3483 return true;
3484 }
3485
3486 //------------------------inline_native_time_funcs--------------
3487 // inline code for System.currentTimeMillis() and System.nanoTime()
3488 // these have the same type and signature
3489 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3490 const TypeFunc* tf = OptoRuntime::void_long_Type();
3491 const TypePtr* no_memory_effects = nullptr;
3492 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3493 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3494 #ifdef ASSERT
3495 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3496 assert(value_top == top(), "second value must be top");
3497 #endif
3498 set_result(value);
3499 return true;
3500 }
3501
3502 //--------------------inline_native_vthread_start_transition--------------------
3503 // inline void startTransition(boolean is_mount);
3504 // inline void startFinalTransition();
3505 // Pseudocode of implementation:
3506 //
3507 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3508 // carrier->set_is_in_vthread_transition(true);
3509 // OrderAccess::storeload();
3510 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3511 // + global_vthread_transition_disable_count();
3512 // if (disable_requests > 0) {
3513 // slow path: runtime call
3514 // }
3515 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3516 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3517 IdealKit ideal(this);
3518
3519 Node* thread = ideal.thread();
3520 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3521 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3522 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3523 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3524 insert_mem_bar(Op_MemBarVolatile);
3525 ideal.sync_kit(this);
3526
3527 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3528 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3529 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3530 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3531 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3532
3533 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3534 sync_kit(ideal);
3535 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3536 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3537 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3538 ideal.sync_kit(this);
3539 }
3540 ideal.end_if();
3541
3542 final_sync(ideal);
3543 return true;
3544 }
3545
3546 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3547 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3548 IdealKit ideal(this);
3549
3550 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3551 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3552
3553 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3554 sync_kit(ideal);
3555 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3556 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3557 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3558 ideal.sync_kit(this);
3559 } ideal.else_(); {
3560 Node* thread = ideal.thread();
3561 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3562 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3563
3564 sync_kit(ideal);
3565 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3566 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3567 ideal.sync_kit(this);
3568 } ideal.end_if();
3569
3570 final_sync(ideal);
3571 return true;
3572 }
3573
3574 #if INCLUDE_JVMTI
3575
3576 // Always update the is_disable_suspend bit.
3577 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3578 if (!DoJVMTIVirtualThreadTransitions) {
3579 return true;
3580 }
3581 IdealKit ideal(this);
3582
3583 {
3584 // unconditionally update the is_disable_suspend bit in current JavaThread
3585 Node* thread = ideal.thread();
3586 Node* arg = _gvn.transform(argument(0)); // argument for notification
3587 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3588 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3589
3590 sync_kit(ideal);
3591 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3592 ideal.sync_kit(this);
3593 }
3594 final_sync(ideal);
3595
3596 return true;
3597 }
3598
3599 #endif // INCLUDE_JVMTI
3600
3601 #ifdef JFR_HAVE_INTRINSICS
3602
3603 /**
3604 * if oop->klass != null
3605 * // normal class
3606 * epoch = _epoch_state ? 2 : 1
3607 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3608 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3609 * }
3610 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3611 * else
3612 * // primitive class
3613 * if oop->array_klass != null
3614 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3615 * else
3616 * id = LAST_TYPE_ID + 1 // void class path
3617 * if (!signaled)
3618 * signaled = true
3619 */
3620 bool LibraryCallKit::inline_native_classID() {
3621 Node* cls = argument(0);
3622
3623 IdealKit ideal(this);
3624 #define __ ideal.
3625 IdealVariable result(ideal); __ declarations_done();
3626 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3627 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3628 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3629
3630
3631 __ if_then(kls, BoolTest::ne, null()); {
3632 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3633 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3634
3635 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3636 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3637 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3638 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3639 mask = _gvn.transform(new OrLNode(mask, epoch));
3640 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3641
3642 float unlikely = PROB_UNLIKELY(0.999);
3643 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3644 sync_kit(ideal);
3645 make_runtime_call(RC_LEAF,
3646 OptoRuntime::class_id_load_barrier_Type(),
3647 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3648 "class id load barrier",
3649 TypePtr::BOTTOM,
3650 kls);
3651 ideal.sync_kit(this);
3652 } __ end_if();
3653
3654 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3655 } __ else_(); {
3656 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3657 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3658 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3659 __ if_then(array_kls, BoolTest::ne, null()); {
3660 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3661 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3662 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3663 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3664 } __ else_(); {
3665 // void class case
3666 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3667 } __ end_if();
3668
3669 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3670 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3671 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3672 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3673 } __ end_if();
3674 } __ end_if();
3675
3676 final_sync(ideal);
3677 set_result(ideal.value(result));
3678 #undef __
3679 return true;
3680 }
3681
3682 //------------------------inline_native_jvm_commit------------------
3683 bool LibraryCallKit::inline_native_jvm_commit() {
3684 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3685
3686 // Save input memory and i_o state.
3687 Node* input_memory_state = reset_memory();
3688 set_all_memory(input_memory_state);
3689 Node* input_io_state = i_o();
3690
3691 // TLS.
3692 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3693 // Jfr java buffer.
3694 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3695 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3696 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3697
3698 // Load the current value of the notified field in the JfrThreadLocal.
3699 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3700 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3701
3702 // Test for notification.
3703 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3704 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3705 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3706
3707 // True branch, is notified.
3708 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3709 set_control(is_notified);
3710
3711 // Reset notified state.
3712 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3713 Node* notified_reset_memory = reset_memory();
3714
3715 // 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.
3716 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3717 // Convert the machine-word to a long.
3718 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3719
3720 // False branch, not notified.
3721 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3722 set_control(not_notified);
3723 set_all_memory(input_memory_state);
3724
3725 // Arg is the next position as a long.
3726 Node* arg = argument(0);
3727 // Convert long to machine-word.
3728 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3729
3730 // Store the next_position to the underlying jfr java buffer.
3731 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3732
3733 Node* commit_memory = reset_memory();
3734 set_all_memory(commit_memory);
3735
3736 // 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.
3737 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3738 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3739 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3740
3741 // And flags with lease constant.
3742 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3743
3744 // Branch on lease to conditionalize returning the leased java buffer.
3745 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3746 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3747 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3748
3749 // False branch, not a lease.
3750 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3751
3752 // True branch, is lease.
3753 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3754 set_control(is_lease);
3755
3756 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3757 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3758 OptoRuntime::void_void_Type(),
3759 SharedRuntime::jfr_return_lease(),
3760 "return_lease", TypePtr::BOTTOM);
3761 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3762
3763 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3764 record_for_igvn(lease_compare_rgn);
3765 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3766 record_for_igvn(lease_compare_mem);
3767 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3768 record_for_igvn(lease_compare_io);
3769 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3770 record_for_igvn(lease_result_value);
3771
3772 // Update control and phi nodes.
3773 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3774 lease_compare_rgn->init_req(_false_path, not_lease);
3775
3776 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3777 lease_compare_mem->init_req(_false_path, commit_memory);
3778
3779 lease_compare_io->init_req(_true_path, i_o());
3780 lease_compare_io->init_req(_false_path, input_io_state);
3781
3782 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3783 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3784
3785 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3786 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3787 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3788 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3789
3790 // Update control and phi nodes.
3791 result_rgn->init_req(_true_path, is_notified);
3792 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3793
3794 result_mem->init_req(_true_path, notified_reset_memory);
3795 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3796
3797 result_io->init_req(_true_path, input_io_state);
3798 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3799
3800 result_value->init_req(_true_path, current_pos);
3801 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3802
3803 // Set output state.
3804 set_control(_gvn.transform(result_rgn));
3805 set_all_memory(_gvn.transform(result_mem));
3806 set_i_o(_gvn.transform(result_io));
3807 set_result(result_rgn, result_value);
3808 return true;
3809 }
3810
3811 /*
3812 * The intrinsic is a model of this pseudo-code:
3813 *
3814 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3815 * jobject h_event_writer = tl->java_event_writer();
3816 * if (h_event_writer == nullptr) {
3817 * return nullptr;
3818 * }
3819 * oop threadObj = Thread::threadObj();
3820 * oop vthread = java_lang_Thread::vthread(threadObj);
3821 * traceid tid;
3822 * bool pinVirtualThread;
3823 * bool excluded;
3824 * if (vthread != threadObj) { // i.e. current thread is virtual
3825 * tid = java_lang_Thread::tid(vthread);
3826 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3827 * pinVirtualThread = VMContinuations;
3828 * excluded = vthread_epoch_raw & excluded_mask;
3829 * if (!excluded) {
3830 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3831 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3832 * if (vthread_epoch != current_epoch) {
3833 * write_checkpoint();
3834 * }
3835 * }
3836 * } else {
3837 * tid = java_lang_Thread::tid(threadObj);
3838 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3839 * pinVirtualThread = false;
3840 * excluded = thread_epoch_raw & excluded_mask;
3841 * }
3842 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3843 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3844 * if (tid_in_event_writer != tid) {
3845 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3846 * setField(event_writer, "excluded", excluded);
3847 * setField(event_writer, "threadID", tid);
3848 * }
3849 * return event_writer
3850 */
3851 bool LibraryCallKit::inline_native_getEventWriter() {
3852 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3853
3854 // Save input memory and i_o state.
3855 Node* input_memory_state = reset_memory();
3856 set_all_memory(input_memory_state);
3857 Node* input_io_state = i_o();
3858
3859 // The most significant bit of the u2 is used to denote thread exclusion
3860 Node* excluded_shift = _gvn.intcon(15);
3861 Node* excluded_mask = _gvn.intcon(1 << 15);
3862 // The epoch generation is the range [1-32767]
3863 Node* epoch_mask = _gvn.intcon(32767);
3864
3865 // TLS
3866 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3867
3868 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3869 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3870
3871 // Load the eventwriter jobject handle.
3872 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3873
3874 // Null check the jobject handle.
3875 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3876 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3877 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3878
3879 // False path, jobj is null.
3880 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3881
3882 // True path, jobj is not null.
3883 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3884
3885 set_control(jobj_is_not_null);
3886
3887 // Load the threadObj for the CarrierThread.
3888 Node* threadObj = generate_current_thread(tls_ptr);
3889
3890 // Load the vthread.
3891 Node* vthread = generate_virtual_thread(tls_ptr);
3892
3893 // If vthread != threadObj, this is a virtual thread.
3894 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3895 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3896 IfNode* iff_vthread_not_equal_threadObj =
3897 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3898
3899 // False branch, fallback to threadObj.
3900 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3901 set_control(vthread_equal_threadObj);
3902
3903 // Load the tid field from the vthread object.
3904 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3905
3906 // Load the raw epoch value from the threadObj.
3907 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3908 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3909 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3910 TypeInt::CHAR, T_CHAR,
3911 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3912
3913 // Mask off the excluded information from the epoch.
3914 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3915
3916 // True branch, this is a virtual thread.
3917 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3918 set_control(vthread_not_equal_threadObj);
3919
3920 // Load the tid field from the vthread object.
3921 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3922
3923 // Continuation support determines if a virtual thread should be pinned.
3924 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3925 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3926
3927 // Load the raw epoch value from the vthread.
3928 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3929 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3930 TypeInt::CHAR, T_CHAR,
3931 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3932
3933 // Mask off the excluded information from the epoch.
3934 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3935
3936 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3937 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3938 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3939 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3940
3941 // False branch, vthread is excluded, no need to write epoch info.
3942 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3943
3944 // True branch, vthread is included, update epoch info.
3945 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3946 set_control(included);
3947
3948 // Get epoch value.
3949 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3950
3951 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3952 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3953 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3954
3955 // Compare the epoch in the vthread to the current epoch generation.
3956 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3957 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3958 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3959
3960 // False path, epoch is equal, checkpoint information is valid.
3961 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3962
3963 // True path, epoch is not equal, write a checkpoint for the vthread.
3964 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3965
3966 set_control(epoch_is_not_equal);
3967
3968 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3969 // The call also updates the native thread local thread id and the vthread with the current epoch.
3970 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3971 OptoRuntime::jfr_write_checkpoint_Type(),
3972 SharedRuntime::jfr_write_checkpoint(),
3973 "write_checkpoint", TypePtr::BOTTOM);
3974 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3975
3976 // vthread epoch != current epoch
3977 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3978 record_for_igvn(epoch_compare_rgn);
3979 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3980 record_for_igvn(epoch_compare_mem);
3981 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3982 record_for_igvn(epoch_compare_io);
3983
3984 // Update control and phi nodes.
3985 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3986 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3987 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3988 epoch_compare_mem->init_req(_false_path, input_memory_state);
3989 epoch_compare_io->init_req(_true_path, i_o());
3990 epoch_compare_io->init_req(_false_path, input_io_state);
3991
3992 // excluded != true
3993 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3994 record_for_igvn(exclude_compare_rgn);
3995 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3996 record_for_igvn(exclude_compare_mem);
3997 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3998 record_for_igvn(exclude_compare_io);
3999
4000 // Update control and phi nodes.
4001 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
4002 exclude_compare_rgn->init_req(_false_path, excluded);
4003 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
4004 exclude_compare_mem->init_req(_false_path, input_memory_state);
4005 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
4006 exclude_compare_io->init_req(_false_path, input_io_state);
4007
4008 // vthread != threadObj
4009 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
4010 record_for_igvn(vthread_compare_rgn);
4011 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4012 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
4013 record_for_igvn(vthread_compare_io);
4014 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
4015 record_for_igvn(tid);
4016 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
4017 record_for_igvn(exclusion);
4018 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
4019 record_for_igvn(pinVirtualThread);
4020
4021 // Update control and phi nodes.
4022 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
4023 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
4024 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
4025 vthread_compare_mem->init_req(_false_path, input_memory_state);
4026 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
4027 vthread_compare_io->init_req(_false_path, input_io_state);
4028 tid->init_req(_true_path, _gvn.transform(vthread_tid));
4029 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
4030 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4031 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
4032 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
4033 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
4034
4035 // Update branch state.
4036 set_control(_gvn.transform(vthread_compare_rgn));
4037 set_all_memory(_gvn.transform(vthread_compare_mem));
4038 set_i_o(_gvn.transform(vthread_compare_io));
4039
4040 // Load the event writer oop by dereferencing the jobject handle.
4041 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
4042 assert(klass_EventWriter->is_loaded(), "invariant");
4043 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
4044 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
4045 const TypeOopPtr* const xtype = aklass->as_instance_type();
4046 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
4047 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
4048
4049 // Load the current thread id from the event writer object.
4050 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
4051 // Get the field offset to, conditionally, store an updated tid value later.
4052 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
4053 // Get the field offset to, conditionally, store an updated exclusion value later.
4054 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
4055 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
4056 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
4057
4058 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
4059 record_for_igvn(event_writer_tid_compare_rgn);
4060 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4061 record_for_igvn(event_writer_tid_compare_mem);
4062 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
4063 record_for_igvn(event_writer_tid_compare_io);
4064
4065 // Compare the current tid from the thread object to what is currently stored in the event writer object.
4066 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
4067 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
4068 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
4069
4070 // False path, tids are the same.
4071 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
4072
4073 // True path, tid is not equal, need to update the tid in the event writer.
4074 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
4075 record_for_igvn(tid_is_not_equal);
4076
4077 // Store the pin state to the event writer.
4078 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
4079
4080 // Store the exclusion state to the event writer.
4081 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
4082 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
4083
4084 // Store the tid to the event writer.
4085 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
4086
4087 // Update control and phi nodes.
4088 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
4089 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
4090 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4091 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
4092 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
4093 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
4094
4095 // Result of top level CFG, Memory, IO and Value.
4096 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4097 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4098 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
4099 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
4100
4101 // Result control.
4102 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
4103 result_rgn->init_req(_false_path, jobj_is_null);
4104
4105 // Result memory.
4106 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
4107 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4108
4109 // Result IO.
4110 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
4111 result_io->init_req(_false_path, _gvn.transform(input_io_state));
4112
4113 // Result value.
4114 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
4115 result_value->init_req(_false_path, null()); // return null
4116
4117 // Set output state.
4118 set_control(_gvn.transform(result_rgn));
4119 set_all_memory(_gvn.transform(result_mem));
4120 set_i_o(_gvn.transform(result_io));
4121 set_result(result_rgn, result_value);
4122 return true;
4123 }
4124
4125 /*
4126 * The intrinsic is a model of this pseudo-code:
4127 *
4128 * JfrThreadLocal* const tl = thread->jfr_thread_local();
4129 * if (carrierThread != thread) { // is virtual thread
4130 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
4131 * bool excluded = vthread_epoch_raw & excluded_mask;
4132 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
4133 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
4134 * if (!excluded) {
4135 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
4136 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
4137 * }
4138 * AtomicAccess::release_store(&tl->_vthread, true);
4139 * return;
4140 * }
4141 * AtomicAccess::release_store(&tl->_vthread, false);
4142 */
4143 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4144 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4145
4146 Node* input_memory_state = reset_memory();
4147 set_all_memory(input_memory_state);
4148
4149 // The most significant bit of the u2 is used to denote thread exclusion
4150 Node* excluded_mask = _gvn.intcon(1 << 15);
4151 // The epoch generation is the range [1-32767]
4152 Node* epoch_mask = _gvn.intcon(32767);
4153
4154 Node* const carrierThread = generate_current_thread(jt);
4155 // If thread != carrierThread, this is a virtual thread.
4156 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4157 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4158 IfNode* iff_thread_not_equal_carrierThread =
4159 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4160
4161 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4162
4163 // False branch, is carrierThread.
4164 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4165 // Store release
4166 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4167
4168 set_all_memory(input_memory_state);
4169
4170 // True branch, is virtual thread.
4171 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4172 set_control(thread_not_equal_carrierThread);
4173
4174 // Load the raw epoch value from the vthread.
4175 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4176 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4177 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4178
4179 // Mask off the excluded information from the epoch.
4180 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4181
4182 // Load the tid field from the thread.
4183 Node* tid = load_field_from_object(thread, "tid", "J");
4184
4185 // Store the vthread tid to the jfr thread local.
4186 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4187 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4188
4189 // Branch is_excluded to conditionalize updating the epoch .
4190 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4191 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4192 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4193
4194 // True branch, vthread is excluded, no need to write epoch info.
4195 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4196 set_control(excluded);
4197 Node* vthread_is_excluded = _gvn.intcon(1);
4198
4199 // False branch, vthread is included, update epoch info.
4200 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4201 set_control(included);
4202 Node* vthread_is_included = _gvn.intcon(0);
4203
4204 // Get epoch value.
4205 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4206
4207 // Store the vthread epoch to the jfr thread local.
4208 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4209 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4210
4211 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4212 record_for_igvn(excluded_rgn);
4213 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4214 record_for_igvn(excluded_mem);
4215 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4216 record_for_igvn(exclusion);
4217
4218 // Merge the excluded control and memory.
4219 excluded_rgn->init_req(_true_path, excluded);
4220 excluded_rgn->init_req(_false_path, included);
4221 excluded_mem->init_req(_true_path, tid_memory);
4222 excluded_mem->init_req(_false_path, included_memory);
4223 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4224 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4225
4226 // Set intermediate state.
4227 set_control(_gvn.transform(excluded_rgn));
4228 set_all_memory(excluded_mem);
4229
4230 // Store the vthread exclusion state to the jfr thread local.
4231 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4232 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4233
4234 // Store release
4235 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4236
4237 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4238 record_for_igvn(thread_compare_rgn);
4239 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4240 record_for_igvn(thread_compare_mem);
4241 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4242 record_for_igvn(vthread);
4243
4244 // Merge the thread_compare control and memory.
4245 thread_compare_rgn->init_req(_true_path, control());
4246 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4247 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4248 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4249
4250 // Set output state.
4251 set_control(_gvn.transform(thread_compare_rgn));
4252 set_all_memory(_gvn.transform(thread_compare_mem));
4253 }
4254
4255 #endif // JFR_HAVE_INTRINSICS
4256
4257 //------------------------inline_native_currentCarrierThread------------------
4258 bool LibraryCallKit::inline_native_currentCarrierThread() {
4259 Node* junk = nullptr;
4260 set_result(generate_current_thread(junk));
4261 return true;
4262 }
4263
4264 //------------------------inline_native_currentThread------------------
4265 bool LibraryCallKit::inline_native_currentThread() {
4266 Node* junk = nullptr;
4267 set_result(generate_virtual_thread(junk));
4268 return true;
4269 }
4270
4271 //------------------------inline_native_setVthread------------------
4272 bool LibraryCallKit::inline_native_setCurrentThread() {
4273 assert(C->method()->changes_current_thread(),
4274 "method changes current Thread but is not annotated ChangesCurrentThread");
4275 Node* arr = argument(1);
4276 Node* thread = _gvn.transform(new ThreadLocalNode());
4277 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4278 Node* thread_obj_handle
4279 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4280 thread_obj_handle = _gvn.transform(thread_obj_handle);
4281 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4282 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4283
4284 // Change the _monitor_owner_id of the JavaThread
4285 Node* tid = load_field_from_object(arr, "tid", "J");
4286 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4287 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4288
4289 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4290 return true;
4291 }
4292
4293 const Type* LibraryCallKit::scopedValueCache_type() {
4294 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4295 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4296 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4297
4298 // Because we create the scopedValue cache lazily we have to make the
4299 // type of the result BotPTR.
4300 bool xk = etype->klass_is_exact();
4301 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4302 return objects_type;
4303 }
4304
4305 Node* LibraryCallKit::scopedValueCache_helper() {
4306 Node* thread = _gvn.transform(new ThreadLocalNode());
4307 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4308 // We cannot use immutable_memory() because we might flip onto a
4309 // different carrier thread, at which point we'll need to use that
4310 // carrier thread's cache.
4311 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4312 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4313 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4314 }
4315
4316 //------------------------inline_native_scopedValueCache------------------
4317 bool LibraryCallKit::inline_native_scopedValueCache() {
4318 Node* cache_obj_handle = scopedValueCache_helper();
4319 const Type* objects_type = scopedValueCache_type();
4320 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4321
4322 return true;
4323 }
4324
4325 //------------------------inline_native_setScopedValueCache------------------
4326 bool LibraryCallKit::inline_native_setScopedValueCache() {
4327 Node* arr = argument(0);
4328 Node* cache_obj_handle = scopedValueCache_helper();
4329 const Type* objects_type = scopedValueCache_type();
4330
4331 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4332 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4333
4334 return true;
4335 }
4336
4337 //------------------------inline_native_Continuation_pin and unpin-----------
4338
4339 // Shared implementation routine for both pin and unpin.
4340 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4341 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4342
4343 // Save input memory.
4344 Node* input_memory_state = reset_memory();
4345 set_all_memory(input_memory_state);
4346
4347 // TLS
4348 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4349 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4350 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4351
4352 // Null check the last continuation object.
4353 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4354 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4355 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4356
4357 // False path, last continuation is null.
4358 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4359
4360 // True path, last continuation is not null.
4361 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4362
4363 set_control(continuation_is_not_null);
4364
4365 // Load the pin count from the last continuation.
4366 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4367 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4368
4369 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4370 Node* pin_count_rhs;
4371 if (unpin) {
4372 pin_count_rhs = _gvn.intcon(0);
4373 } else {
4374 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4375 }
4376 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4377 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4378 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4379
4380 // True branch, pin count over/underflow.
4381 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4382 {
4383 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4384 // which will throw IllegalStateException for pin count over/underflow.
4385 // No memory changed so far - we can use memory create by reset_memory()
4386 // at the beginning of this intrinsic. No need to call reset_memory() again.
4387 PreserveJVMState pjvms(this);
4388 set_control(pin_count_over_underflow);
4389 uncommon_trap(Deoptimization::Reason_intrinsic,
4390 Deoptimization::Action_none);
4391 assert(stopped(), "invariant");
4392 }
4393
4394 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4395 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4396 set_control(valid_pin_count);
4397
4398 Node* next_pin_count;
4399 if (unpin) {
4400 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4401 } else {
4402 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4403 }
4404
4405 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4406
4407 // Result of top level CFG and Memory.
4408 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4409 record_for_igvn(result_rgn);
4410 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4411 record_for_igvn(result_mem);
4412
4413 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4414 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4415 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4416 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4417
4418 // Set output state.
4419 set_control(_gvn.transform(result_rgn));
4420 set_all_memory(_gvn.transform(result_mem));
4421
4422 return true;
4423 }
4424
4425 //---------------------------load_mirror_from_klass----------------------------
4426 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4427 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4428 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
4429 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4430 // mirror = ((OopHandle)mirror)->resolve();
4431 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4432 }
4433
4434 //-----------------------load_klass_from_mirror_common-------------------------
4435 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4436 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4437 // and branch to the given path on the region.
4438 // If never_see_null, take an uncommon trap on null, so we can optimistically
4439 // compile for the non-null case.
4440 // If the region is null, force never_see_null = true.
4441 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4442 bool never_see_null,
4443 RegionNode* region,
4444 int null_path,
4445 int offset) {
4446 if (region == nullptr) never_see_null = true;
4447 Node* p = basic_plus_adr(mirror, offset);
4448 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4449 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4450 Node* null_ctl = top();
4451 kls = null_check_oop(kls, &null_ctl, never_see_null);
4452 if (region != nullptr) {
4453 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4454 region->init_req(null_path, null_ctl);
4455 } else {
4456 assert(null_ctl == top(), "no loose ends");
4457 }
4458 return kls;
4459 }
4460
4461 //--------------------(inline_native_Class_query helpers)---------------------
4462 // Use this for JVM_ACC_INTERFACE.
4463 // Fall through if (mods & mask) == bits, take the guard otherwise.
4464 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4465 ByteSize offset, const Type* type, BasicType bt) {
4466 // Branch around if the given klass has the given modifier bit set.
4467 // Like generate_guard, adds a new path onto the region.
4468 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4469 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4470 Node* mask = intcon(modifier_mask);
4471 Node* bits = intcon(modifier_bits);
4472 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4473 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4474 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4475 return generate_fair_guard(bol, region);
4476 }
4477
4478 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4479 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4480 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4481 }
4482
4483 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4484 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4485 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4486 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4487 }
4488
4489 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4490 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4491 }
4492
4493 //-------------------------inline_native_Class_query-------------------
4494 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4495 const Type* return_type = TypeInt::BOOL;
4496 Node* prim_return_value = top(); // what happens if it's a primitive class?
4497 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4498 bool expect_prim = false; // most of these guys expect to work on refs
4499
4500 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4501
4502 Node* mirror = argument(0);
4503 Node* obj = top();
4504
4505 switch (id) {
4506 case vmIntrinsics::_isInstance:
4507 // nothing is an instance of a primitive type
4508 prim_return_value = intcon(0);
4509 obj = argument(1);
4510 break;
4511 case vmIntrinsics::_isHidden:
4512 prim_return_value = intcon(0);
4513 break;
4514 case vmIntrinsics::_getSuperclass:
4515 prim_return_value = null();
4516 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4517 break;
4518 default:
4519 fatal_unexpected_iid(id);
4520 break;
4521 }
4522
4523 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4524 if (mirror_con == nullptr) return false; // cannot happen?
4525
4526 #ifndef PRODUCT
4527 if (C->print_intrinsics() || C->print_inlining()) {
4528 ciType* k = mirror_con->java_mirror_type();
4529 if (k) {
4530 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4531 k->print_name();
4532 tty->cr();
4533 }
4534 }
4535 #endif
4536
4537 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4538 RegionNode* region = new RegionNode(PATH_LIMIT);
4539 record_for_igvn(region);
4540 PhiNode* phi = new PhiNode(region, return_type);
4541
4542 // The mirror will never be null of Reflection.getClassAccessFlags, however
4543 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4544 // if it is. See bug 4774291.
4545
4546 // For Reflection.getClassAccessFlags(), the null check occurs in
4547 // the wrong place; see inline_unsafe_access(), above, for a similar
4548 // situation.
4549 mirror = null_check(mirror);
4550 // If mirror or obj is dead, only null-path is taken.
4551 if (stopped()) return true;
4552
4553 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4554
4555 // Now load the mirror's klass metaobject, and null-check it.
4556 // Side-effects region with the control path if the klass is null.
4557 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4558 // If kls is null, we have a primitive mirror.
4559 phi->init_req(_prim_path, prim_return_value);
4560 if (stopped()) { set_result(region, phi); return true; }
4561 bool safe_for_replace = (region->in(_prim_path) == top());
4562
4563 Node* p; // handy temp
4564 Node* null_ctl;
4565
4566 // Now that we have the non-null klass, we can perform the real query.
4567 // For constant classes, the query will constant-fold in LoadNode::Value.
4568 Node* query_value = top();
4569 switch (id) {
4570 case vmIntrinsics::_isInstance:
4571 // nothing is an instance of a primitive type
4572 query_value = gen_instanceof(obj, kls, safe_for_replace);
4573 break;
4574
4575 case vmIntrinsics::_isHidden:
4576 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4577 if (generate_hidden_class_guard(kls, region) != nullptr)
4578 // A guard was added. If the guard is taken, it was an hidden class.
4579 phi->add_req(intcon(1));
4580 // If we fall through, it's a plain class.
4581 query_value = intcon(0);
4582 break;
4583
4584
4585 case vmIntrinsics::_getSuperclass:
4586 // The rules here are somewhat unfortunate, but we can still do better
4587 // with random logic than with a JNI call.
4588 // Interfaces store null or Object as _super, but must report null.
4589 // Arrays store an intermediate super as _super, but must report Object.
4590 // Other types can report the actual _super.
4591 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4592 if (generate_array_guard(kls, region) != nullptr) {
4593 // A guard was added. If the guard is taken, it was an array.
4594 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4595 }
4596 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4597 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4598 if (generate_interface_guard(kls, region) != nullptr) {
4599 // A guard was added. If the guard is taken, it was an interface.
4600 phi->add_req(null());
4601 }
4602 // If we fall through, it's a plain class. Get its _super.
4603 if (!stopped()) {
4604 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4605 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4606 null_ctl = top();
4607 kls = null_check_oop(kls, &null_ctl);
4608 if (null_ctl != top()) {
4609 // If the guard is taken, Object.superClass is null (both klass and mirror).
4610 region->add_req(null_ctl);
4611 phi ->add_req(null());
4612 }
4613 if (!stopped()) {
4614 query_value = load_mirror_from_klass(kls);
4615 }
4616 }
4617 break;
4618
4619 default:
4620 fatal_unexpected_iid(id);
4621 break;
4622 }
4623
4624 // Fall-through is the normal case of a query to a real class.
4625 phi->init_req(1, query_value);
4626 region->init_req(1, control());
4627
4628 C->set_has_split_ifs(true); // Has chance for split-if optimization
4629 set_result(region, phi);
4630 return true;
4631 }
4632
4633
4634 //-------------------------inline_Class_cast-------------------
4635 bool LibraryCallKit::inline_Class_cast() {
4636 Node* mirror = argument(0); // Class
4637 Node* obj = argument(1);
4638 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4639 if (mirror_con == nullptr) {
4640 return false; // dead path (mirror->is_top()).
4641 }
4642 if (obj == nullptr || obj->is_top()) {
4643 return false; // dead path
4644 }
4645 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4646
4647 // First, see if Class.cast() can be folded statically.
4648 // java_mirror_type() returns non-null for compile-time Class constants.
4649 ciType* tm = mirror_con->java_mirror_type();
4650 if (tm != nullptr && tm->is_klass() &&
4651 tp != nullptr) {
4652 if (!tp->is_loaded()) {
4653 // Don't use intrinsic when class is not loaded.
4654 return false;
4655 } else {
4656 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4657 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4658 if (static_res == Compile::SSC_always_true) {
4659 // isInstance() is true - fold the code.
4660 set_result(obj);
4661 return true;
4662 } else if (static_res == Compile::SSC_always_false) {
4663 // Don't use intrinsic, have to throw ClassCastException.
4664 // If the reference is null, the non-intrinsic bytecode will
4665 // be optimized appropriately.
4666 return false;
4667 }
4668 }
4669 }
4670
4671 // Bailout intrinsic and do normal inlining if exception path is frequent.
4672 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4673 return false;
4674 }
4675
4676 // Generate dynamic checks.
4677 // Class.cast() is java implementation of _checkcast bytecode.
4678 // Do checkcast (Parse::do_checkcast()) optimizations here.
4679
4680 mirror = null_check(mirror);
4681 // If mirror is dead, only null-path is taken.
4682 if (stopped()) {
4683 return true;
4684 }
4685
4686 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4687 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4688 RegionNode* region = new RegionNode(PATH_LIMIT);
4689 record_for_igvn(region);
4690
4691 // Now load the mirror's klass metaobject, and null-check it.
4692 // If kls is null, we have a primitive mirror and
4693 // nothing is an instance of a primitive type.
4694 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4695
4696 Node* res = top();
4697 Node* io = i_o();
4698 Node* mem = merged_memory();
4699 if (!stopped()) {
4700
4701 Node* bad_type_ctrl = top();
4702 // Do checkcast optimizations.
4703 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4704 region->init_req(_bad_type_path, bad_type_ctrl);
4705 }
4706 if (region->in(_prim_path) != top() ||
4707 region->in(_bad_type_path) != top() ||
4708 region->in(_npe_path) != top()) {
4709 // Let Interpreter throw ClassCastException.
4710 PreserveJVMState pjvms(this);
4711 set_control(_gvn.transform(region));
4712 // Set IO and memory because gen_checkcast may override them when buffering inline types
4713 set_i_o(io);
4714 set_all_memory(mem);
4715 uncommon_trap(Deoptimization::Reason_intrinsic,
4716 Deoptimization::Action_maybe_recompile);
4717 }
4718 if (!stopped()) {
4719 set_result(res);
4720 }
4721 return true;
4722 }
4723
4724
4725 //--------------------------inline_native_subtype_check------------------------
4726 // This intrinsic takes the JNI calls out of the heart of
4727 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4728 bool LibraryCallKit::inline_native_subtype_check() {
4729 // Pull both arguments off the stack.
4730 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4731 args[0] = argument(0);
4732 args[1] = argument(1);
4733 Node* klasses[2]; // corresponding Klasses: superk, subk
4734 klasses[0] = klasses[1] = top();
4735
4736 enum {
4737 // A full decision tree on {superc is prim, subc is prim}:
4738 _prim_0_path = 1, // {P,N} => false
4739 // {P,P} & superc!=subc => false
4740 _prim_same_path, // {P,P} & superc==subc => true
4741 _prim_1_path, // {N,P} => false
4742 _ref_subtype_path, // {N,N} & subtype check wins => true
4743 _both_ref_path, // {N,N} & subtype check loses => false
4744 PATH_LIMIT
4745 };
4746
4747 RegionNode* region = new RegionNode(PATH_LIMIT);
4748 RegionNode* prim_region = new RegionNode(2);
4749 Node* phi = new PhiNode(region, TypeInt::BOOL);
4750 record_for_igvn(region);
4751 record_for_igvn(prim_region);
4752
4753 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4754 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4755 int class_klass_offset = java_lang_Class::klass_offset();
4756
4757 // First null-check both mirrors and load each mirror's klass metaobject.
4758 int which_arg;
4759 for (which_arg = 0; which_arg <= 1; which_arg++) {
4760 Node* arg = args[which_arg];
4761 arg = null_check(arg);
4762 if (stopped()) break;
4763 args[which_arg] = arg;
4764
4765 Node* p = basic_plus_adr(arg, class_klass_offset);
4766 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4767 klasses[which_arg] = _gvn.transform(kls);
4768 }
4769
4770 // Having loaded both klasses, test each for null.
4771 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4772 for (which_arg = 0; which_arg <= 1; which_arg++) {
4773 Node* kls = klasses[which_arg];
4774 Node* null_ctl = top();
4775 kls = null_check_oop(kls, &null_ctl, never_see_null);
4776 if (which_arg == 0) {
4777 prim_region->init_req(1, null_ctl);
4778 } else {
4779 region->init_req(_prim_1_path, null_ctl);
4780 }
4781 if (stopped()) break;
4782 klasses[which_arg] = kls;
4783 }
4784
4785 if (!stopped()) {
4786 // now we have two reference types, in klasses[0..1]
4787 Node* subk = klasses[1]; // the argument to isAssignableFrom
4788 Node* superk = klasses[0]; // the receiver
4789 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4790 region->set_req(_ref_subtype_path, control());
4791 }
4792
4793 // If both operands are primitive (both klasses null), then
4794 // we must return true when they are identical primitives.
4795 // It is convenient to test this after the first null klass check.
4796 // This path is also used if superc is a value mirror.
4797 set_control(_gvn.transform(prim_region));
4798 if (!stopped()) {
4799 // Since superc is primitive, make a guard for the superc==subc case.
4800 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4801 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4802 generate_fair_guard(bol_eq, region);
4803 if (region->req() == PATH_LIMIT+1) {
4804 // A guard was added. If the added guard is taken, superc==subc.
4805 region->swap_edges(PATH_LIMIT, _prim_same_path);
4806 region->del_req(PATH_LIMIT);
4807 }
4808 region->set_req(_prim_0_path, control()); // Not equal after all.
4809 }
4810
4811 // these are the only paths that produce 'true':
4812 phi->set_req(_prim_same_path, intcon(1));
4813 phi->set_req(_ref_subtype_path, intcon(1));
4814
4815 // pull together the cases:
4816 assert(region->req() == PATH_LIMIT, "sane region");
4817 for (uint i = 1; i < region->req(); i++) {
4818 Node* ctl = region->in(i);
4819 if (ctl == nullptr || ctl == top()) {
4820 region->set_req(i, top());
4821 phi ->set_req(i, top());
4822 } else if (phi->in(i) == nullptr) {
4823 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4824 }
4825 }
4826
4827 set_control(_gvn.transform(region));
4828 set_result(_gvn.transform(phi));
4829 return true;
4830 }
4831
4832 //---------------------generate_array_guard_common------------------------
4833 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4834
4835 if (stopped()) {
4836 return nullptr;
4837 }
4838
4839 // Like generate_guard, adds a new path onto the region.
4840 jint layout_con = 0;
4841 Node* layout_val = get_layout_helper(kls, layout_con);
4842 if (layout_val == nullptr) {
4843 bool query = 0;
4844 switch(kind) {
4845 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4846 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4847 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4848 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4849 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4850 default:
4851 ShouldNotReachHere();
4852 }
4853 if (!query) {
4854 return nullptr; // never a branch
4855 } else { // always a branch
4856 Node* always_branch = control();
4857 if (region != nullptr)
4858 region->add_req(always_branch);
4859 set_control(top());
4860 return always_branch;
4861 }
4862 }
4863 unsigned int value = 0;
4864 BoolTest::mask btest = BoolTest::illegal;
4865 switch(kind) {
4866 case RefArray:
4867 case NonRefArray: {
4868 value = Klass::_lh_array_tag_ref_value;
4869 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4870 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4871 break;
4872 }
4873 case TypeArray: {
4874 value = Klass::_lh_array_tag_type_value;
4875 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4876 btest = BoolTest::eq;
4877 break;
4878 }
4879 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4880 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4881 default:
4882 ShouldNotReachHere();
4883 }
4884 // Now test the correct condition.
4885 jint nval = (jint)value;
4886 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4887 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4888 Node* ctrl = generate_fair_guard(bol, region);
4889 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4890 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4891 // Keep track of the fact that 'obj' is an array to prevent
4892 // array specific accesses from floating above the guard.
4893 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4894 }
4895 return ctrl;
4896 }
4897
4898 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4899 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4900 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4901 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4902 assert(null_free || atomic, "nullable implies atomic");
4903 Node* componentType = argument(0);
4904 Node* length = argument(1);
4905 Node* init_val = null_free ? argument(2) : nullptr;
4906
4907 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4908 if (tp != nullptr) {
4909 ciInstanceKlass* ik = tp->instance_klass();
4910 if (ik == C->env()->Class_klass()) {
4911 ciType* t = tp->java_mirror_type();
4912 if (t != nullptr && t->is_inlinetype()) {
4913
4914 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4915 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4916
4917 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4918 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4919 return false;
4920 }
4921
4922 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4923 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4924 if (null_free) {
4925 if (init_val->is_InlineType()) {
4926 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4927 // Zeroing is enough because the init value is the all-zero value
4928 init_val = nullptr;
4929 } else {
4930 init_val = init_val->as_InlineType()->buffer(this);
4931 }
4932 }
4933 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4934 // If we insert a checkcast here, we can be sure that init_val is an InlineTypeNode, so
4935 // when we folded a field load from an allocation (e.g. during escape analysis), we can
4936 // remove the check init_val->is_InlineType().
4937 }
4938 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4939 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4940 assert(arytype->is_null_free() == null_free, "inconsistency");
4941 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4942 set_result(obj);
4943 return true;
4944 }
4945 }
4946 }
4947 }
4948 return false;
4949 }
4950
4951 // public static native boolean ValueClass::isFlatArray(Object array);
4952 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4953 // public static native boolean ValueClass::isAtomicArray(Object array);
4954 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4955 Node* array = argument(0);
4956
4957 Node* bol;
4958 switch(check) {
4959 case IsFlat:
4960 // TODO 8350865 Use the object version here instead of loading the klass
4961 // The problem is that PhaseMacroExpand::expand_flatarraycheck_node can only handle some IR shapes and will fail, for example, if the bol is directly wired to a ReturnNode
4962 bol = flat_array_test(load_object_klass(array));
4963 break;
4964 case IsNullRestricted:
4965 bol = null_free_array_test(array);
4966 break;
4967 case IsAtomic:
4968 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4969 // Enable TestIntrinsics::test87/88 once this is implemented
4970 // bol = null_free_atomic_array_test
4971 return false;
4972 default:
4973 ShouldNotReachHere();
4974 }
4975
4976 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4977 set_result(res);
4978 return true;
4979 }
4980
4981 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4982 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4983 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4984 RegionNode* region = new RegionNode(2);
4985 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4986
4987 if (type_array_guard) {
4988 generate_typeArray_guard(klass_node, region);
4989 if (region->req() == 3) {
4990 phi->add_req(klass_node);
4991 }
4992 }
4993 Node* adr_refined_klass = basic_plus_adr(klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4994 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4995
4996 // Can be null if not initialized yet, just deopt
4997 Node* null_ctl = top();
4998 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4999
5000 region->init_req(1, control());
5001 phi->init_req(1, refined_klass);
5002
5003 set_control(_gvn.transform(region));
5004 return _gvn.transform(phi);
5005 }
5006
5007 // Load the non-refined array klass from an ObjArrayKlass.
5008 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
5009 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
5010 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
5011 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
5012 }
5013
5014 RegionNode* region = new RegionNode(2);
5015 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
5016
5017 generate_typeArray_guard(klass_node, region);
5018 if (region->req() == 3) {
5019 phi->add_req(klass_node);
5020 }
5021 Node* super_adr = basic_plus_adr(klass_node, in_bytes(Klass::super_offset()));
5022 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5023
5024 region->init_req(1, control());
5025 phi->init_req(1, super_klass);
5026
5027 set_control(_gvn.transform(region));
5028 return _gvn.transform(phi);
5029 }
5030
5031 //-----------------------inline_native_newArray--------------------------
5032 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
5033 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
5034 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
5035 Node* mirror;
5036 Node* count_val;
5037 if (uninitialized) {
5038 null_check_receiver();
5039 mirror = argument(1);
5040 count_val = argument(2);
5041 } else {
5042 mirror = argument(0);
5043 count_val = argument(1);
5044 }
5045
5046 mirror = null_check(mirror);
5047 // If mirror or obj is dead, only null-path is taken.
5048 if (stopped()) return true;
5049
5050 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
5051 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5052 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5053 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5054 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5055
5056 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
5057 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
5058 result_reg, _slow_path);
5059 Node* normal_ctl = control();
5060 Node* no_array_ctl = result_reg->in(_slow_path);
5061
5062 // Generate code for the slow case. We make a call to newArray().
5063 set_control(no_array_ctl);
5064 if (!stopped()) {
5065 // Either the input type is void.class, or else the
5066 // array klass has not yet been cached. Either the
5067 // ensuing call will throw an exception, or else it
5068 // will cache the array klass for next time.
5069 PreserveJVMState pjvms(this);
5070 CallJavaNode* slow_call = nullptr;
5071 if (uninitialized) {
5072 // Generate optimized virtual call (holder class 'Unsafe' is final)
5073 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5074 } else {
5075 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5076 }
5077 Node* slow_result = set_results_for_java_call(slow_call);
5078 // this->control() comes from set_results_for_java_call
5079 result_reg->set_req(_slow_path, control());
5080 result_val->set_req(_slow_path, slow_result);
5081 result_io ->set_req(_slow_path, i_o());
5082 result_mem->set_req(_slow_path, reset_memory());
5083 }
5084
5085 set_control(normal_ctl);
5086 if (!stopped()) {
5087 // Normal case: The array type has been cached in the java.lang.Class.
5088 // The following call works fine even if the array type is polymorphic.
5089 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5090
5091 klass_node = load_default_refined_array_klass(klass_node);
5092
5093 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5094 result_reg->init_req(_normal_path, control());
5095 result_val->init_req(_normal_path, obj);
5096 result_io ->init_req(_normal_path, i_o());
5097 result_mem->init_req(_normal_path, reset_memory());
5098
5099 if (uninitialized) {
5100 // Mark the allocation so that zeroing is skipped
5101 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5102 alloc->maybe_set_complete(&_gvn);
5103 }
5104 }
5105
5106 // Return the combined state.
5107 set_i_o( _gvn.transform(result_io) );
5108 set_all_memory( _gvn.transform(result_mem));
5109
5110 C->set_has_split_ifs(true); // Has chance for split-if optimization
5111 set_result(result_reg, result_val);
5112 return true;
5113 }
5114
5115 //----------------------inline_native_getLength--------------------------
5116 // public static native int java.lang.reflect.Array.getLength(Object array);
5117 bool LibraryCallKit::inline_native_getLength() {
5118 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5119
5120 Node* array = null_check(argument(0));
5121 // If array is dead, only null-path is taken.
5122 if (stopped()) return true;
5123
5124 // Deoptimize if it is a non-array.
5125 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5126
5127 if (non_array != nullptr) {
5128 PreserveJVMState pjvms(this);
5129 set_control(non_array);
5130 uncommon_trap(Deoptimization::Reason_intrinsic,
5131 Deoptimization::Action_maybe_recompile);
5132 }
5133
5134 // If control is dead, only non-array-path is taken.
5135 if (stopped()) return true;
5136
5137 // The works fine even if the array type is polymorphic.
5138 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5139 Node* result = load_array_length(array);
5140
5141 C->set_has_split_ifs(true); // Has chance for split-if optimization
5142 set_result(result);
5143 return true;
5144 }
5145
5146 //------------------------inline_array_copyOf----------------------------
5147 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5148 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5149 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5150 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5151
5152 // Get the arguments.
5153 Node* original = argument(0);
5154 Node* start = is_copyOfRange? argument(1): intcon(0);
5155 Node* end = is_copyOfRange? argument(2): argument(1);
5156 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5157
5158 Node* newcopy = nullptr;
5159
5160 // Set the original stack and the reexecute bit for the interpreter to reexecute
5161 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5162 { PreserveReexecuteState preexecs(this);
5163 jvms()->set_should_reexecute(true);
5164
5165 array_type_mirror = null_check(array_type_mirror);
5166 original = null_check(original);
5167
5168 // Check if a null path was taken unconditionally.
5169 if (stopped()) return true;
5170
5171 Node* orig_length = load_array_length(original);
5172
5173 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5174 klass_node = null_check(klass_node);
5175
5176 RegionNode* bailout = new RegionNode(1);
5177 record_for_igvn(bailout);
5178
5179 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5180 // Bail out if that is so.
5181 // Inline type array may have object field that would require a
5182 // write barrier. Conservatively, go to slow path.
5183 // TODO 8251971: Optimize for the case when flat src/dst are later found
5184 // to not contain oops (i.e., move this check to the macro expansion phase).
5185 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5186 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5187 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5188 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5189 // Can src array be flat and contain oops?
5190 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5191 // Can dest array be flat and contain oops?
5192 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5193 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5194
5195 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5196
5197 if (not_objArray != nullptr) {
5198 // Improve the klass node's type from the new optimistic assumption:
5199 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5200 bool not_flat = !UseArrayFlattening;
5201 bool not_null_free = !Arguments::is_valhalla_enabled();
5202 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5203 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5204 refined_klass_node = _gvn.transform(cast);
5205 }
5206
5207 // Bail out if either start or end is negative.
5208 generate_negative_guard(start, bailout, &start);
5209 generate_negative_guard(end, bailout, &end);
5210
5211 Node* length = end;
5212 if (_gvn.type(start) != TypeInt::ZERO) {
5213 length = _gvn.transform(new SubINode(end, start));
5214 }
5215
5216 // Bail out if length is negative (i.e., if start > end).
5217 // Without this the new_array would throw
5218 // NegativeArraySizeException but IllegalArgumentException is what
5219 // should be thrown
5220 generate_negative_guard(length, bailout, &length);
5221
5222 // Handle inline type arrays
5223 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5224 if (!stopped()) {
5225 // TODO 8251971
5226 if (!orig_t->is_null_free()) {
5227 // Not statically known to be null free, add a check
5228 generate_fair_guard(null_free_array_test(original), bailout);
5229 }
5230 orig_t = _gvn.type(original)->isa_aryptr();
5231 if (orig_t != nullptr && orig_t->is_flat()) {
5232 // Src is flat, check that dest is flat as well
5233 if (exclude_flat) {
5234 // Dest can't be flat, bail out
5235 bailout->add_req(control());
5236 set_control(top());
5237 } else {
5238 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5239 }
5240 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5241 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5242 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5243 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5244 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5245 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5246 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5247 if (orig_t != nullptr) {
5248 orig_t = orig_t->cast_to_not_flat();
5249 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5250 }
5251 }
5252 if (!can_validate) {
5253 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5254 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5255 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5256 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5257 generate_fair_guard(null_free_array_test(original), bailout);
5258 }
5259 }
5260
5261 // Bail out if start is larger than the original length
5262 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5263 generate_negative_guard(orig_tail, bailout, &orig_tail);
5264
5265 if (bailout->req() > 1) {
5266 PreserveJVMState pjvms(this);
5267 set_control(_gvn.transform(bailout));
5268 uncommon_trap(Deoptimization::Reason_intrinsic,
5269 Deoptimization::Action_maybe_recompile);
5270 }
5271
5272 if (!stopped()) {
5273 // How many elements will we copy from the original?
5274 // The answer is MinI(orig_tail, length).
5275 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5276
5277 // Generate a direct call to the right arraycopy function(s).
5278 // We know the copy is disjoint but we might not know if the
5279 // oop stores need checking.
5280 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5281 // This will fail a store-check if x contains any non-nulls.
5282
5283 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5284 // loads/stores but it is legal only if we're sure the
5285 // Arrays.copyOf would succeed. So we need all input arguments
5286 // to the copyOf to be validated, including that the copy to the
5287 // new array won't trigger an ArrayStoreException. That subtype
5288 // check can be optimized if we know something on the type of
5289 // the input array from type speculation.
5290 if (_gvn.type(klass_node)->singleton()) {
5291 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5292 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5293
5294 int test = C->static_subtype_check(superk, subk);
5295 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5296 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5297 if (t_original->speculative_type() != nullptr) {
5298 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5299 }
5300 }
5301 }
5302
5303 bool validated = false;
5304 // Reason_class_check rather than Reason_intrinsic because we
5305 // want to intrinsify even if this traps.
5306 if (can_validate) {
5307 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5308
5309 if (not_subtype_ctrl != top()) {
5310 PreserveJVMState pjvms(this);
5311 set_control(not_subtype_ctrl);
5312 uncommon_trap(Deoptimization::Reason_class_check,
5313 Deoptimization::Action_make_not_entrant);
5314 assert(stopped(), "Should be stopped");
5315 }
5316 validated = true;
5317 }
5318
5319 if (!stopped()) {
5320 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5321
5322 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5323 load_object_klass(original), klass_node);
5324 if (!is_copyOfRange) {
5325 ac->set_copyof(validated);
5326 } else {
5327 ac->set_copyofrange(validated);
5328 }
5329 Node* n = _gvn.transform(ac);
5330 if (n == ac) {
5331 ac->connect_outputs(this);
5332 } else {
5333 assert(validated, "shouldn't transform if all arguments not validated");
5334 set_all_memory(n);
5335 }
5336 }
5337 }
5338 } // original reexecute is set back here
5339
5340 C->set_has_split_ifs(true); // Has chance for split-if optimization
5341 if (!stopped()) {
5342 set_result(newcopy);
5343 }
5344 return true;
5345 }
5346
5347
5348 //----------------------generate_virtual_guard---------------------------
5349 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5350 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5351 RegionNode* slow_region) {
5352 ciMethod* method = callee();
5353 int vtable_index = method->vtable_index();
5354 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5355 "bad index %d", vtable_index);
5356 // Get the Method* out of the appropriate vtable entry.
5357 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5358 vtable_index*vtableEntry::size_in_bytes() +
5359 in_bytes(vtableEntry::method_offset());
5360 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5361 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5362
5363 // Compare the target method with the expected method (e.g., Object.hashCode).
5364 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5365
5366 Node* native_call = makecon(native_call_addr);
5367 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5368 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5369
5370 return generate_slow_guard(test_native, slow_region);
5371 }
5372
5373 //-----------------------generate_method_call----------------------------
5374 // Use generate_method_call to make a slow-call to the real
5375 // method if the fast path fails. An alternative would be to
5376 // use a stub like OptoRuntime::slow_arraycopy_Java.
5377 // This only works for expanding the current library call,
5378 // not another intrinsic. (E.g., don't use this for making an
5379 // arraycopy call inside of the copyOf intrinsic.)
5380 CallJavaNode*
5381 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5382 // When compiling the intrinsic method itself, do not use this technique.
5383 guarantee(callee() != C->method(), "cannot make slow-call to self");
5384
5385 ciMethod* method = callee();
5386 // ensure the JVMS we have will be correct for this call
5387 guarantee(method_id == method->intrinsic_id(), "must match");
5388
5389 const TypeFunc* tf = TypeFunc::make(method);
5390 if (res_not_null) {
5391 assert(tf->return_type() == T_OBJECT, "");
5392 const TypeTuple* range = tf->range_cc();
5393 const Type** fields = TypeTuple::fields(range->cnt());
5394 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5395 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5396 tf = TypeFunc::make(tf->domain_cc(), new_range);
5397 }
5398 CallJavaNode* slow_call;
5399 if (is_static) {
5400 assert(!is_virtual, "");
5401 slow_call = new CallStaticJavaNode(C, tf,
5402 SharedRuntime::get_resolve_static_call_stub(), method);
5403 } else if (is_virtual) {
5404 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5405 int vtable_index = Method::invalid_vtable_index;
5406 if (UseInlineCaches) {
5407 // Suppress the vtable call
5408 } else {
5409 // hashCode and clone are not a miranda methods,
5410 // so the vtable index is fixed.
5411 // No need to use the linkResolver to get it.
5412 vtable_index = method->vtable_index();
5413 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5414 "bad index %d", vtable_index);
5415 }
5416 slow_call = new CallDynamicJavaNode(tf,
5417 SharedRuntime::get_resolve_virtual_call_stub(),
5418 method, vtable_index);
5419 } else { // neither virtual nor static: opt_virtual
5420 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5421 slow_call = new CallStaticJavaNode(C, tf,
5422 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5423 slow_call->set_optimized_virtual(true);
5424 }
5425 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5426 // To be able to issue a direct call (optimized virtual or virtual)
5427 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5428 // about the method being invoked should be attached to the call site to
5429 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5430 slow_call->set_override_symbolic_info(true);
5431 }
5432 set_arguments_for_java_call(slow_call);
5433 set_edges_for_java_call(slow_call);
5434 return slow_call;
5435 }
5436
5437
5438 /**
5439 * Build special case code for calls to hashCode on an object. This call may
5440 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5441 * slightly different code.
5442 */
5443 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5444 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5445 assert(!(is_virtual && is_static), "either virtual, special, or static");
5446
5447 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5448
5449 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5450 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5451 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5452 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5453 Node* obj = argument(0);
5454
5455 // Don't intrinsify hashcode on inline types for now.
5456 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5457 if (gvn().type(obj)->is_inlinetypeptr()) {
5458 return false;
5459 }
5460
5461 if (!is_static) {
5462 // Check for hashing null object
5463 obj = null_check_receiver();
5464 if (stopped()) return true; // unconditionally null
5465 result_reg->init_req(_null_path, top());
5466 result_val->init_req(_null_path, top());
5467 } else {
5468 // Do a null check, and return zero if null.
5469 // System.identityHashCode(null) == 0
5470 Node* null_ctl = top();
5471 obj = null_check_oop(obj, &null_ctl);
5472 result_reg->init_req(_null_path, null_ctl);
5473 result_val->init_req(_null_path, _gvn.intcon(0));
5474 }
5475
5476 // Unconditionally null? Then return right away.
5477 if (stopped()) {
5478 set_control( result_reg->in(_null_path));
5479 if (!stopped())
5480 set_result(result_val->in(_null_path));
5481 return true;
5482 }
5483
5484 // We only go to the fast case code if we pass a number of guards. The
5485 // paths which do not pass are accumulated in the slow_region.
5486 RegionNode* slow_region = new RegionNode(1);
5487 record_for_igvn(slow_region);
5488
5489 // If this is a virtual call, we generate a funny guard. We pull out
5490 // the vtable entry corresponding to hashCode() from the target object.
5491 // If the target method which we are calling happens to be the native
5492 // Object hashCode() method, we pass the guard. We do not need this
5493 // guard for non-virtual calls -- the caller is known to be the native
5494 // Object hashCode().
5495 if (is_virtual) {
5496 // After null check, get the object's klass.
5497 Node* obj_klass = load_object_klass(obj);
5498 generate_virtual_guard(obj_klass, slow_region);
5499 }
5500
5501 // Get the header out of the object, use LoadMarkNode when available
5502 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5503 // The control of the load must be null. Otherwise, the load can move before
5504 // the null check after castPP removal.
5505 Node* no_ctrl = nullptr;
5506 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5507
5508 if (!UseObjectMonitorTable) {
5509 // Test the header to see if it is safe to read w.r.t. locking.
5510 // We cannot use the inline type mask as this may check bits that are overriden
5511 // by an object monitor's pointer when inflating locking.
5512 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5513 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5514 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5515 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5516 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5517
5518 generate_slow_guard(test_monitor, slow_region);
5519 }
5520
5521 // Get the hash value and check to see that it has been properly assigned.
5522 // We depend on hash_mask being at most 32 bits and avoid the use of
5523 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5524 // vm: see markWord.hpp.
5525 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5526 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5527 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5528 // This hack lets the hash bits live anywhere in the mark object now, as long
5529 // as the shift drops the relevant bits into the low 32 bits. Note that
5530 // Java spec says that HashCode is an int so there's no point in capturing
5531 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5532 hshifted_header = ConvX2I(hshifted_header);
5533 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5534
5535 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5536 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5537 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5538
5539 generate_slow_guard(test_assigned, slow_region);
5540
5541 Node* init_mem = reset_memory();
5542 // fill in the rest of the null path:
5543 result_io ->init_req(_null_path, i_o());
5544 result_mem->init_req(_null_path, init_mem);
5545
5546 result_val->init_req(_fast_path, hash_val);
5547 result_reg->init_req(_fast_path, control());
5548 result_io ->init_req(_fast_path, i_o());
5549 result_mem->init_req(_fast_path, init_mem);
5550
5551 // Generate code for the slow case. We make a call to hashCode().
5552 set_control(_gvn.transform(slow_region));
5553 if (!stopped()) {
5554 // No need for PreserveJVMState, because we're using up the present state.
5555 set_all_memory(init_mem);
5556 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5557 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5558 Node* slow_result = set_results_for_java_call(slow_call);
5559 // this->control() comes from set_results_for_java_call
5560 result_reg->init_req(_slow_path, control());
5561 result_val->init_req(_slow_path, slow_result);
5562 result_io ->set_req(_slow_path, i_o());
5563 result_mem ->set_req(_slow_path, reset_memory());
5564 }
5565
5566 // Return the combined state.
5567 set_i_o( _gvn.transform(result_io) );
5568 set_all_memory( _gvn.transform(result_mem));
5569
5570 set_result(result_reg, result_val);
5571 return true;
5572 }
5573
5574 //---------------------------inline_native_getClass----------------------------
5575 // public final native Class<?> java.lang.Object.getClass();
5576 //
5577 // Build special case code for calls to getClass on an object.
5578 bool LibraryCallKit::inline_native_getClass() {
5579 Node* obj = argument(0);
5580 if (obj->is_InlineType()) {
5581 const Type* t = _gvn.type(obj);
5582 if (t->maybe_null()) {
5583 null_check(obj);
5584 }
5585 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5586 return true;
5587 }
5588 obj = null_check_receiver();
5589 if (stopped()) return true;
5590 set_result(load_mirror_from_klass(load_object_klass(obj)));
5591 return true;
5592 }
5593
5594 //-----------------inline_native_Reflection_getCallerClass---------------------
5595 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5596 //
5597 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5598 //
5599 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5600 // in that it must skip particular security frames and checks for
5601 // caller sensitive methods.
5602 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5603 #ifndef PRODUCT
5604 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5605 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5606 }
5607 #endif
5608
5609 if (!jvms()->has_method()) {
5610 #ifndef PRODUCT
5611 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5612 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5613 }
5614 #endif
5615 return false;
5616 }
5617
5618 // Walk back up the JVM state to find the caller at the required
5619 // depth.
5620 JVMState* caller_jvms = jvms();
5621
5622 // Cf. JVM_GetCallerClass
5623 // NOTE: Start the loop at depth 1 because the current JVM state does
5624 // not include the Reflection.getCallerClass() frame.
5625 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5626 ciMethod* m = caller_jvms->method();
5627 switch (n) {
5628 case 0:
5629 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5630 break;
5631 case 1:
5632 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5633 if (!m->caller_sensitive()) {
5634 #ifndef PRODUCT
5635 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5636 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5637 }
5638 #endif
5639 return false; // bail-out; let JVM_GetCallerClass do the work
5640 }
5641 break;
5642 default:
5643 if (!m->is_ignored_by_security_stack_walk()) {
5644 // We have reached the desired frame; return the holder class.
5645 // Acquire method holder as java.lang.Class and push as constant.
5646 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5647 ciInstance* caller_mirror = caller_klass->java_mirror();
5648 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5649
5650 #ifndef PRODUCT
5651 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5652 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());
5653 tty->print_cr(" JVM state at this point:");
5654 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5655 ciMethod* m = jvms()->of_depth(i)->method();
5656 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5657 }
5658 }
5659 #endif
5660 return true;
5661 }
5662 break;
5663 }
5664 }
5665
5666 #ifndef PRODUCT
5667 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5668 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5669 tty->print_cr(" JVM state at this point:");
5670 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5671 ciMethod* m = jvms()->of_depth(i)->method();
5672 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5673 }
5674 }
5675 #endif
5676
5677 return false; // bail-out; let JVM_GetCallerClass do the work
5678 }
5679
5680 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5681 Node* arg = argument(0);
5682 Node* result = nullptr;
5683
5684 switch (id) {
5685 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5686 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5687 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5688 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5689 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5690 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5691
5692 case vmIntrinsics::_doubleToLongBits: {
5693 // two paths (plus control) merge in a wood
5694 RegionNode *r = new RegionNode(3);
5695 Node *phi = new PhiNode(r, TypeLong::LONG);
5696
5697 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5698 // Build the boolean node
5699 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5700
5701 // Branch either way.
5702 // NaN case is less traveled, which makes all the difference.
5703 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5704 Node *opt_isnan = _gvn.transform(ifisnan);
5705 assert( opt_isnan->is_If(), "Expect an IfNode");
5706 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5707 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5708
5709 set_control(iftrue);
5710
5711 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5712 Node *slow_result = longcon(nan_bits); // return NaN
5713 phi->init_req(1, _gvn.transform( slow_result ));
5714 r->init_req(1, iftrue);
5715
5716 // Else fall through
5717 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5718 set_control(iffalse);
5719
5720 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5721 r->init_req(2, iffalse);
5722
5723 // Post merge
5724 set_control(_gvn.transform(r));
5725 record_for_igvn(r);
5726
5727 C->set_has_split_ifs(true); // Has chance for split-if optimization
5728 result = phi;
5729 assert(result->bottom_type()->isa_long(), "must be");
5730 break;
5731 }
5732
5733 case vmIntrinsics::_floatToIntBits: {
5734 // two paths (plus control) merge in a wood
5735 RegionNode *r = new RegionNode(3);
5736 Node *phi = new PhiNode(r, TypeInt::INT);
5737
5738 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5739 // Build the boolean node
5740 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5741
5742 // Branch either way.
5743 // NaN case is less traveled, which makes all the difference.
5744 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5745 Node *opt_isnan = _gvn.transform(ifisnan);
5746 assert( opt_isnan->is_If(), "Expect an IfNode");
5747 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5748 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5749
5750 set_control(iftrue);
5751
5752 static const jint nan_bits = 0x7fc00000;
5753 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5754 phi->init_req(1, _gvn.transform( slow_result ));
5755 r->init_req(1, iftrue);
5756
5757 // Else fall through
5758 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5759 set_control(iffalse);
5760
5761 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5762 r->init_req(2, iffalse);
5763
5764 // Post merge
5765 set_control(_gvn.transform(r));
5766 record_for_igvn(r);
5767
5768 C->set_has_split_ifs(true); // Has chance for split-if optimization
5769 result = phi;
5770 assert(result->bottom_type()->isa_int(), "must be");
5771 break;
5772 }
5773
5774 default:
5775 fatal_unexpected_iid(id);
5776 break;
5777 }
5778 set_result(_gvn.transform(result));
5779 return true;
5780 }
5781
5782 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5783 Node* arg = argument(0);
5784 Node* result = nullptr;
5785
5786 switch (id) {
5787 case vmIntrinsics::_floatIsInfinite:
5788 result = new IsInfiniteFNode(arg);
5789 break;
5790 case vmIntrinsics::_floatIsFinite:
5791 result = new IsFiniteFNode(arg);
5792 break;
5793 case vmIntrinsics::_doubleIsInfinite:
5794 result = new IsInfiniteDNode(arg);
5795 break;
5796 case vmIntrinsics::_doubleIsFinite:
5797 result = new IsFiniteDNode(arg);
5798 break;
5799 default:
5800 fatal_unexpected_iid(id);
5801 break;
5802 }
5803 set_result(_gvn.transform(result));
5804 return true;
5805 }
5806
5807 //----------------------inline_unsafe_copyMemory-------------------------
5808 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5809
5810 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5811 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5812 const Type* base_t = gvn.type(base);
5813
5814 bool in_native = (base_t == TypePtr::NULL_PTR);
5815 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5816 bool is_mixed = !in_heap && !in_native;
5817
5818 if (is_mixed) {
5819 return true; // mixed accesses can touch both on-heap and off-heap memory
5820 }
5821 if (in_heap) {
5822 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5823 if (!is_prim_array) {
5824 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5825 // there's not enough type information available to determine proper memory slice for it.
5826 return true;
5827 }
5828 }
5829 return false;
5830 }
5831
5832 bool LibraryCallKit::inline_unsafe_copyMemory() {
5833 if (callee()->is_static()) return false; // caller must have the capability!
5834 null_check_receiver(); // null-check receiver
5835 if (stopped()) return true;
5836
5837 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5838
5839 Node* src_base = argument(1); // type: oop
5840 Node* src_off = ConvL2X(argument(2)); // type: long
5841 Node* dst_base = argument(4); // type: oop
5842 Node* dst_off = ConvL2X(argument(5)); // type: long
5843 Node* size = ConvL2X(argument(7)); // type: long
5844
5845 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5846 "fieldOffset must be byte-scaled");
5847
5848 Node* src_addr = make_unsafe_address(src_base, src_off);
5849 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5850
5851 Node* thread = _gvn.transform(new ThreadLocalNode());
5852 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5853 BasicType doing_unsafe_access_bt = T_BYTE;
5854 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5855
5856 // update volatile field
5857 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5858
5859 int flags = RC_LEAF | RC_NO_FP;
5860
5861 const TypePtr* dst_type = TypePtr::BOTTOM;
5862
5863 // Adjust memory effects of the runtime call based on input values.
5864 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5865 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5866 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5867
5868 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5869 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5870 flags |= RC_NARROW_MEM; // narrow in memory
5871 }
5872 }
5873
5874 // Call it. Note that the length argument is not scaled.
5875 make_runtime_call(flags,
5876 OptoRuntime::fast_arraycopy_Type(),
5877 StubRoutines::unsafe_arraycopy(),
5878 "unsafe_arraycopy",
5879 dst_type,
5880 src_addr, dst_addr, size XTOP);
5881
5882 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5883
5884 return true;
5885 }
5886
5887 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5888 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5889 bool LibraryCallKit::inline_unsafe_setMemory() {
5890 if (callee()->is_static()) return false; // caller must have the capability!
5891 null_check_receiver(); // null-check receiver
5892 if (stopped()) return true;
5893
5894 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5895
5896 Node* dst_base = argument(1); // type: oop
5897 Node* dst_off = ConvL2X(argument(2)); // type: long
5898 Node* size = ConvL2X(argument(4)); // type: long
5899 Node* byte = argument(6); // type: byte
5900
5901 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5902 "fieldOffset must be byte-scaled");
5903
5904 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5905
5906 Node* thread = _gvn.transform(new ThreadLocalNode());
5907 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5908 BasicType doing_unsafe_access_bt = T_BYTE;
5909 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5910
5911 // update volatile field
5912 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5913
5914 int flags = RC_LEAF | RC_NO_FP;
5915
5916 const TypePtr* dst_type = TypePtr::BOTTOM;
5917
5918 // Adjust memory effects of the runtime call based on input values.
5919 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5920 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5921
5922 flags |= RC_NARROW_MEM; // narrow in memory
5923 }
5924
5925 // Call it. Note that the length argument is not scaled.
5926 make_runtime_call(flags,
5927 OptoRuntime::unsafe_setmemory_Type(),
5928 StubRoutines::unsafe_setmemory(),
5929 "unsafe_setmemory",
5930 dst_type,
5931 dst_addr, size XTOP, byte);
5932
5933 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5934
5935 return true;
5936 }
5937
5938 #undef XTOP
5939
5940 //------------------------clone_coping-----------------------------------
5941 // Helper function for inline_native_clone.
5942 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5943 assert(obj_size != nullptr, "");
5944 Node* raw_obj = alloc_obj->in(1);
5945 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5946
5947 AllocateNode* alloc = nullptr;
5948 if (ReduceBulkZeroing &&
5949 // If we are implementing an array clone without knowing its source type
5950 // (can happen when compiling the array-guarded branch of a reflective
5951 // Object.clone() invocation), initialize the array within the allocation.
5952 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5953 // to a runtime clone call that assumes fully initialized source arrays.
5954 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5955 // We will be completely responsible for initializing this object -
5956 // mark Initialize node as complete.
5957 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5958 // The object was just allocated - there should be no any stores!
5959 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5960 // Mark as complete_with_arraycopy so that on AllocateNode
5961 // expansion, we know this AllocateNode is initialized by an array
5962 // copy and a StoreStore barrier exists after the array copy.
5963 alloc->initialization()->set_complete_with_arraycopy();
5964 }
5965
5966 Node* size = _gvn.transform(obj_size);
5967 access_clone(obj, alloc_obj, size, is_array);
5968
5969 // Do not let reads from the cloned object float above the arraycopy.
5970 if (alloc != nullptr) {
5971 // Do not let stores that initialize this object be reordered with
5972 // a subsequent store that would make this object accessible by
5973 // other threads.
5974 // Record what AllocateNode this StoreStore protects so that
5975 // escape analysis can go from the MemBarStoreStoreNode to the
5976 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5977 // based on the escape status of the AllocateNode.
5978 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5979 } else {
5980 insert_mem_bar(Op_MemBarCPUOrder);
5981 }
5982 }
5983
5984 //------------------------inline_native_clone----------------------------
5985 // protected native Object java.lang.Object.clone();
5986 //
5987 // Here are the simple edge cases:
5988 // null receiver => normal trap
5989 // virtual and clone was overridden => slow path to out-of-line clone
5990 // not cloneable or finalizer => slow path to out-of-line Object.clone
5991 //
5992 // The general case has two steps, allocation and copying.
5993 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5994 //
5995 // Copying also has two cases, oop arrays and everything else.
5996 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5997 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5998 //
5999 // These steps fold up nicely if and when the cloned object's klass
6000 // can be sharply typed as an object array, a type array, or an instance.
6001 //
6002 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
6003 PhiNode* result_val;
6004
6005 // Set the reexecute bit for the interpreter to reexecute
6006 // the bytecode that invokes Object.clone if deoptimization happens.
6007 { PreserveReexecuteState preexecs(this);
6008 jvms()->set_should_reexecute(true);
6009
6010 Node* obj = argument(0);
6011 obj = null_check_receiver();
6012 if (stopped()) return true;
6013
6014 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
6015 if (obj_type->is_inlinetypeptr()) {
6016 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
6017 // no identity.
6018 set_result(obj);
6019 return true;
6020 }
6021
6022 // If we are going to clone an instance, we need its exact type to
6023 // know the number and types of fields to convert the clone to
6024 // loads/stores. Maybe a speculative type can help us.
6025 if (!obj_type->klass_is_exact() &&
6026 obj_type->speculative_type() != nullptr &&
6027 obj_type->speculative_type()->is_instance_klass() &&
6028 !obj_type->speculative_type()->is_inlinetype()) {
6029 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
6030 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
6031 !spec_ik->has_injected_fields()) {
6032 if (!obj_type->isa_instptr() ||
6033 obj_type->is_instptr()->instance_klass()->has_subklass()) {
6034 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
6035 }
6036 }
6037 }
6038
6039 // Conservatively insert a memory barrier on all memory slices.
6040 // Do not let writes into the original float below the clone.
6041 insert_mem_bar(Op_MemBarCPUOrder);
6042
6043 // paths into result_reg:
6044 enum {
6045 _slow_path = 1, // out-of-line call to clone method (virtual or not)
6046 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
6047 _array_path, // plain array allocation, plus arrayof_long_arraycopy
6048 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
6049 PATH_LIMIT
6050 };
6051 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
6052 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
6053 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
6054 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
6055 record_for_igvn(result_reg);
6056
6057 Node* obj_klass = load_object_klass(obj);
6058 // We only go to the fast case code if we pass a number of guards.
6059 // The paths which do not pass are accumulated in the slow_region.
6060 RegionNode* slow_region = new RegionNode(1);
6061 record_for_igvn(slow_region);
6062
6063 Node* array_obj = obj;
6064 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
6065 if (array_ctl != nullptr) {
6066 // It's an array.
6067 PreserveJVMState pjvms(this);
6068 set_control(array_ctl);
6069
6070 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6071 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6072 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6073 obj_type->can_be_inline_array() &&
6074 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6075 // Flat inline type array may have object field that would require a
6076 // write barrier. Conservatively, go to slow path.
6077 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6078 }
6079
6080 if (!stopped()) {
6081 Node* obj_length = load_array_length(array_obj);
6082 Node* array_size = nullptr; // Size of the array without object alignment padding.
6083 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6084
6085 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6086 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6087 // If it is an oop array, it requires very special treatment,
6088 // because gc barriers are required when accessing the array.
6089 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6090 if (is_obja != nullptr) {
6091 PreserveJVMState pjvms2(this);
6092 set_control(is_obja);
6093 // Generate a direct call to the right arraycopy function(s).
6094 // Clones are always tightly coupled.
6095 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6096 ac->set_clone_oop_array();
6097 Node* n = _gvn.transform(ac);
6098 assert(n == ac, "cannot disappear");
6099 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6100
6101 result_reg->init_req(_objArray_path, control());
6102 result_val->init_req(_objArray_path, alloc_obj);
6103 result_i_o ->set_req(_objArray_path, i_o());
6104 result_mem ->set_req(_objArray_path, reset_memory());
6105 }
6106 }
6107 // Otherwise, there are no barriers to worry about.
6108 // (We can dispense with card marks if we know the allocation
6109 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6110 // causes the non-eden paths to take compensating steps to
6111 // simulate a fresh allocation, so that no further
6112 // card marks are required in compiled code to initialize
6113 // the object.)
6114
6115 if (!stopped()) {
6116 copy_to_clone(obj, alloc_obj, array_size, true);
6117
6118 // Present the results of the copy.
6119 result_reg->init_req(_array_path, control());
6120 result_val->init_req(_array_path, alloc_obj);
6121 result_i_o ->set_req(_array_path, i_o());
6122 result_mem ->set_req(_array_path, reset_memory());
6123 }
6124 }
6125 }
6126
6127 if (!stopped()) {
6128 // It's an instance (we did array above). Make the slow-path tests.
6129 // If this is a virtual call, we generate a funny guard. We grab
6130 // the vtable entry corresponding to clone() from the target object.
6131 // If the target method which we are calling happens to be the
6132 // Object clone() method, we pass the guard. We do not need this
6133 // guard for non-virtual calls; the caller is known to be the native
6134 // Object clone().
6135 if (is_virtual) {
6136 generate_virtual_guard(obj_klass, slow_region);
6137 }
6138
6139 // The object must be easily cloneable and must not have a finalizer.
6140 // Both of these conditions may be checked in a single test.
6141 // We could optimize the test further, but we don't care.
6142 generate_misc_flags_guard(obj_klass,
6143 // Test both conditions:
6144 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6145 // Must be cloneable but not finalizer:
6146 KlassFlags::_misc_is_cloneable_fast,
6147 slow_region);
6148 }
6149
6150 if (!stopped()) {
6151 // It's an instance, and it passed the slow-path tests.
6152 PreserveJVMState pjvms(this);
6153 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6154 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6155 // is reexecuted if deoptimization occurs and there could be problems when merging
6156 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6157 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6158
6159 copy_to_clone(obj, alloc_obj, obj_size, false);
6160
6161 // Present the results of the slow call.
6162 result_reg->init_req(_instance_path, control());
6163 result_val->init_req(_instance_path, alloc_obj);
6164 result_i_o ->set_req(_instance_path, i_o());
6165 result_mem ->set_req(_instance_path, reset_memory());
6166 }
6167
6168 // Generate code for the slow case. We make a call to clone().
6169 set_control(_gvn.transform(slow_region));
6170 if (!stopped()) {
6171 PreserveJVMState pjvms(this);
6172 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6173 // We need to deoptimize on exception (see comment above)
6174 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6175 // this->control() comes from set_results_for_java_call
6176 result_reg->init_req(_slow_path, control());
6177 result_val->init_req(_slow_path, slow_result);
6178 result_i_o ->set_req(_slow_path, i_o());
6179 result_mem ->set_req(_slow_path, reset_memory());
6180 }
6181
6182 // Return the combined state.
6183 set_control( _gvn.transform(result_reg));
6184 set_i_o( _gvn.transform(result_i_o));
6185 set_all_memory( _gvn.transform(result_mem));
6186 } // original reexecute is set back here
6187
6188 set_result(_gvn.transform(result_val));
6189 return true;
6190 }
6191
6192 // If we have a tightly coupled allocation, the arraycopy may take care
6193 // of the array initialization. If one of the guards we insert between
6194 // the allocation and the arraycopy causes a deoptimization, an
6195 // uninitialized array will escape the compiled method. To prevent that
6196 // we set the JVM state for uncommon traps between the allocation and
6197 // the arraycopy to the state before the allocation so, in case of
6198 // deoptimization, we'll reexecute the allocation and the
6199 // initialization.
6200 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6201 if (alloc != nullptr) {
6202 ciMethod* trap_method = alloc->jvms()->method();
6203 int trap_bci = alloc->jvms()->bci();
6204
6205 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6206 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6207 // Make sure there's no store between the allocation and the
6208 // arraycopy otherwise visible side effects could be rexecuted
6209 // in case of deoptimization and cause incorrect execution.
6210 bool no_interfering_store = true;
6211 Node* mem = alloc->in(TypeFunc::Memory);
6212 if (mem->is_MergeMem()) {
6213 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6214 Node* n = mms.memory();
6215 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6216 assert(n->is_Store(), "what else?");
6217 no_interfering_store = false;
6218 break;
6219 }
6220 }
6221 } else {
6222 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6223 Node* n = mms.memory();
6224 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6225 assert(n->is_Store(), "what else?");
6226 no_interfering_store = false;
6227 break;
6228 }
6229 }
6230 }
6231
6232 if (no_interfering_store) {
6233 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6234
6235 JVMState* saved_jvms = jvms();
6236 saved_reexecute_sp = _reexecute_sp;
6237
6238 set_jvms(sfpt->jvms());
6239 _reexecute_sp = jvms()->sp();
6240
6241 return saved_jvms;
6242 }
6243 }
6244 }
6245 return nullptr;
6246 }
6247
6248 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6249 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6250 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6251 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6252 uint size = alloc->req();
6253 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6254 old_jvms->set_map(sfpt);
6255 for (uint i = 0; i < size; i++) {
6256 sfpt->init_req(i, alloc->in(i));
6257 }
6258 int adjustment = 1;
6259 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6260 if (ary_klass_ptr->is_null_free()) {
6261 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6262 // also requires the componentType and initVal on stack for re-execution.
6263 // Re-create and push the componentType.
6264 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6265 ciInstance* instance = klass->component_mirror_instance();
6266 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6267 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6268 adjustment++;
6269 }
6270 // re-push array length for deoptimization
6271 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6272 if (ary_klass_ptr->is_null_free()) {
6273 // Re-create and push the initVal.
6274 Node* init_val = alloc->in(AllocateNode::InitValue);
6275 if (init_val == nullptr) {
6276 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6277 } else if (UseCompressedOops) {
6278 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6279 }
6280 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6281 adjustment++;
6282 }
6283 old_jvms->set_sp(old_jvms->sp() + adjustment);
6284 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6285 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6286 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6287 old_jvms->set_should_reexecute(true);
6288
6289 sfpt->set_i_o(map()->i_o());
6290 sfpt->set_memory(map()->memory());
6291 sfpt->set_control(map()->control());
6292 return sfpt;
6293 }
6294
6295 // In case of a deoptimization, we restart execution at the
6296 // allocation, allocating a new array. We would leave an uninitialized
6297 // array in the heap that GCs wouldn't expect. Move the allocation
6298 // after the traps so we don't allocate the array if we
6299 // deoptimize. This is possible because tightly_coupled_allocation()
6300 // guarantees there's no observer of the allocated array at this point
6301 // and the control flow is simple enough.
6302 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6303 int saved_reexecute_sp, uint new_idx) {
6304 if (saved_jvms_before_guards != nullptr && !stopped()) {
6305 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6306
6307 assert(alloc != nullptr, "only with a tightly coupled allocation");
6308 // restore JVM state to the state at the arraycopy
6309 saved_jvms_before_guards->map()->set_control(map()->control());
6310 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6311 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6312 // If we've improved the types of some nodes (null check) while
6313 // emitting the guards, propagate them to the current state
6314 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6315 set_jvms(saved_jvms_before_guards);
6316 _reexecute_sp = saved_reexecute_sp;
6317
6318 // Remove the allocation from above the guards
6319 CallProjections* callprojs = alloc->extract_projections(true);
6320 InitializeNode* init = alloc->initialization();
6321 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6322 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6323 init->replace_mem_projs_by(alloc_mem, C);
6324
6325 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6326 // the allocation (i.e. is only valid if the allocation succeeds):
6327 // 1) replace CastIINode with AllocateArrayNode's length here
6328 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6329 //
6330 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6331 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6332 Node* init_control = init->proj_out(TypeFunc::Control);
6333 Node* alloc_length = alloc->Ideal_length();
6334 #ifdef ASSERT
6335 Node* prev_cast = nullptr;
6336 #endif
6337 for (uint i = 0; i < init_control->outcnt(); i++) {
6338 Node* init_out = init_control->raw_out(i);
6339 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6340 #ifdef ASSERT
6341 if (prev_cast == nullptr) {
6342 prev_cast = init_out;
6343 } else {
6344 if (prev_cast->cmp(*init_out) == false) {
6345 prev_cast->dump();
6346 init_out->dump();
6347 assert(false, "not equal CastIINode");
6348 }
6349 }
6350 #endif
6351 C->gvn_replace_by(init_out, alloc_length);
6352 }
6353 }
6354 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6355
6356 // move the allocation here (after the guards)
6357 _gvn.hash_delete(alloc);
6358 alloc->set_req(TypeFunc::Control, control());
6359 alloc->set_req(TypeFunc::I_O, i_o());
6360 Node *mem = reset_memory();
6361 set_all_memory(mem);
6362 alloc->set_req(TypeFunc::Memory, mem);
6363 set_control(init->proj_out_or_null(TypeFunc::Control));
6364 set_i_o(callprojs->fallthrough_ioproj);
6365
6366 // Update memory as done in GraphKit::set_output_for_allocation()
6367 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6368 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6369 if (ary_type->isa_aryptr() && length_type != nullptr) {
6370 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6371 }
6372 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6373 int elemidx = C->get_alias_index(telemref);
6374 // Need to properly move every memory projection for the Initialize
6375 #ifdef ASSERT
6376 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6377 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6378 #endif
6379 auto move_proj = [&](ProjNode* proj) {
6380 int alias_idx = C->get_alias_index(proj->adr_type());
6381 assert(alias_idx == Compile::AliasIdxRaw ||
6382 alias_idx == elemidx ||
6383 alias_idx == mark_idx ||
6384 alias_idx == klass_idx, "should be raw memory or array element type");
6385 set_memory(proj, alias_idx);
6386 };
6387 init->for_each_proj(move_proj, TypeFunc::Memory);
6388
6389 Node* allocx = _gvn.transform(alloc);
6390 assert(allocx == alloc, "where has the allocation gone?");
6391 assert(dest->is_CheckCastPP(), "not an allocation result?");
6392
6393 _gvn.hash_delete(dest);
6394 dest->set_req(0, control());
6395 Node* destx = _gvn.transform(dest);
6396 assert(destx == dest, "where has the allocation result gone?");
6397
6398 array_ideal_length(alloc, ary_type, true);
6399 }
6400 }
6401
6402 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6403 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6404 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6405 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6406 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6407 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6408 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6409 JVMState* saved_jvms_before_guards) {
6410 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6411 // There is at least one unrelated uncommon trap which needs to be replaced.
6412 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6413
6414 JVMState* saved_jvms = jvms();
6415 const int saved_reexecute_sp = _reexecute_sp;
6416 set_jvms(sfpt->jvms());
6417 _reexecute_sp = jvms()->sp();
6418
6419 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6420
6421 // Restore state
6422 set_jvms(saved_jvms);
6423 _reexecute_sp = saved_reexecute_sp;
6424 }
6425 }
6426
6427 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6428 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6429 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6430 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6431 while (if_proj->is_IfProj()) {
6432 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6433 if (uncommon_trap != nullptr) {
6434 create_new_uncommon_trap(uncommon_trap);
6435 }
6436 assert(if_proj->in(0)->is_If(), "must be If");
6437 if_proj = if_proj->in(0)->in(0);
6438 }
6439 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6440 "must have reached control projection of init node");
6441 }
6442
6443 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6444 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6445 assert(trap_request != 0, "no valid UCT trap request");
6446 PreserveJVMState pjvms(this);
6447 set_control(uncommon_trap_call->in(0));
6448 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6449 Deoptimization::trap_request_action(trap_request));
6450 assert(stopped(), "Should be stopped");
6451 _gvn.hash_delete(uncommon_trap_call);
6452 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6453 }
6454
6455 // Common checks for array sorting intrinsics arguments.
6456 // Returns `true` if checks passed.
6457 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6458 // check address of the class
6459 if (elementType == nullptr || elementType->is_top()) {
6460 return false; // dead path
6461 }
6462 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6463 if (elem_klass == nullptr) {
6464 return false; // dead path
6465 }
6466 // java_mirror_type() returns non-null for compile-time Class constants only
6467 ciType* elem_type = elem_klass->java_mirror_type();
6468 if (elem_type == nullptr) {
6469 return false;
6470 }
6471 bt = elem_type->basic_type();
6472 // Disable the intrinsic if the CPU does not support SIMD sort
6473 if (!Matcher::supports_simd_sort(bt)) {
6474 return false;
6475 }
6476 // check address of the array
6477 if (obj == nullptr || obj->is_top()) {
6478 return false; // dead path
6479 }
6480 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6481 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6482 return false; // failed input validation
6483 }
6484 return true;
6485 }
6486
6487 //------------------------------inline_array_partition-----------------------
6488 bool LibraryCallKit::inline_array_partition() {
6489 address stubAddr = StubRoutines::select_array_partition_function();
6490 if (stubAddr == nullptr) {
6491 return false; // Intrinsic's stub is not implemented on this platform
6492 }
6493 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6494
6495 // no receiver because it is a static method
6496 Node* elementType = argument(0);
6497 Node* obj = argument(1);
6498 Node* offset = argument(2); // long
6499 Node* fromIndex = argument(4);
6500 Node* toIndex = argument(5);
6501 Node* indexPivot1 = argument(6);
6502 Node* indexPivot2 = argument(7);
6503 // PartitionOperation: argument(8) is ignored
6504
6505 Node* pivotIndices = nullptr;
6506 BasicType bt = T_ILLEGAL;
6507
6508 if (!check_array_sort_arguments(elementType, obj, bt)) {
6509 return false;
6510 }
6511 null_check(obj);
6512 // If obj is dead, only null-path is taken.
6513 if (stopped()) {
6514 return true;
6515 }
6516 // Set the original stack and the reexecute bit for the interpreter to reexecute
6517 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6518 { PreserveReexecuteState preexecs(this);
6519 jvms()->set_should_reexecute(true);
6520
6521 Node* obj_adr = make_unsafe_address(obj, offset);
6522
6523 // create the pivotIndices array of type int and size = 2
6524 Node* size = intcon(2);
6525 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6526 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6527 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6528 guarantee(alloc != nullptr, "created above");
6529 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6530
6531 // pass the basic type enum to the stub
6532 Node* elemType = intcon(bt);
6533
6534 // Call the stub
6535 const char *stubName = "array_partition_stub";
6536 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6537 stubAddr, stubName, TypePtr::BOTTOM,
6538 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6539 indexPivot1, indexPivot2);
6540
6541 } // original reexecute is set back here
6542
6543 if (!stopped()) {
6544 set_result(pivotIndices);
6545 }
6546
6547 return true;
6548 }
6549
6550
6551 //------------------------------inline_array_sort-----------------------
6552 bool LibraryCallKit::inline_array_sort() {
6553 address stubAddr = StubRoutines::select_arraysort_function();
6554 if (stubAddr == nullptr) {
6555 return false; // Intrinsic's stub is not implemented on this platform
6556 }
6557 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6558
6559 // no receiver because it is a static method
6560 Node* elementType = argument(0);
6561 Node* obj = argument(1);
6562 Node* offset = argument(2); // long
6563 Node* fromIndex = argument(4);
6564 Node* toIndex = argument(5);
6565 // SortOperation: argument(6) is ignored
6566
6567 BasicType bt = T_ILLEGAL;
6568
6569 if (!check_array_sort_arguments(elementType, obj, bt)) {
6570 return false;
6571 }
6572 null_check(obj);
6573 // If obj is dead, only null-path is taken.
6574 if (stopped()) {
6575 return true;
6576 }
6577 Node* obj_adr = make_unsafe_address(obj, offset);
6578
6579 // pass the basic type enum to the stub
6580 Node* elemType = intcon(bt);
6581
6582 // Call the stub.
6583 const char *stubName = "arraysort_stub";
6584 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6585 stubAddr, stubName, TypePtr::BOTTOM,
6586 obj_adr, elemType, fromIndex, toIndex);
6587
6588 return true;
6589 }
6590
6591
6592 //------------------------------inline_arraycopy-----------------------
6593 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6594 // Object dest, int destPos,
6595 // int length);
6596 bool LibraryCallKit::inline_arraycopy() {
6597 // Get the arguments.
6598 Node* src = argument(0); // type: oop
6599 Node* src_offset = argument(1); // type: int
6600 Node* dest = argument(2); // type: oop
6601 Node* dest_offset = argument(3); // type: int
6602 Node* length = argument(4); // type: int
6603
6604 uint new_idx = C->unique();
6605
6606 // Check for allocation before we add nodes that would confuse
6607 // tightly_coupled_allocation()
6608 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6609
6610 int saved_reexecute_sp = -1;
6611 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6612 // See arraycopy_restore_alloc_state() comment
6613 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6614 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6615 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6616 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6617
6618 // The following tests must be performed
6619 // (1) src and dest are arrays.
6620 // (2) src and dest arrays must have elements of the same BasicType
6621 // (3) src and dest must not be null.
6622 // (4) src_offset must not be negative.
6623 // (5) dest_offset must not be negative.
6624 // (6) length must not be negative.
6625 // (7) src_offset + length must not exceed length of src.
6626 // (8) dest_offset + length must not exceed length of dest.
6627 // (9) each element of an oop array must be assignable
6628
6629 // (3) src and dest must not be null.
6630 // always do this here because we need the JVM state for uncommon traps
6631 Node* null_ctl = top();
6632 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6633 assert(null_ctl->is_top(), "no null control here");
6634 dest = null_check(dest, T_ARRAY);
6635
6636 if (!can_emit_guards) {
6637 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6638 // guards but the arraycopy node could still take advantage of a
6639 // tightly allocated allocation. tightly_coupled_allocation() is
6640 // called again to make sure it takes the null check above into
6641 // account: the null check is mandatory and if it caused an
6642 // uncommon trap to be emitted then the allocation can't be
6643 // considered tightly coupled in this context.
6644 alloc = tightly_coupled_allocation(dest);
6645 }
6646
6647 bool validated = false;
6648
6649 const Type* src_type = _gvn.type(src);
6650 const Type* dest_type = _gvn.type(dest);
6651 const TypeAryPtr* top_src = src_type->isa_aryptr();
6652 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6653
6654 // Do we have the type of src?
6655 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6656 // Do we have the type of dest?
6657 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6658 // Is the type for src from speculation?
6659 bool src_spec = false;
6660 // Is the type for dest from speculation?
6661 bool dest_spec = false;
6662
6663 if ((!has_src || !has_dest) && can_emit_guards) {
6664 // We don't have sufficient type information, let's see if
6665 // speculative types can help. We need to have types for both src
6666 // and dest so that it pays off.
6667
6668 // Do we already have or could we have type information for src
6669 bool could_have_src = has_src;
6670 // Do we already have or could we have type information for dest
6671 bool could_have_dest = has_dest;
6672
6673 ciKlass* src_k = nullptr;
6674 if (!has_src) {
6675 src_k = src_type->speculative_type_not_null();
6676 if (src_k != nullptr && src_k->is_array_klass()) {
6677 could_have_src = true;
6678 }
6679 }
6680
6681 ciKlass* dest_k = nullptr;
6682 if (!has_dest) {
6683 dest_k = dest_type->speculative_type_not_null();
6684 if (dest_k != nullptr && dest_k->is_array_klass()) {
6685 could_have_dest = true;
6686 }
6687 }
6688
6689 if (could_have_src && could_have_dest) {
6690 // This is going to pay off so emit the required guards
6691 if (!has_src) {
6692 src = maybe_cast_profiled_obj(src, src_k, true);
6693 src_type = _gvn.type(src);
6694 top_src = src_type->isa_aryptr();
6695 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6696 src_spec = true;
6697 }
6698 if (!has_dest) {
6699 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6700 dest_type = _gvn.type(dest);
6701 top_dest = dest_type->isa_aryptr();
6702 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6703 dest_spec = true;
6704 }
6705 }
6706 }
6707
6708 if (has_src && has_dest && can_emit_guards) {
6709 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6710 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6711 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6712 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6713
6714 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6715 // If both arrays are object arrays then having the exact types
6716 // for both will remove the need for a subtype check at runtime
6717 // before the call and may make it possible to pick a faster copy
6718 // routine (without a subtype check on every element)
6719 // Do we have the exact type of src?
6720 bool could_have_src = src_spec;
6721 // Do we have the exact type of dest?
6722 bool could_have_dest = dest_spec;
6723 ciKlass* src_k = nullptr;
6724 ciKlass* dest_k = nullptr;
6725 if (!src_spec) {
6726 src_k = src_type->speculative_type_not_null();
6727 if (src_k != nullptr && src_k->is_array_klass()) {
6728 could_have_src = true;
6729 }
6730 }
6731 if (!dest_spec) {
6732 dest_k = dest_type->speculative_type_not_null();
6733 if (dest_k != nullptr && dest_k->is_array_klass()) {
6734 could_have_dest = true;
6735 }
6736 }
6737 if (could_have_src && could_have_dest) {
6738 // If we can have both exact types, emit the missing guards
6739 if (could_have_src && !src_spec) {
6740 src = maybe_cast_profiled_obj(src, src_k, true);
6741 src_type = _gvn.type(src);
6742 top_src = src_type->isa_aryptr();
6743 }
6744 if (could_have_dest && !dest_spec) {
6745 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6746 dest_type = _gvn.type(dest);
6747 top_dest = dest_type->isa_aryptr();
6748 }
6749 }
6750 }
6751 }
6752
6753 ciMethod* trap_method = method();
6754 int trap_bci = bci();
6755 if (saved_jvms_before_guards != nullptr) {
6756 trap_method = alloc->jvms()->method();
6757 trap_bci = alloc->jvms()->bci();
6758 }
6759
6760 bool negative_length_guard_generated = false;
6761
6762 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6763 can_emit_guards && !src->is_top() && !dest->is_top()) {
6764 // validate arguments: enables transformation the ArrayCopyNode
6765 validated = true;
6766
6767 RegionNode* slow_region = new RegionNode(1);
6768 record_for_igvn(slow_region);
6769
6770 // (1) src and dest are arrays.
6771 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6772 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6773
6774 // (2) src and dest arrays must have elements of the same BasicType
6775 // done at macro expansion or at Ideal transformation time
6776
6777 // (4) src_offset must not be negative.
6778 generate_negative_guard(src_offset, slow_region);
6779
6780 // (5) dest_offset must not be negative.
6781 generate_negative_guard(dest_offset, slow_region);
6782
6783 // (7) src_offset + length must not exceed length of src.
6784 generate_limit_guard(src_offset, length,
6785 load_array_length(src),
6786 slow_region);
6787
6788 // (8) dest_offset + length must not exceed length of dest.
6789 generate_limit_guard(dest_offset, length,
6790 load_array_length(dest),
6791 slow_region);
6792
6793 // (6) length must not be negative.
6794 // This is also checked in generate_arraycopy() during macro expansion, but
6795 // we also have to check it here for the case where the ArrayCopyNode will
6796 // be eliminated by Escape Analysis.
6797 if (EliminateAllocations) {
6798 generate_negative_guard(length, slow_region);
6799 negative_length_guard_generated = true;
6800 }
6801
6802 // (9) each element of an oop array must be assignable
6803 Node* dest_klass = load_object_klass(dest);
6804 Node* refined_dest_klass = dest_klass;
6805 if (src != dest) {
6806 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6807 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6808 slow_region->add_req(not_subtype_ctrl);
6809 }
6810
6811 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6812 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6813 Node* src_klass = load_object_klass(src);
6814 Node* adr_prop_src = basic_plus_adr(src_klass, in_bytes(ArrayKlass::properties_offset()));
6815 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6816 Node* adr_prop_dest = basic_plus_adr(refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6817 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6818
6819 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6820 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6821 prop_src = _gvn.transform(new AndINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6822
6823 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6824 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6825 generate_fair_guard(tst, slow_region);
6826
6827 // TODO 8350865 This is too strong
6828 generate_fair_guard(flat_array_test(src), slow_region);
6829 generate_fair_guard(flat_array_test(dest), slow_region);
6830
6831 {
6832 PreserveJVMState pjvms(this);
6833 set_control(_gvn.transform(slow_region));
6834 uncommon_trap(Deoptimization::Reason_intrinsic,
6835 Deoptimization::Action_make_not_entrant);
6836 assert(stopped(), "Should be stopped");
6837 }
6838
6839 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6840 if (dest_klass_t == nullptr) {
6841 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6842 // are in a dead path.
6843 uncommon_trap(Deoptimization::Reason_intrinsic,
6844 Deoptimization::Action_make_not_entrant);
6845 return true;
6846 }
6847
6848 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6849 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6850 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6851 }
6852
6853 if (stopped()) {
6854 return true;
6855 }
6856
6857 Node* dest_klass = load_object_klass(dest);
6858 dest_klass = load_non_refined_array_klass(dest_klass);
6859
6860 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6861 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6862 // so the compiler has a chance to eliminate them: during macro expansion,
6863 // we have to set their control (CastPP nodes are eliminated).
6864 load_object_klass(src), dest_klass,
6865 load_array_length(src), load_array_length(dest));
6866
6867 ac->set_arraycopy(validated);
6868
6869 Node* n = _gvn.transform(ac);
6870 if (n == ac) {
6871 ac->connect_outputs(this);
6872 } else {
6873 assert(validated, "shouldn't transform if all arguments not validated");
6874 set_all_memory(n);
6875 }
6876 clear_upper_avx();
6877
6878
6879 return true;
6880 }
6881
6882
6883 // Helper function which determines if an arraycopy immediately follows
6884 // an allocation, with no intervening tests or other escapes for the object.
6885 AllocateArrayNode*
6886 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6887 if (stopped()) return nullptr; // no fast path
6888 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6889
6890 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6891 if (alloc == nullptr) return nullptr;
6892
6893 Node* rawmem = memory(Compile::AliasIdxRaw);
6894 // Is the allocation's memory state untouched?
6895 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6896 // Bail out if there have been raw-memory effects since the allocation.
6897 // (Example: There might have been a call or safepoint.)
6898 return nullptr;
6899 }
6900 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6901 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6902 return nullptr;
6903 }
6904
6905 // There must be no unexpected observers of this allocation.
6906 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6907 Node* obs = ptr->fast_out(i);
6908 if (obs != this->map()) {
6909 return nullptr;
6910 }
6911 }
6912
6913 // This arraycopy must unconditionally follow the allocation of the ptr.
6914 Node* alloc_ctl = ptr->in(0);
6915 Node* ctl = control();
6916 while (ctl != alloc_ctl) {
6917 // There may be guards which feed into the slow_region.
6918 // Any other control flow means that we might not get a chance
6919 // to finish initializing the allocated object.
6920 // Various low-level checks bottom out in uncommon traps. These
6921 // are considered safe since we've already checked above that
6922 // there is no unexpected observer of this allocation.
6923 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6924 assert(ctl->in(0)->is_If(), "must be If");
6925 ctl = ctl->in(0)->in(0);
6926 } else {
6927 return nullptr;
6928 }
6929 }
6930
6931 // If we get this far, we have an allocation which immediately
6932 // precedes the arraycopy, and we can take over zeroing the new object.
6933 // The arraycopy will finish the initialization, and provide
6934 // a new control state to which we will anchor the destination pointer.
6935
6936 return alloc;
6937 }
6938
6939 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6940 if (node->is_IfProj()) {
6941 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6942 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6943 Node* obs = other_proj->fast_out(j);
6944 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6945 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6946 return obs->as_CallStaticJava();
6947 }
6948 }
6949 }
6950 return nullptr;
6951 }
6952
6953 //-------------inline_encodeISOArray-----------------------------------
6954 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6955 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6956 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6957 // encode char[] to byte[] in ISO_8859_1 or ASCII
6958 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6959 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6960 // no receiver since it is static method
6961 Node *src = argument(0);
6962 Node *src_offset = argument(1);
6963 Node *dst = argument(2);
6964 Node *dst_offset = argument(3);
6965 Node *length = argument(4);
6966
6967 // Cast source & target arrays to not-null
6968 if (VerifyIntrinsicChecks) {
6969 src = must_be_not_null(src, true);
6970 dst = must_be_not_null(dst, true);
6971 if (stopped()) {
6972 return true;
6973 }
6974 }
6975
6976 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6977 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6978 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6979 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6980 // failed array check
6981 return false;
6982 }
6983
6984 // Figure out the size and type of the elements we will be copying.
6985 BasicType src_elem = src_type->elem()->array_element_basic_type();
6986 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6987 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6988 return false;
6989 }
6990
6991 // Check source & target bounds
6992 if (VerifyIntrinsicChecks) {
6993 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6994 generate_string_range_check(dst, dst_offset, length, false, true);
6995 if (stopped()) {
6996 return true;
6997 }
6998 }
6999
7000 Node* src_start = array_element_address(src, src_offset, T_CHAR);
7001 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
7002 // 'src_start' points to src array + scaled offset
7003 // 'dst_start' points to dst array + scaled offset
7004
7005 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
7006 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
7007 enc = _gvn.transform(enc);
7008 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
7009 set_memory(res_mem, mtype);
7010 set_result(enc);
7011 clear_upper_avx();
7012
7013 return true;
7014 }
7015
7016 //-------------inline_multiplyToLen-----------------------------------
7017 bool LibraryCallKit::inline_multiplyToLen() {
7018 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
7019
7020 address stubAddr = StubRoutines::multiplyToLen();
7021 if (stubAddr == nullptr) {
7022 return false; // Intrinsic's stub is not implemented on this platform
7023 }
7024 const char* stubName = "multiplyToLen";
7025
7026 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
7027
7028 // no receiver because it is a static method
7029 Node* x = argument(0);
7030 Node* xlen = argument(1);
7031 Node* y = argument(2);
7032 Node* ylen = argument(3);
7033 Node* z = argument(4);
7034
7035 x = must_be_not_null(x, true);
7036 y = must_be_not_null(y, true);
7037
7038 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7039 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
7040 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7041 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
7042 // failed array check
7043 return false;
7044 }
7045
7046 BasicType x_elem = x_type->elem()->array_element_basic_type();
7047 BasicType y_elem = y_type->elem()->array_element_basic_type();
7048 if (x_elem != T_INT || y_elem != T_INT) {
7049 return false;
7050 }
7051
7052 Node* x_start = array_element_address(x, intcon(0), x_elem);
7053 Node* y_start = array_element_address(y, intcon(0), y_elem);
7054 // 'x_start' points to x array + scaled xlen
7055 // 'y_start' points to y array + scaled ylen
7056
7057 Node* z_start = array_element_address(z, intcon(0), T_INT);
7058
7059 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7060 OptoRuntime::multiplyToLen_Type(),
7061 stubAddr, stubName, TypePtr::BOTTOM,
7062 x_start, xlen, y_start, ylen, z_start);
7063
7064 C->set_has_split_ifs(true); // Has chance for split-if optimization
7065 set_result(z);
7066 return true;
7067 }
7068
7069 //-------------inline_squareToLen------------------------------------
7070 bool LibraryCallKit::inline_squareToLen() {
7071 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7072
7073 address stubAddr = StubRoutines::squareToLen();
7074 if (stubAddr == nullptr) {
7075 return false; // Intrinsic's stub is not implemented on this platform
7076 }
7077 const char* stubName = "squareToLen";
7078
7079 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7080
7081 Node* x = argument(0);
7082 Node* len = argument(1);
7083 Node* z = argument(2);
7084 Node* zlen = argument(3);
7085
7086 x = must_be_not_null(x, true);
7087 z = must_be_not_null(z, true);
7088
7089 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7090 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7091 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7092 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7093 // failed array check
7094 return false;
7095 }
7096
7097 BasicType x_elem = x_type->elem()->array_element_basic_type();
7098 BasicType z_elem = z_type->elem()->array_element_basic_type();
7099 if (x_elem != T_INT || z_elem != T_INT) {
7100 return false;
7101 }
7102
7103
7104 Node* x_start = array_element_address(x, intcon(0), x_elem);
7105 Node* z_start = array_element_address(z, intcon(0), z_elem);
7106
7107 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7108 OptoRuntime::squareToLen_Type(),
7109 stubAddr, stubName, TypePtr::BOTTOM,
7110 x_start, len, z_start, zlen);
7111
7112 set_result(z);
7113 return true;
7114 }
7115
7116 //-------------inline_mulAdd------------------------------------------
7117 bool LibraryCallKit::inline_mulAdd() {
7118 assert(UseMulAddIntrinsic, "not implemented on this platform");
7119
7120 address stubAddr = StubRoutines::mulAdd();
7121 if (stubAddr == nullptr) {
7122 return false; // Intrinsic's stub is not implemented on this platform
7123 }
7124 const char* stubName = "mulAdd";
7125
7126 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7127
7128 Node* out = argument(0);
7129 Node* in = argument(1);
7130 Node* offset = argument(2);
7131 Node* len = argument(3);
7132 Node* k = argument(4);
7133
7134 in = must_be_not_null(in, true);
7135 out = must_be_not_null(out, true);
7136
7137 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7138 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7139 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7140 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7141 // failed array check
7142 return false;
7143 }
7144
7145 BasicType out_elem = out_type->elem()->array_element_basic_type();
7146 BasicType in_elem = in_type->elem()->array_element_basic_type();
7147 if (out_elem != T_INT || in_elem != T_INT) {
7148 return false;
7149 }
7150
7151 Node* outlen = load_array_length(out);
7152 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7153 Node* out_start = array_element_address(out, intcon(0), out_elem);
7154 Node* in_start = array_element_address(in, intcon(0), in_elem);
7155
7156 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7157 OptoRuntime::mulAdd_Type(),
7158 stubAddr, stubName, TypePtr::BOTTOM,
7159 out_start,in_start, new_offset, len, k);
7160 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7161 set_result(result);
7162 return true;
7163 }
7164
7165 //-------------inline_montgomeryMultiply-----------------------------------
7166 bool LibraryCallKit::inline_montgomeryMultiply() {
7167 address stubAddr = StubRoutines::montgomeryMultiply();
7168 if (stubAddr == nullptr) {
7169 return false; // Intrinsic's stub is not implemented on this platform
7170 }
7171
7172 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7173 const char* stubName = "montgomery_multiply";
7174
7175 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7176
7177 Node* a = argument(0);
7178 Node* b = argument(1);
7179 Node* n = argument(2);
7180 Node* len = argument(3);
7181 Node* inv = argument(4);
7182 Node* m = argument(6);
7183
7184 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7185 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7186 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7187 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7188 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7189 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7190 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7191 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7192 // failed array check
7193 return false;
7194 }
7195
7196 BasicType a_elem = a_type->elem()->array_element_basic_type();
7197 BasicType b_elem = b_type->elem()->array_element_basic_type();
7198 BasicType n_elem = n_type->elem()->array_element_basic_type();
7199 BasicType m_elem = m_type->elem()->array_element_basic_type();
7200 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7201 return false;
7202 }
7203
7204 // Make the call
7205 {
7206 Node* a_start = array_element_address(a, intcon(0), a_elem);
7207 Node* b_start = array_element_address(b, intcon(0), b_elem);
7208 Node* n_start = array_element_address(n, intcon(0), n_elem);
7209 Node* m_start = array_element_address(m, intcon(0), m_elem);
7210
7211 Node* call = make_runtime_call(RC_LEAF,
7212 OptoRuntime::montgomeryMultiply_Type(),
7213 stubAddr, stubName, TypePtr::BOTTOM,
7214 a_start, b_start, n_start, len, inv, top(),
7215 m_start);
7216 set_result(m);
7217 }
7218
7219 return true;
7220 }
7221
7222 bool LibraryCallKit::inline_montgomerySquare() {
7223 address stubAddr = StubRoutines::montgomerySquare();
7224 if (stubAddr == nullptr) {
7225 return false; // Intrinsic's stub is not implemented on this platform
7226 }
7227
7228 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7229 const char* stubName = "montgomery_square";
7230
7231 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7232
7233 Node* a = argument(0);
7234 Node* n = argument(1);
7235 Node* len = argument(2);
7236 Node* inv = argument(3);
7237 Node* m = argument(5);
7238
7239 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7240 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7241 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7242 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7243 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7244 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7245 // failed array check
7246 return false;
7247 }
7248
7249 BasicType a_elem = a_type->elem()->array_element_basic_type();
7250 BasicType n_elem = n_type->elem()->array_element_basic_type();
7251 BasicType m_elem = m_type->elem()->array_element_basic_type();
7252 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7253 return false;
7254 }
7255
7256 // Make the call
7257 {
7258 Node* a_start = array_element_address(a, intcon(0), a_elem);
7259 Node* n_start = array_element_address(n, intcon(0), n_elem);
7260 Node* m_start = array_element_address(m, intcon(0), m_elem);
7261
7262 Node* call = make_runtime_call(RC_LEAF,
7263 OptoRuntime::montgomerySquare_Type(),
7264 stubAddr, stubName, TypePtr::BOTTOM,
7265 a_start, n_start, len, inv, top(),
7266 m_start);
7267 set_result(m);
7268 }
7269
7270 return true;
7271 }
7272
7273 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7274 address stubAddr = nullptr;
7275 const char* stubName = nullptr;
7276
7277 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7278 if (stubAddr == nullptr) {
7279 return false; // Intrinsic's stub is not implemented on this platform
7280 }
7281
7282 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7283
7284 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7285
7286 Node* newArr = argument(0);
7287 Node* oldArr = argument(1);
7288 Node* newIdx = argument(2);
7289 Node* shiftCount = argument(3);
7290 Node* numIter = argument(4);
7291
7292 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7293 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7294 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7295 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7296 return false;
7297 }
7298
7299 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7300 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7301 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7302 return false;
7303 }
7304
7305 // Make the call
7306 {
7307 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7308 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7309
7310 Node* call = make_runtime_call(RC_LEAF,
7311 OptoRuntime::bigIntegerShift_Type(),
7312 stubAddr,
7313 stubName,
7314 TypePtr::BOTTOM,
7315 newArr_start,
7316 oldArr_start,
7317 newIdx,
7318 shiftCount,
7319 numIter);
7320 }
7321
7322 return true;
7323 }
7324
7325 //-------------inline_vectorizedMismatch------------------------------
7326 bool LibraryCallKit::inline_vectorizedMismatch() {
7327 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7328
7329 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7330 Node* obja = argument(0); // Object
7331 Node* aoffset = argument(1); // long
7332 Node* objb = argument(3); // Object
7333 Node* boffset = argument(4); // long
7334 Node* length = argument(6); // int
7335 Node* scale = argument(7); // int
7336
7337 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7338 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7339 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7340 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7341 scale == top()) {
7342 return false; // failed input validation
7343 }
7344
7345 Node* obja_adr = make_unsafe_address(obja, aoffset);
7346 Node* objb_adr = make_unsafe_address(objb, boffset);
7347
7348 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7349 //
7350 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7351 // if (length <= inline_limit) {
7352 // inline_path:
7353 // vmask = VectorMaskGen length
7354 // vload1 = LoadVectorMasked obja, vmask
7355 // vload2 = LoadVectorMasked objb, vmask
7356 // result1 = VectorCmpMasked vload1, vload2, vmask
7357 // } else {
7358 // call_stub_path:
7359 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7360 // }
7361 // exit_block:
7362 // return Phi(result1, result2);
7363 //
7364 enum { inline_path = 1, // input is small enough to process it all at once
7365 stub_path = 2, // input is too large; call into the VM
7366 PATH_LIMIT = 3
7367 };
7368
7369 Node* exit_block = new RegionNode(PATH_LIMIT);
7370 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7371 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7372
7373 Node* call_stub_path = control();
7374
7375 BasicType elem_bt = T_ILLEGAL;
7376
7377 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7378 if (scale_t->is_con()) {
7379 switch (scale_t->get_con()) {
7380 case 0: elem_bt = T_BYTE; break;
7381 case 1: elem_bt = T_SHORT; break;
7382 case 2: elem_bt = T_INT; break;
7383 case 3: elem_bt = T_LONG; break;
7384
7385 default: elem_bt = T_ILLEGAL; break; // not supported
7386 }
7387 }
7388
7389 int inline_limit = 0;
7390 bool do_partial_inline = false;
7391
7392 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7393 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7394 do_partial_inline = inline_limit >= 16;
7395 }
7396
7397 if (do_partial_inline) {
7398 assert(elem_bt != T_ILLEGAL, "sanity");
7399
7400 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7401 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7402 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7403
7404 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7405 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7406 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7407
7408 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7409
7410 if (!stopped()) {
7411 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7412
7413 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7414 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7415 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7416 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7417
7418 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7419 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7420 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7421 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7422
7423 exit_block->init_req(inline_path, control());
7424 memory_phi->init_req(inline_path, map()->memory());
7425 result_phi->init_req(inline_path, result);
7426
7427 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7428 clear_upper_avx();
7429 }
7430 }
7431 }
7432
7433 if (call_stub_path != nullptr) {
7434 set_control(call_stub_path);
7435
7436 Node* call = make_runtime_call(RC_LEAF,
7437 OptoRuntime::vectorizedMismatch_Type(),
7438 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7439 obja_adr, objb_adr, length, scale);
7440
7441 exit_block->init_req(stub_path, control());
7442 memory_phi->init_req(stub_path, map()->memory());
7443 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7444 }
7445
7446 exit_block = _gvn.transform(exit_block);
7447 memory_phi = _gvn.transform(memory_phi);
7448 result_phi = _gvn.transform(result_phi);
7449
7450 record_for_igvn(exit_block);
7451 record_for_igvn(memory_phi);
7452 record_for_igvn(result_phi);
7453
7454 set_control(exit_block);
7455 set_all_memory(memory_phi);
7456 set_result(result_phi);
7457
7458 return true;
7459 }
7460
7461 //------------------------------inline_vectorizedHashcode----------------------------
7462 bool LibraryCallKit::inline_vectorizedHashCode() {
7463 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7464
7465 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7466 Node* array = argument(0);
7467 Node* offset = argument(1);
7468 Node* length = argument(2);
7469 Node* initialValue = argument(3);
7470 Node* basic_type = argument(4);
7471
7472 if (basic_type == top()) {
7473 return false; // failed input validation
7474 }
7475
7476 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7477 if (!basic_type_t->is_con()) {
7478 return false; // Only intrinsify if mode argument is constant
7479 }
7480
7481 array = must_be_not_null(array, true);
7482
7483 BasicType bt = (BasicType)basic_type_t->get_con();
7484
7485 // Resolve address of first element
7486 Node* array_start = array_element_address(array, offset, bt);
7487
7488 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7489 array_start, length, initialValue, basic_type)));
7490 clear_upper_avx();
7491
7492 return true;
7493 }
7494
7495 /**
7496 * Calculate CRC32 for byte.
7497 * int java.util.zip.CRC32.update(int crc, int b)
7498 */
7499 bool LibraryCallKit::inline_updateCRC32() {
7500 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7501 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7502 // no receiver since it is static method
7503 Node* crc = argument(0); // type: int
7504 Node* b = argument(1); // type: int
7505
7506 /*
7507 * int c = ~ crc;
7508 * b = timesXtoThe32[(b ^ c) & 0xFF];
7509 * b = b ^ (c >>> 8);
7510 * crc = ~b;
7511 */
7512
7513 Node* M1 = intcon(-1);
7514 crc = _gvn.transform(new XorINode(crc, M1));
7515 Node* result = _gvn.transform(new XorINode(crc, b));
7516 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7517
7518 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7519 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7520 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7521 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7522
7523 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7524 result = _gvn.transform(new XorINode(crc, result));
7525 result = _gvn.transform(new XorINode(result, M1));
7526 set_result(result);
7527 return true;
7528 }
7529
7530 /**
7531 * Calculate CRC32 for byte[] array.
7532 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7533 */
7534 bool LibraryCallKit::inline_updateBytesCRC32() {
7535 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7536 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7537 // no receiver since it is static method
7538 Node* crc = argument(0); // type: int
7539 Node* src = argument(1); // type: oop
7540 Node* offset = argument(2); // type: int
7541 Node* length = argument(3); // type: int
7542
7543 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7544 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7545 // failed array check
7546 return false;
7547 }
7548
7549 // Figure out the size and type of the elements we will be copying.
7550 BasicType src_elem = src_type->elem()->array_element_basic_type();
7551 if (src_elem != T_BYTE) {
7552 return false;
7553 }
7554
7555 // 'src_start' points to src array + scaled offset
7556 src = must_be_not_null(src, true);
7557 Node* src_start = array_element_address(src, offset, src_elem);
7558
7559 // We assume that range check is done by caller.
7560 // TODO: generate range check (offset+length < src.length) in debug VM.
7561
7562 // Call the stub.
7563 address stubAddr = StubRoutines::updateBytesCRC32();
7564 const char *stubName = "updateBytesCRC32";
7565
7566 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7567 stubAddr, stubName, TypePtr::BOTTOM,
7568 crc, src_start, length);
7569 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7570 set_result(result);
7571 return true;
7572 }
7573
7574 /**
7575 * Calculate CRC32 for ByteBuffer.
7576 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7577 */
7578 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7579 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7580 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7581 // no receiver since it is static method
7582 Node* crc = argument(0); // type: int
7583 Node* src = argument(1); // type: long
7584 Node* offset = argument(3); // type: int
7585 Node* length = argument(4); // type: int
7586
7587 src = ConvL2X(src); // adjust Java long to machine word
7588 Node* base = _gvn.transform(new CastX2PNode(src));
7589 offset = ConvI2X(offset);
7590
7591 // 'src_start' points to src array + scaled offset
7592 Node* src_start = basic_plus_adr(top(), base, offset);
7593
7594 // Call the stub.
7595 address stubAddr = StubRoutines::updateBytesCRC32();
7596 const char *stubName = "updateBytesCRC32";
7597
7598 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7599 stubAddr, stubName, TypePtr::BOTTOM,
7600 crc, src_start, length);
7601 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7602 set_result(result);
7603 return true;
7604 }
7605
7606 //------------------------------get_table_from_crc32c_class-----------------------
7607 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7608 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7609 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7610
7611 return table;
7612 }
7613
7614 //------------------------------inline_updateBytesCRC32C-----------------------
7615 //
7616 // Calculate CRC32C for byte[] array.
7617 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7618 //
7619 bool LibraryCallKit::inline_updateBytesCRC32C() {
7620 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7621 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7622 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7623 // no receiver since it is a static method
7624 Node* crc = argument(0); // type: int
7625 Node* src = argument(1); // type: oop
7626 Node* offset = argument(2); // type: int
7627 Node* end = argument(3); // type: int
7628
7629 Node* length = _gvn.transform(new SubINode(end, offset));
7630
7631 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7632 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7633 // failed array check
7634 return false;
7635 }
7636
7637 // Figure out the size and type of the elements we will be copying.
7638 BasicType src_elem = src_type->elem()->array_element_basic_type();
7639 if (src_elem != T_BYTE) {
7640 return false;
7641 }
7642
7643 // 'src_start' points to src array + scaled offset
7644 src = must_be_not_null(src, true);
7645 Node* src_start = array_element_address(src, offset, src_elem);
7646
7647 // static final int[] byteTable in class CRC32C
7648 Node* table = get_table_from_crc32c_class(callee()->holder());
7649 table = must_be_not_null(table, true);
7650 Node* table_start = array_element_address(table, intcon(0), T_INT);
7651
7652 // We assume that range check is done by caller.
7653 // TODO: generate range check (offset+length < src.length) in debug VM.
7654
7655 // Call the stub.
7656 address stubAddr = StubRoutines::updateBytesCRC32C();
7657 const char *stubName = "updateBytesCRC32C";
7658
7659 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7660 stubAddr, stubName, TypePtr::BOTTOM,
7661 crc, src_start, length, table_start);
7662 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7663 set_result(result);
7664 return true;
7665 }
7666
7667 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7668 //
7669 // Calculate CRC32C for DirectByteBuffer.
7670 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7671 //
7672 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7673 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7674 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7675 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7676 // no receiver since it is a static method
7677 Node* crc = argument(0); // type: int
7678 Node* src = argument(1); // type: long
7679 Node* offset = argument(3); // type: int
7680 Node* end = argument(4); // type: int
7681
7682 Node* length = _gvn.transform(new SubINode(end, offset));
7683
7684 src = ConvL2X(src); // adjust Java long to machine word
7685 Node* base = _gvn.transform(new CastX2PNode(src));
7686 offset = ConvI2X(offset);
7687
7688 // 'src_start' points to src array + scaled offset
7689 Node* src_start = basic_plus_adr(top(), base, offset);
7690
7691 // static final int[] byteTable in class CRC32C
7692 Node* table = get_table_from_crc32c_class(callee()->holder());
7693 table = must_be_not_null(table, true);
7694 Node* table_start = array_element_address(table, intcon(0), T_INT);
7695
7696 // Call the stub.
7697 address stubAddr = StubRoutines::updateBytesCRC32C();
7698 const char *stubName = "updateBytesCRC32C";
7699
7700 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7701 stubAddr, stubName, TypePtr::BOTTOM,
7702 crc, src_start, length, table_start);
7703 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7704 set_result(result);
7705 return true;
7706 }
7707
7708 //------------------------------inline_updateBytesAdler32----------------------
7709 //
7710 // Calculate Adler32 checksum for byte[] array.
7711 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7712 //
7713 bool LibraryCallKit::inline_updateBytesAdler32() {
7714 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7715 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7716 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7717 // no receiver since it is static method
7718 Node* crc = argument(0); // type: int
7719 Node* src = argument(1); // type: oop
7720 Node* offset = argument(2); // type: int
7721 Node* length = argument(3); // type: int
7722
7723 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7724 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7725 // failed array check
7726 return false;
7727 }
7728
7729 // Figure out the size and type of the elements we will be copying.
7730 BasicType src_elem = src_type->elem()->array_element_basic_type();
7731 if (src_elem != T_BYTE) {
7732 return false;
7733 }
7734
7735 // 'src_start' points to src array + scaled offset
7736 Node* src_start = array_element_address(src, offset, src_elem);
7737
7738 // We assume that range check is done by caller.
7739 // TODO: generate range check (offset+length < src.length) in debug VM.
7740
7741 // Call the stub.
7742 address stubAddr = StubRoutines::updateBytesAdler32();
7743 const char *stubName = "updateBytesAdler32";
7744
7745 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7746 stubAddr, stubName, TypePtr::BOTTOM,
7747 crc, src_start, length);
7748 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7749 set_result(result);
7750 return true;
7751 }
7752
7753 //------------------------------inline_updateByteBufferAdler32---------------
7754 //
7755 // Calculate Adler32 checksum for DirectByteBuffer.
7756 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7757 //
7758 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7759 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7760 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7761 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7762 // no receiver since it is static method
7763 Node* crc = argument(0); // type: int
7764 Node* src = argument(1); // type: long
7765 Node* offset = argument(3); // type: int
7766 Node* length = argument(4); // type: int
7767
7768 src = ConvL2X(src); // adjust Java long to machine word
7769 Node* base = _gvn.transform(new CastX2PNode(src));
7770 offset = ConvI2X(offset);
7771
7772 // 'src_start' points to src array + scaled offset
7773 Node* src_start = basic_plus_adr(top(), base, offset);
7774
7775 // Call the stub.
7776 address stubAddr = StubRoutines::updateBytesAdler32();
7777 const char *stubName = "updateBytesAdler32";
7778
7779 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7780 stubAddr, stubName, TypePtr::BOTTOM,
7781 crc, src_start, length);
7782
7783 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7784 set_result(result);
7785 return true;
7786 }
7787
7788 //----------------------------inline_reference_get0----------------------------
7789 // public T java.lang.ref.Reference.get();
7790 bool LibraryCallKit::inline_reference_get0() {
7791 const int referent_offset = java_lang_ref_Reference::referent_offset();
7792
7793 // Get the argument:
7794 Node* reference_obj = null_check_receiver();
7795 if (stopped()) return true;
7796
7797 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7798 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7799 decorators, /*is_static*/ false, nullptr);
7800 if (result == nullptr) return false;
7801
7802 // Add memory barrier to prevent commoning reads from this field
7803 // across safepoint since GC can change its value.
7804 insert_mem_bar(Op_MemBarCPUOrder);
7805
7806 set_result(result);
7807 return true;
7808 }
7809
7810 //----------------------------inline_reference_refersTo0----------------------------
7811 // bool java.lang.ref.Reference.refersTo0();
7812 // bool java.lang.ref.PhantomReference.refersTo0();
7813 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7814 // Get arguments:
7815 Node* reference_obj = null_check_receiver();
7816 Node* other_obj = argument(1);
7817 if (stopped()) return true;
7818
7819 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7820 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7821 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7822 decorators, /*is_static*/ false, nullptr);
7823 if (referent == nullptr) return false;
7824
7825 // Add memory barrier to prevent commoning reads from this field
7826 // across safepoint since GC can change its value.
7827 insert_mem_bar(Op_MemBarCPUOrder);
7828
7829 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7830 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7831 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7832
7833 RegionNode* region = new RegionNode(3);
7834 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7835
7836 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7837 region->init_req(1, if_true);
7838 phi->init_req(1, intcon(1));
7839
7840 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7841 region->init_req(2, if_false);
7842 phi->init_req(2, intcon(0));
7843
7844 set_control(_gvn.transform(region));
7845 record_for_igvn(region);
7846 set_result(_gvn.transform(phi));
7847 return true;
7848 }
7849
7850 //----------------------------inline_reference_clear0----------------------------
7851 // void java.lang.ref.Reference.clear0();
7852 // void java.lang.ref.PhantomReference.clear0();
7853 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7854 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7855
7856 // Get arguments
7857 Node* reference_obj = null_check_receiver();
7858 if (stopped()) return true;
7859
7860 // Common access parameters
7861 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7862 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7863 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7864 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7865 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7866
7867 Node* referent = access_load_at(reference_obj,
7868 referent_field_addr,
7869 referent_field_addr_type,
7870 val_type,
7871 T_OBJECT,
7872 decorators);
7873
7874 IdealKit ideal(this);
7875 #define __ ideal.
7876 __ if_then(referent, BoolTest::ne, null());
7877 sync_kit(ideal);
7878 access_store_at(reference_obj,
7879 referent_field_addr,
7880 referent_field_addr_type,
7881 null(),
7882 val_type,
7883 T_OBJECT,
7884 decorators);
7885 __ sync_kit(this);
7886 __ end_if();
7887 final_sync(ideal);
7888 #undef __
7889
7890 return true;
7891 }
7892
7893 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7894 DecoratorSet decorators, bool is_static,
7895 ciInstanceKlass* fromKls) {
7896 if (fromKls == nullptr) {
7897 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7898 assert(tinst != nullptr, "obj is null");
7899 assert(tinst->is_loaded(), "obj is not loaded");
7900 fromKls = tinst->instance_klass();
7901 } else {
7902 assert(is_static, "only for static field access");
7903 }
7904 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7905 ciSymbol::make(fieldTypeString),
7906 is_static);
7907
7908 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7909 if (field == nullptr) return (Node *) nullptr;
7910
7911 if (is_static) {
7912 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7913 fromObj = makecon(tip);
7914 }
7915
7916 // Next code copied from Parse::do_get_xxx():
7917
7918 // Compute address and memory type.
7919 int offset = field->offset_in_bytes();
7920 bool is_vol = field->is_volatile();
7921 ciType* field_klass = field->type();
7922 assert(field_klass->is_loaded(), "should be loaded");
7923 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7924 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7925 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7926 "slice of address and input slice don't match");
7927 BasicType bt = field->layout_type();
7928
7929 // Build the resultant type of the load
7930 const Type *type;
7931 if (bt == T_OBJECT) {
7932 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7933 } else {
7934 type = Type::get_const_basic_type(bt);
7935 }
7936
7937 if (is_vol) {
7938 decorators |= MO_SEQ_CST;
7939 }
7940
7941 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7942 }
7943
7944 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7945 bool is_exact /* true */, bool is_static /* false */,
7946 ciInstanceKlass * fromKls /* nullptr */) {
7947 if (fromKls == nullptr) {
7948 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7949 assert(tinst != nullptr, "obj is null");
7950 assert(tinst->is_loaded(), "obj is not loaded");
7951 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7952 fromKls = tinst->instance_klass();
7953 }
7954 else {
7955 assert(is_static, "only for static field access");
7956 }
7957 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7958 ciSymbol::make(fieldTypeString),
7959 is_static);
7960
7961 assert(field != nullptr, "undefined field");
7962 assert(!field->is_volatile(), "not defined for volatile fields");
7963
7964 if (is_static) {
7965 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7966 fromObj = makecon(tip);
7967 }
7968
7969 // Next code copied from Parse::do_get_xxx():
7970
7971 // Compute address and memory type.
7972 int offset = field->offset_in_bytes();
7973 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7974
7975 return adr;
7976 }
7977
7978 //------------------------------inline_aescrypt_Block-----------------------
7979 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7980 address stubAddr = nullptr;
7981 const char *stubName;
7982 bool is_decrypt = false;
7983 assert(UseAES, "need AES instruction support");
7984
7985 switch(id) {
7986 case vmIntrinsics::_aescrypt_encryptBlock:
7987 stubAddr = StubRoutines::aescrypt_encryptBlock();
7988 stubName = "aescrypt_encryptBlock";
7989 break;
7990 case vmIntrinsics::_aescrypt_decryptBlock:
7991 stubAddr = StubRoutines::aescrypt_decryptBlock();
7992 stubName = "aescrypt_decryptBlock";
7993 is_decrypt = true;
7994 break;
7995 default:
7996 break;
7997 }
7998 if (stubAddr == nullptr) return false;
7999
8000 Node* aescrypt_object = argument(0);
8001 Node* src = argument(1);
8002 Node* src_offset = argument(2);
8003 Node* dest = argument(3);
8004 Node* dest_offset = argument(4);
8005
8006 src = must_be_not_null(src, true);
8007 dest = must_be_not_null(dest, true);
8008
8009 // (1) src and dest are arrays.
8010 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8011 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8012 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8013 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8014
8015 // for the quick and dirty code we will skip all the checks.
8016 // we are just trying to get the call to be generated.
8017 Node* src_start = src;
8018 Node* dest_start = dest;
8019 if (src_offset != nullptr || dest_offset != nullptr) {
8020 assert(src_offset != nullptr && dest_offset != nullptr, "");
8021 src_start = array_element_address(src, src_offset, T_BYTE);
8022 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8023 }
8024
8025 // now need to get the start of its expanded key array
8026 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8027 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8028 if (k_start == nullptr) return false;
8029
8030 // Call the stub.
8031 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
8032 stubAddr, stubName, TypePtr::BOTTOM,
8033 src_start, dest_start, k_start);
8034
8035 return true;
8036 }
8037
8038 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
8039 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
8040 address stubAddr = nullptr;
8041 const char *stubName = nullptr;
8042 bool is_decrypt = false;
8043 assert(UseAES, "need AES instruction support");
8044
8045 switch(id) {
8046 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
8047 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
8048 stubName = "cipherBlockChaining_encryptAESCrypt";
8049 break;
8050 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
8051 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
8052 stubName = "cipherBlockChaining_decryptAESCrypt";
8053 is_decrypt = true;
8054 break;
8055 default:
8056 break;
8057 }
8058 if (stubAddr == nullptr) return false;
8059
8060 Node* cipherBlockChaining_object = argument(0);
8061 Node* src = argument(1);
8062 Node* src_offset = argument(2);
8063 Node* len = argument(3);
8064 Node* dest = argument(4);
8065 Node* dest_offset = argument(5);
8066
8067 src = must_be_not_null(src, false);
8068 dest = must_be_not_null(dest, false);
8069
8070 // (1) src and dest are arrays.
8071 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8072 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8073 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8074 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8075
8076 // checks are the responsibility of the caller
8077 Node* src_start = src;
8078 Node* dest_start = dest;
8079 if (src_offset != nullptr || dest_offset != nullptr) {
8080 assert(src_offset != nullptr && dest_offset != nullptr, "");
8081 src_start = array_element_address(src, src_offset, T_BYTE);
8082 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8083 }
8084
8085 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8086 // (because of the predicated logic executed earlier).
8087 // so we cast it here safely.
8088 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8089
8090 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8091 if (embeddedCipherObj == nullptr) return false;
8092
8093 // cast it to what we know it will be at runtime
8094 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8095 assert(tinst != nullptr, "CBC obj is null");
8096 assert(tinst->is_loaded(), "CBC obj is not loaded");
8097 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8098 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8099
8100 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8101 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8102 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8103 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8104 aescrypt_object = _gvn.transform(aescrypt_object);
8105
8106 // we need to get the start of the aescrypt_object's expanded key array
8107 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8108 if (k_start == nullptr) return false;
8109
8110 // similarly, get the start address of the r vector
8111 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8112 if (objRvec == nullptr) return false;
8113 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8114
8115 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8116 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8117 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8118 stubAddr, stubName, TypePtr::BOTTOM,
8119 src_start, dest_start, k_start, r_start, len);
8120
8121 // return cipher length (int)
8122 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8123 set_result(retvalue);
8124 return true;
8125 }
8126
8127 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8128 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8129 address stubAddr = nullptr;
8130 const char *stubName = nullptr;
8131 bool is_decrypt = false;
8132 assert(UseAES, "need AES instruction support");
8133
8134 switch (id) {
8135 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8136 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8137 stubName = "electronicCodeBook_encryptAESCrypt";
8138 break;
8139 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8140 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8141 stubName = "electronicCodeBook_decryptAESCrypt";
8142 is_decrypt = true;
8143 break;
8144 default:
8145 break;
8146 }
8147
8148 if (stubAddr == nullptr) return false;
8149
8150 Node* electronicCodeBook_object = argument(0);
8151 Node* src = argument(1);
8152 Node* src_offset = argument(2);
8153 Node* len = argument(3);
8154 Node* dest = argument(4);
8155 Node* dest_offset = argument(5);
8156
8157 // (1) src and dest are arrays.
8158 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8159 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8160 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8161 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8162
8163 // checks are the responsibility of the caller
8164 Node* src_start = src;
8165 Node* dest_start = dest;
8166 if (src_offset != nullptr || dest_offset != nullptr) {
8167 assert(src_offset != nullptr && dest_offset != nullptr, "");
8168 src_start = array_element_address(src, src_offset, T_BYTE);
8169 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8170 }
8171
8172 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8173 // (because of the predicated logic executed earlier).
8174 // so we cast it here safely.
8175 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8176
8177 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8178 if (embeddedCipherObj == nullptr) return false;
8179
8180 // cast it to what we know it will be at runtime
8181 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8182 assert(tinst != nullptr, "ECB obj is null");
8183 assert(tinst->is_loaded(), "ECB obj is not loaded");
8184 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8185 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8186
8187 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8188 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8189 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8190 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8191 aescrypt_object = _gvn.transform(aescrypt_object);
8192
8193 // we need to get the start of the aescrypt_object's expanded key array
8194 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8195 if (k_start == nullptr) return false;
8196
8197 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8198 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8199 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8200 stubAddr, stubName, TypePtr::BOTTOM,
8201 src_start, dest_start, k_start, len);
8202
8203 // return cipher length (int)
8204 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8205 set_result(retvalue);
8206 return true;
8207 }
8208
8209 //------------------------------inline_counterMode_AESCrypt-----------------------
8210 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8211 assert(UseAES, "need AES instruction support");
8212 if (!UseAESCTRIntrinsics) return false;
8213
8214 address stubAddr = nullptr;
8215 const char *stubName = nullptr;
8216 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8217 stubAddr = StubRoutines::counterMode_AESCrypt();
8218 stubName = "counterMode_AESCrypt";
8219 }
8220 if (stubAddr == nullptr) return false;
8221
8222 Node* counterMode_object = argument(0);
8223 Node* src = argument(1);
8224 Node* src_offset = argument(2);
8225 Node* len = argument(3);
8226 Node* dest = argument(4);
8227 Node* dest_offset = argument(5);
8228
8229 // (1) src and dest are arrays.
8230 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8231 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8232 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8233 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8234
8235 // checks are the responsibility of the caller
8236 Node* src_start = src;
8237 Node* dest_start = dest;
8238 if (src_offset != nullptr || dest_offset != nullptr) {
8239 assert(src_offset != nullptr && dest_offset != nullptr, "");
8240 src_start = array_element_address(src, src_offset, T_BYTE);
8241 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8242 }
8243
8244 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8245 // (because of the predicated logic executed earlier).
8246 // so we cast it here safely.
8247 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8248 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8249 if (embeddedCipherObj == nullptr) return false;
8250 // cast it to what we know it will be at runtime
8251 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8252 assert(tinst != nullptr, "CTR obj is null");
8253 assert(tinst->is_loaded(), "CTR obj is not loaded");
8254 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8255 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8256 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8257 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8258 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8259 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8260 aescrypt_object = _gvn.transform(aescrypt_object);
8261 // we need to get the start of the aescrypt_object's expanded key array
8262 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8263 if (k_start == nullptr) return false;
8264 // similarly, get the start address of the r vector
8265 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8266 if (obj_counter == nullptr) return false;
8267 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8268
8269 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8270 if (saved_encCounter == nullptr) return false;
8271 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8272 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8273
8274 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8275 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8276 OptoRuntime::counterMode_aescrypt_Type(),
8277 stubAddr, stubName, TypePtr::BOTTOM,
8278 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8279
8280 // return cipher length (int)
8281 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8282 set_result(retvalue);
8283 return true;
8284 }
8285
8286 //------------------------------get_key_start_from_aescrypt_object-----------------------
8287 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8288 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8289 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8290 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8291 // The following platform specific stubs of encryption and decryption use the same round keys.
8292 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8293 bool use_decryption_key = false;
8294 #else
8295 bool use_decryption_key = is_decrypt;
8296 #endif
8297 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8298 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8299 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8300
8301 // now have the array, need to get the start address of the selected key array
8302 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8303 return k_start;
8304 }
8305
8306 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8307 // Return node representing slow path of predicate check.
8308 // the pseudo code we want to emulate with this predicate is:
8309 // for encryption:
8310 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8311 // for decryption:
8312 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8313 // note cipher==plain is more conservative than the original java code but that's OK
8314 //
8315 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8316 // The receiver was checked for null already.
8317 Node* objCBC = argument(0);
8318
8319 Node* src = argument(1);
8320 Node* dest = argument(4);
8321
8322 // Load embeddedCipher field of CipherBlockChaining object.
8323 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8324
8325 // get AESCrypt klass for instanceOf check
8326 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8327 // will have same classloader as CipherBlockChaining object
8328 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8329 assert(tinst != nullptr, "CBCobj is null");
8330 assert(tinst->is_loaded(), "CBCobj is not loaded");
8331
8332 // we want to do an instanceof comparison against the AESCrypt class
8333 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8334 if (!klass_AESCrypt->is_loaded()) {
8335 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8336 Node* ctrl = control();
8337 set_control(top()); // no regular fast path
8338 return ctrl;
8339 }
8340
8341 src = must_be_not_null(src, true);
8342 dest = must_be_not_null(dest, true);
8343
8344 // Resolve oops to stable for CmpP below.
8345 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8346
8347 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8348 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8349 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8350
8351 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8352
8353 // for encryption, we are done
8354 if (!decrypting)
8355 return instof_false; // even if it is null
8356
8357 // for decryption, we need to add a further check to avoid
8358 // taking the intrinsic path when cipher and plain are the same
8359 // see the original java code for why.
8360 RegionNode* region = new RegionNode(3);
8361 region->init_req(1, instof_false);
8362
8363 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8364 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8365 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8366 region->init_req(2, src_dest_conjoint);
8367
8368 record_for_igvn(region);
8369 return _gvn.transform(region);
8370 }
8371
8372 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8373 // Return node representing slow path of predicate check.
8374 // the pseudo code we want to emulate with this predicate is:
8375 // for encryption:
8376 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8377 // for decryption:
8378 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8379 // note cipher==plain is more conservative than the original java code but that's OK
8380 //
8381 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8382 // The receiver was checked for null already.
8383 Node* objECB = argument(0);
8384
8385 // Load embeddedCipher field of ElectronicCodeBook object.
8386 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8387
8388 // get AESCrypt klass for instanceOf check
8389 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8390 // will have same classloader as ElectronicCodeBook object
8391 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8392 assert(tinst != nullptr, "ECBobj is null");
8393 assert(tinst->is_loaded(), "ECBobj is not loaded");
8394
8395 // we want to do an instanceof comparison against the AESCrypt class
8396 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8397 if (!klass_AESCrypt->is_loaded()) {
8398 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8399 Node* ctrl = control();
8400 set_control(top()); // no regular fast path
8401 return ctrl;
8402 }
8403 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8404
8405 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8406 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8407 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8408
8409 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8410
8411 // for encryption, we are done
8412 if (!decrypting)
8413 return instof_false; // even if it is null
8414
8415 // for decryption, we need to add a further check to avoid
8416 // taking the intrinsic path when cipher and plain are the same
8417 // see the original java code for why.
8418 RegionNode* region = new RegionNode(3);
8419 region->init_req(1, instof_false);
8420 Node* src = argument(1);
8421 Node* dest = argument(4);
8422 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8423 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8424 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8425 region->init_req(2, src_dest_conjoint);
8426
8427 record_for_igvn(region);
8428 return _gvn.transform(region);
8429 }
8430
8431 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8432 // Return node representing slow path of predicate check.
8433 // the pseudo code we want to emulate with this predicate is:
8434 // for encryption:
8435 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8436 // for decryption:
8437 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8438 // note cipher==plain is more conservative than the original java code but that's OK
8439 //
8440
8441 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8442 // The receiver was checked for null already.
8443 Node* objCTR = argument(0);
8444
8445 // Load embeddedCipher field of CipherBlockChaining object.
8446 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8447
8448 // get AESCrypt klass for instanceOf check
8449 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8450 // will have same classloader as CipherBlockChaining object
8451 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8452 assert(tinst != nullptr, "CTRobj is null");
8453 assert(tinst->is_loaded(), "CTRobj is not loaded");
8454
8455 // we want to do an instanceof comparison against the AESCrypt class
8456 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8457 if (!klass_AESCrypt->is_loaded()) {
8458 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8459 Node* ctrl = control();
8460 set_control(top()); // no regular fast path
8461 return ctrl;
8462 }
8463
8464 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8465 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8466 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8467 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8468 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8469
8470 return instof_false; // even if it is null
8471 }
8472
8473 //------------------------------inline_ghash_processBlocks
8474 bool LibraryCallKit::inline_ghash_processBlocks() {
8475 address stubAddr;
8476 const char *stubName;
8477 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8478
8479 stubAddr = StubRoutines::ghash_processBlocks();
8480 stubName = "ghash_processBlocks";
8481
8482 Node* data = argument(0);
8483 Node* offset = argument(1);
8484 Node* len = argument(2);
8485 Node* state = argument(3);
8486 Node* subkeyH = argument(4);
8487
8488 state = must_be_not_null(state, true);
8489 subkeyH = must_be_not_null(subkeyH, true);
8490 data = must_be_not_null(data, true);
8491
8492 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8493 assert(state_start, "state is null");
8494 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8495 assert(subkeyH_start, "subkeyH is null");
8496 Node* data_start = array_element_address(data, offset, T_BYTE);
8497 assert(data_start, "data is null");
8498
8499 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8500 OptoRuntime::ghash_processBlocks_Type(),
8501 stubAddr, stubName, TypePtr::BOTTOM,
8502 state_start, subkeyH_start, data_start, len);
8503 return true;
8504 }
8505
8506 //------------------------------inline_chacha20Block
8507 bool LibraryCallKit::inline_chacha20Block() {
8508 address stubAddr;
8509 const char *stubName;
8510 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8511
8512 stubAddr = StubRoutines::chacha20Block();
8513 stubName = "chacha20Block";
8514
8515 Node* state = argument(0);
8516 Node* result = argument(1);
8517
8518 state = must_be_not_null(state, true);
8519 result = must_be_not_null(result, true);
8520
8521 Node* state_start = array_element_address(state, intcon(0), T_INT);
8522 assert(state_start, "state is null");
8523 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8524 assert(result_start, "result is null");
8525
8526 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8527 OptoRuntime::chacha20Block_Type(),
8528 stubAddr, stubName, TypePtr::BOTTOM,
8529 state_start, result_start);
8530 // return key stream length (int)
8531 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8532 set_result(retvalue);
8533 return true;
8534 }
8535
8536 //------------------------------inline_kyberNtt
8537 bool LibraryCallKit::inline_kyberNtt() {
8538 address stubAddr;
8539 const char *stubName;
8540 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8541 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8542
8543 stubAddr = StubRoutines::kyberNtt();
8544 stubName = "kyberNtt";
8545 if (!stubAddr) return false;
8546
8547 Node* coeffs = argument(0);
8548 Node* ntt_zetas = argument(1);
8549
8550 coeffs = must_be_not_null(coeffs, true);
8551 ntt_zetas = must_be_not_null(ntt_zetas, true);
8552
8553 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8554 assert(coeffs_start, "coeffs is null");
8555 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8556 assert(ntt_zetas_start, "ntt_zetas is null");
8557 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8558 OptoRuntime::kyberNtt_Type(),
8559 stubAddr, stubName, TypePtr::BOTTOM,
8560 coeffs_start, ntt_zetas_start);
8561 // return an int
8562 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8563 set_result(retvalue);
8564 return true;
8565 }
8566
8567 //------------------------------inline_kyberInverseNtt
8568 bool LibraryCallKit::inline_kyberInverseNtt() {
8569 address stubAddr;
8570 const char *stubName;
8571 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8572 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8573
8574 stubAddr = StubRoutines::kyberInverseNtt();
8575 stubName = "kyberInverseNtt";
8576 if (!stubAddr) return false;
8577
8578 Node* coeffs = argument(0);
8579 Node* zetas = argument(1);
8580
8581 coeffs = must_be_not_null(coeffs, true);
8582 zetas = must_be_not_null(zetas, true);
8583
8584 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8585 assert(coeffs_start, "coeffs is null");
8586 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8587 assert(zetas_start, "inverseNtt_zetas is null");
8588 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8589 OptoRuntime::kyberInverseNtt_Type(),
8590 stubAddr, stubName, TypePtr::BOTTOM,
8591 coeffs_start, zetas_start);
8592
8593 // return an int
8594 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8595 set_result(retvalue);
8596 return true;
8597 }
8598
8599 //------------------------------inline_kyberNttMult
8600 bool LibraryCallKit::inline_kyberNttMult() {
8601 address stubAddr;
8602 const char *stubName;
8603 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8604 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8605
8606 stubAddr = StubRoutines::kyberNttMult();
8607 stubName = "kyberNttMult";
8608 if (!stubAddr) return false;
8609
8610 Node* result = argument(0);
8611 Node* ntta = argument(1);
8612 Node* nttb = argument(2);
8613 Node* zetas = argument(3);
8614
8615 result = must_be_not_null(result, true);
8616 ntta = must_be_not_null(ntta, true);
8617 nttb = must_be_not_null(nttb, true);
8618 zetas = must_be_not_null(zetas, true);
8619
8620 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8621 assert(result_start, "result is null");
8622 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8623 assert(ntta_start, "ntta is null");
8624 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8625 assert(nttb_start, "nttb is null");
8626 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8627 assert(zetas_start, "nttMult_zetas is null");
8628 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8629 OptoRuntime::kyberNttMult_Type(),
8630 stubAddr, stubName, TypePtr::BOTTOM,
8631 result_start, ntta_start, nttb_start,
8632 zetas_start);
8633
8634 // return an int
8635 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8636 set_result(retvalue);
8637
8638 return true;
8639 }
8640
8641 //------------------------------inline_kyberAddPoly_2
8642 bool LibraryCallKit::inline_kyberAddPoly_2() {
8643 address stubAddr;
8644 const char *stubName;
8645 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8646 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8647
8648 stubAddr = StubRoutines::kyberAddPoly_2();
8649 stubName = "kyberAddPoly_2";
8650 if (!stubAddr) return false;
8651
8652 Node* result = argument(0);
8653 Node* a = argument(1);
8654 Node* b = argument(2);
8655
8656 result = must_be_not_null(result, true);
8657 a = must_be_not_null(a, true);
8658 b = must_be_not_null(b, true);
8659
8660 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8661 assert(result_start, "result is null");
8662 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8663 assert(a_start, "a is null");
8664 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8665 assert(b_start, "b is null");
8666 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8667 OptoRuntime::kyberAddPoly_2_Type(),
8668 stubAddr, stubName, TypePtr::BOTTOM,
8669 result_start, a_start, b_start);
8670 // return an int
8671 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8672 set_result(retvalue);
8673 return true;
8674 }
8675
8676 //------------------------------inline_kyberAddPoly_3
8677 bool LibraryCallKit::inline_kyberAddPoly_3() {
8678 address stubAddr;
8679 const char *stubName;
8680 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8681 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8682
8683 stubAddr = StubRoutines::kyberAddPoly_3();
8684 stubName = "kyberAddPoly_3";
8685 if (!stubAddr) return false;
8686
8687 Node* result = argument(0);
8688 Node* a = argument(1);
8689 Node* b = argument(2);
8690 Node* c = argument(3);
8691
8692 result = must_be_not_null(result, true);
8693 a = must_be_not_null(a, true);
8694 b = must_be_not_null(b, true);
8695 c = must_be_not_null(c, true);
8696
8697 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8698 assert(result_start, "result is null");
8699 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8700 assert(a_start, "a is null");
8701 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8702 assert(b_start, "b is null");
8703 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8704 assert(c_start, "c is null");
8705 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8706 OptoRuntime::kyberAddPoly_3_Type(),
8707 stubAddr, stubName, TypePtr::BOTTOM,
8708 result_start, a_start, b_start, c_start);
8709 // return an int
8710 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8711 set_result(retvalue);
8712 return true;
8713 }
8714
8715 //------------------------------inline_kyber12To16
8716 bool LibraryCallKit::inline_kyber12To16() {
8717 address stubAddr;
8718 const char *stubName;
8719 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8720 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8721
8722 stubAddr = StubRoutines::kyber12To16();
8723 stubName = "kyber12To16";
8724 if (!stubAddr) return false;
8725
8726 Node* condensed = argument(0);
8727 Node* condensedOffs = argument(1);
8728 Node* parsed = argument(2);
8729 Node* parsedLength = argument(3);
8730
8731 condensed = must_be_not_null(condensed, true);
8732 parsed = must_be_not_null(parsed, true);
8733
8734 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8735 assert(condensed_start, "condensed is null");
8736 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8737 assert(parsed_start, "parsed is null");
8738 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8739 OptoRuntime::kyber12To16_Type(),
8740 stubAddr, stubName, TypePtr::BOTTOM,
8741 condensed_start, condensedOffs, parsed_start, parsedLength);
8742 // return an int
8743 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8744 set_result(retvalue);
8745 return true;
8746
8747 }
8748
8749 //------------------------------inline_kyberBarrettReduce
8750 bool LibraryCallKit::inline_kyberBarrettReduce() {
8751 address stubAddr;
8752 const char *stubName;
8753 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8754 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8755
8756 stubAddr = StubRoutines::kyberBarrettReduce();
8757 stubName = "kyberBarrettReduce";
8758 if (!stubAddr) return false;
8759
8760 Node* coeffs = argument(0);
8761
8762 coeffs = must_be_not_null(coeffs, true);
8763
8764 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8765 assert(coeffs_start, "coeffs is null");
8766 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8767 OptoRuntime::kyberBarrettReduce_Type(),
8768 stubAddr, stubName, TypePtr::BOTTOM,
8769 coeffs_start);
8770 // return an int
8771 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8772 set_result(retvalue);
8773 return true;
8774 }
8775
8776 //------------------------------inline_dilithiumAlmostNtt
8777 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8778 address stubAddr;
8779 const char *stubName;
8780 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8781 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8782
8783 stubAddr = StubRoutines::dilithiumAlmostNtt();
8784 stubName = "dilithiumAlmostNtt";
8785 if (!stubAddr) return false;
8786
8787 Node* coeffs = argument(0);
8788 Node* ntt_zetas = argument(1);
8789
8790 coeffs = must_be_not_null(coeffs, true);
8791 ntt_zetas = must_be_not_null(ntt_zetas, true);
8792
8793 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8794 assert(coeffs_start, "coeffs is null");
8795 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8796 assert(ntt_zetas_start, "ntt_zetas is null");
8797 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8798 OptoRuntime::dilithiumAlmostNtt_Type(),
8799 stubAddr, stubName, TypePtr::BOTTOM,
8800 coeffs_start, ntt_zetas_start);
8801 // return an int
8802 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8803 set_result(retvalue);
8804 return true;
8805 }
8806
8807 //------------------------------inline_dilithiumAlmostInverseNtt
8808 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8809 address stubAddr;
8810 const char *stubName;
8811 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8812 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8813
8814 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8815 stubName = "dilithiumAlmostInverseNtt";
8816 if (!stubAddr) return false;
8817
8818 Node* coeffs = argument(0);
8819 Node* zetas = argument(1);
8820
8821 coeffs = must_be_not_null(coeffs, true);
8822 zetas = must_be_not_null(zetas, true);
8823
8824 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8825 assert(coeffs_start, "coeffs is null");
8826 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8827 assert(zetas_start, "inverseNtt_zetas is null");
8828 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8829 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8830 stubAddr, stubName, TypePtr::BOTTOM,
8831 coeffs_start, zetas_start);
8832 // return an int
8833 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8834 set_result(retvalue);
8835 return true;
8836 }
8837
8838 //------------------------------inline_dilithiumNttMult
8839 bool LibraryCallKit::inline_dilithiumNttMult() {
8840 address stubAddr;
8841 const char *stubName;
8842 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8843 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8844
8845 stubAddr = StubRoutines::dilithiumNttMult();
8846 stubName = "dilithiumNttMult";
8847 if (!stubAddr) return false;
8848
8849 Node* result = argument(0);
8850 Node* ntta = argument(1);
8851 Node* nttb = argument(2);
8852 Node* zetas = argument(3);
8853
8854 result = must_be_not_null(result, true);
8855 ntta = must_be_not_null(ntta, true);
8856 nttb = must_be_not_null(nttb, true);
8857 zetas = must_be_not_null(zetas, true);
8858
8859 Node* result_start = array_element_address(result, intcon(0), T_INT);
8860 assert(result_start, "result is null");
8861 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8862 assert(ntta_start, "ntta is null");
8863 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8864 assert(nttb_start, "nttb is null");
8865 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8866 OptoRuntime::dilithiumNttMult_Type(),
8867 stubAddr, stubName, TypePtr::BOTTOM,
8868 result_start, ntta_start, nttb_start);
8869
8870 // return an int
8871 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8872 set_result(retvalue);
8873
8874 return true;
8875 }
8876
8877 //------------------------------inline_dilithiumMontMulByConstant
8878 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8879 address stubAddr;
8880 const char *stubName;
8881 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8882 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8883
8884 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8885 stubName = "dilithiumMontMulByConstant";
8886 if (!stubAddr) return false;
8887
8888 Node* coeffs = argument(0);
8889 Node* constant = argument(1);
8890
8891 coeffs = must_be_not_null(coeffs, true);
8892
8893 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8894 assert(coeffs_start, "coeffs is null");
8895 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8896 OptoRuntime::dilithiumMontMulByConstant_Type(),
8897 stubAddr, stubName, TypePtr::BOTTOM,
8898 coeffs_start, constant);
8899
8900 // return an int
8901 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8902 set_result(retvalue);
8903 return true;
8904 }
8905
8906
8907 //------------------------------inline_dilithiumDecomposePoly
8908 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8909 address stubAddr;
8910 const char *stubName;
8911 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8912 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8913
8914 stubAddr = StubRoutines::dilithiumDecomposePoly();
8915 stubName = "dilithiumDecomposePoly";
8916 if (!stubAddr) return false;
8917
8918 Node* input = argument(0);
8919 Node* lowPart = argument(1);
8920 Node* highPart = argument(2);
8921 Node* twoGamma2 = argument(3);
8922 Node* multiplier = argument(4);
8923
8924 input = must_be_not_null(input, true);
8925 lowPart = must_be_not_null(lowPart, true);
8926 highPart = must_be_not_null(highPart, true);
8927
8928 Node* input_start = array_element_address(input, intcon(0), T_INT);
8929 assert(input_start, "input is null");
8930 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8931 assert(lowPart_start, "lowPart is null");
8932 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8933 assert(highPart_start, "highPart is null");
8934
8935 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8936 OptoRuntime::dilithiumDecomposePoly_Type(),
8937 stubAddr, stubName, TypePtr::BOTTOM,
8938 input_start, lowPart_start, highPart_start,
8939 twoGamma2, multiplier);
8940
8941 // return an int
8942 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8943 set_result(retvalue);
8944 return true;
8945 }
8946
8947 bool LibraryCallKit::inline_base64_encodeBlock() {
8948 address stubAddr;
8949 const char *stubName;
8950 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8951 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8952 stubAddr = StubRoutines::base64_encodeBlock();
8953 stubName = "encodeBlock";
8954
8955 if (!stubAddr) return false;
8956 Node* base64obj = argument(0);
8957 Node* src = argument(1);
8958 Node* offset = argument(2);
8959 Node* len = argument(3);
8960 Node* dest = argument(4);
8961 Node* dp = argument(5);
8962 Node* isURL = argument(6);
8963
8964 src = must_be_not_null(src, true);
8965 dest = must_be_not_null(dest, true);
8966
8967 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8968 assert(src_start, "source array is null");
8969 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8970 assert(dest_start, "destination array is null");
8971
8972 Node* base64 = make_runtime_call(RC_LEAF,
8973 OptoRuntime::base64_encodeBlock_Type(),
8974 stubAddr, stubName, TypePtr::BOTTOM,
8975 src_start, offset, len, dest_start, dp, isURL);
8976 return true;
8977 }
8978
8979 bool LibraryCallKit::inline_base64_decodeBlock() {
8980 address stubAddr;
8981 const char *stubName;
8982 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8983 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8984 stubAddr = StubRoutines::base64_decodeBlock();
8985 stubName = "decodeBlock";
8986
8987 if (!stubAddr) return false;
8988 Node* base64obj = argument(0);
8989 Node* src = argument(1);
8990 Node* src_offset = argument(2);
8991 Node* len = argument(3);
8992 Node* dest = argument(4);
8993 Node* dest_offset = argument(5);
8994 Node* isURL = argument(6);
8995 Node* isMIME = argument(7);
8996
8997 src = must_be_not_null(src, true);
8998 dest = must_be_not_null(dest, true);
8999
9000 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
9001 assert(src_start, "source array is null");
9002 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
9003 assert(dest_start, "destination array is null");
9004
9005 Node* call = make_runtime_call(RC_LEAF,
9006 OptoRuntime::base64_decodeBlock_Type(),
9007 stubAddr, stubName, TypePtr::BOTTOM,
9008 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
9009 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9010 set_result(result);
9011 return true;
9012 }
9013
9014 bool LibraryCallKit::inline_poly1305_processBlocks() {
9015 address stubAddr;
9016 const char *stubName;
9017 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
9018 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
9019 stubAddr = StubRoutines::poly1305_processBlocks();
9020 stubName = "poly1305_processBlocks";
9021
9022 if (!stubAddr) return false;
9023 null_check_receiver(); // null-check receiver
9024 if (stopped()) return true;
9025
9026 Node* input = argument(1);
9027 Node* input_offset = argument(2);
9028 Node* len = argument(3);
9029 Node* alimbs = argument(4);
9030 Node* rlimbs = argument(5);
9031
9032 input = must_be_not_null(input, true);
9033 alimbs = must_be_not_null(alimbs, true);
9034 rlimbs = must_be_not_null(rlimbs, true);
9035
9036 Node* input_start = array_element_address(input, input_offset, T_BYTE);
9037 assert(input_start, "input array is null");
9038 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
9039 assert(acc_start, "acc array is null");
9040 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
9041 assert(r_start, "r array is null");
9042
9043 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9044 OptoRuntime::poly1305_processBlocks_Type(),
9045 stubAddr, stubName, TypePtr::BOTTOM,
9046 input_start, len, acc_start, r_start);
9047 return true;
9048 }
9049
9050 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
9051 address stubAddr;
9052 const char *stubName;
9053 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9054 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
9055 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
9056 stubName = "intpoly_montgomeryMult_P256";
9057
9058 if (!stubAddr) return false;
9059 null_check_receiver(); // null-check receiver
9060 if (stopped()) return true;
9061
9062 Node* a = argument(1);
9063 Node* b = argument(2);
9064 Node* r = argument(3);
9065
9066 a = must_be_not_null(a, true);
9067 b = must_be_not_null(b, true);
9068 r = must_be_not_null(r, true);
9069
9070 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9071 assert(a_start, "a array is null");
9072 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9073 assert(b_start, "b array is null");
9074 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9075 assert(r_start, "r array is null");
9076
9077 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9078 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9079 stubAddr, stubName, TypePtr::BOTTOM,
9080 a_start, b_start, r_start);
9081 return true;
9082 }
9083
9084 bool LibraryCallKit::inline_intpoly_assign() {
9085 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9086 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9087 const char *stubName = "intpoly_assign";
9088 address stubAddr = StubRoutines::intpoly_assign();
9089 if (!stubAddr) return false;
9090
9091 Node* set = argument(0);
9092 Node* a = argument(1);
9093 Node* b = argument(2);
9094 Node* arr_length = load_array_length(a);
9095
9096 a = must_be_not_null(a, true);
9097 b = must_be_not_null(b, true);
9098
9099 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9100 assert(a_start, "a array is null");
9101 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9102 assert(b_start, "b array is null");
9103
9104 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9105 OptoRuntime::intpoly_assign_Type(),
9106 stubAddr, stubName, TypePtr::BOTTOM,
9107 set, a_start, b_start, arr_length);
9108 return true;
9109 }
9110
9111 //------------------------------inline_digestBase_implCompress-----------------------
9112 //
9113 // Calculate MD5 for single-block byte[] array.
9114 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9115 //
9116 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9117 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9118 //
9119 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9120 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9121 //
9122 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9123 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9124 //
9125 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9126 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9127 //
9128 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9129 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9130
9131 Node* digestBase_obj = argument(0);
9132 Node* src = argument(1); // type oop
9133 Node* ofs = argument(2); // type int
9134
9135 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9136 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9137 // failed array check
9138 return false;
9139 }
9140 // Figure out the size and type of the elements we will be copying.
9141 BasicType src_elem = src_type->elem()->array_element_basic_type();
9142 if (src_elem != T_BYTE) {
9143 return false;
9144 }
9145 // 'src_start' points to src array + offset
9146 src = must_be_not_null(src, true);
9147 Node* src_start = array_element_address(src, ofs, src_elem);
9148 Node* state = nullptr;
9149 Node* block_size = nullptr;
9150 address stubAddr;
9151 const char *stubName;
9152
9153 switch(id) {
9154 case vmIntrinsics::_md5_implCompress:
9155 assert(UseMD5Intrinsics, "need MD5 instruction support");
9156 state = get_state_from_digest_object(digestBase_obj, T_INT);
9157 stubAddr = StubRoutines::md5_implCompress();
9158 stubName = "md5_implCompress";
9159 break;
9160 case vmIntrinsics::_sha_implCompress:
9161 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9162 state = get_state_from_digest_object(digestBase_obj, T_INT);
9163 stubAddr = StubRoutines::sha1_implCompress();
9164 stubName = "sha1_implCompress";
9165 break;
9166 case vmIntrinsics::_sha2_implCompress:
9167 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9168 state = get_state_from_digest_object(digestBase_obj, T_INT);
9169 stubAddr = StubRoutines::sha256_implCompress();
9170 stubName = "sha256_implCompress";
9171 break;
9172 case vmIntrinsics::_sha5_implCompress:
9173 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9174 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9175 stubAddr = StubRoutines::sha512_implCompress();
9176 stubName = "sha512_implCompress";
9177 break;
9178 case vmIntrinsics::_sha3_implCompress:
9179 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9180 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9181 stubAddr = StubRoutines::sha3_implCompress();
9182 stubName = "sha3_implCompress";
9183 block_size = get_block_size_from_digest_object(digestBase_obj);
9184 if (block_size == nullptr) return false;
9185 break;
9186 default:
9187 fatal_unexpected_iid(id);
9188 return false;
9189 }
9190 if (state == nullptr) return false;
9191
9192 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9193 if (stubAddr == nullptr) return false;
9194
9195 // Call the stub.
9196 Node* call;
9197 if (block_size == nullptr) {
9198 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9199 stubAddr, stubName, TypePtr::BOTTOM,
9200 src_start, state);
9201 } else {
9202 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9203 stubAddr, stubName, TypePtr::BOTTOM,
9204 src_start, state, block_size);
9205 }
9206
9207 return true;
9208 }
9209
9210 //------------------------------inline_double_keccak
9211 bool LibraryCallKit::inline_double_keccak() {
9212 address stubAddr;
9213 const char *stubName;
9214 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9215 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9216
9217 stubAddr = StubRoutines::double_keccak();
9218 stubName = "double_keccak";
9219 if (!stubAddr) return false;
9220
9221 Node* status0 = argument(0);
9222 Node* status1 = argument(1);
9223
9224 status0 = must_be_not_null(status0, true);
9225 status1 = must_be_not_null(status1, true);
9226
9227 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9228 assert(status0_start, "status0 is null");
9229 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9230 assert(status1_start, "status1 is null");
9231 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9232 OptoRuntime::double_keccak_Type(),
9233 stubAddr, stubName, TypePtr::BOTTOM,
9234 status0_start, status1_start);
9235 // return an int
9236 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9237 set_result(retvalue);
9238 return true;
9239 }
9240
9241
9242 //------------------------------inline_digestBase_implCompressMB-----------------------
9243 //
9244 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9245 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9246 //
9247 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9248 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9249 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9250 assert((uint)predicate < 5, "sanity");
9251 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9252
9253 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9254 Node* src = argument(1); // byte[] array
9255 Node* ofs = argument(2); // type int
9256 Node* limit = argument(3); // type int
9257
9258 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9259 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9260 // failed array check
9261 return false;
9262 }
9263 // Figure out the size and type of the elements we will be copying.
9264 BasicType src_elem = src_type->elem()->array_element_basic_type();
9265 if (src_elem != T_BYTE) {
9266 return false;
9267 }
9268 // 'src_start' points to src array + offset
9269 src = must_be_not_null(src, false);
9270 Node* src_start = array_element_address(src, ofs, src_elem);
9271
9272 const char* klass_digestBase_name = nullptr;
9273 const char* stub_name = nullptr;
9274 address stub_addr = nullptr;
9275 BasicType elem_type = T_INT;
9276
9277 switch (predicate) {
9278 case 0:
9279 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9280 klass_digestBase_name = "sun/security/provider/MD5";
9281 stub_name = "md5_implCompressMB";
9282 stub_addr = StubRoutines::md5_implCompressMB();
9283 }
9284 break;
9285 case 1:
9286 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9287 klass_digestBase_name = "sun/security/provider/SHA";
9288 stub_name = "sha1_implCompressMB";
9289 stub_addr = StubRoutines::sha1_implCompressMB();
9290 }
9291 break;
9292 case 2:
9293 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9294 klass_digestBase_name = "sun/security/provider/SHA2";
9295 stub_name = "sha256_implCompressMB";
9296 stub_addr = StubRoutines::sha256_implCompressMB();
9297 }
9298 break;
9299 case 3:
9300 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9301 klass_digestBase_name = "sun/security/provider/SHA5";
9302 stub_name = "sha512_implCompressMB";
9303 stub_addr = StubRoutines::sha512_implCompressMB();
9304 elem_type = T_LONG;
9305 }
9306 break;
9307 case 4:
9308 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9309 klass_digestBase_name = "sun/security/provider/SHA3";
9310 stub_name = "sha3_implCompressMB";
9311 stub_addr = StubRoutines::sha3_implCompressMB();
9312 elem_type = T_LONG;
9313 }
9314 break;
9315 default:
9316 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9317 }
9318 if (klass_digestBase_name != nullptr) {
9319 assert(stub_addr != nullptr, "Stub is generated");
9320 if (stub_addr == nullptr) return false;
9321
9322 // get DigestBase klass to lookup for SHA klass
9323 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9324 assert(tinst != nullptr, "digestBase_obj is not instance???");
9325 assert(tinst->is_loaded(), "DigestBase is not loaded");
9326
9327 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9328 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9329 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9330 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9331 }
9332 return false;
9333 }
9334
9335 //------------------------------inline_digestBase_implCompressMB-----------------------
9336 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9337 BasicType elem_type, address stubAddr, const char *stubName,
9338 Node* src_start, Node* ofs, Node* limit) {
9339 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9340 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9341 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9342 digest_obj = _gvn.transform(digest_obj);
9343
9344 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9345 if (state == nullptr) return false;
9346
9347 Node* block_size = nullptr;
9348 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9349 block_size = get_block_size_from_digest_object(digest_obj);
9350 if (block_size == nullptr) return false;
9351 }
9352
9353 // Call the stub.
9354 Node* call;
9355 if (block_size == nullptr) {
9356 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9357 OptoRuntime::digestBase_implCompressMB_Type(false),
9358 stubAddr, stubName, TypePtr::BOTTOM,
9359 src_start, state, ofs, limit);
9360 } else {
9361 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9362 OptoRuntime::digestBase_implCompressMB_Type(true),
9363 stubAddr, stubName, TypePtr::BOTTOM,
9364 src_start, state, block_size, ofs, limit);
9365 }
9366
9367 // return ofs (int)
9368 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9369 set_result(result);
9370
9371 return true;
9372 }
9373
9374 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9375 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9376 assert(UseAES, "need AES instruction support");
9377 address stubAddr = nullptr;
9378 const char *stubName = nullptr;
9379 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9380 stubName = "galoisCounterMode_AESCrypt";
9381
9382 if (stubAddr == nullptr) return false;
9383
9384 Node* in = argument(0);
9385 Node* inOfs = argument(1);
9386 Node* len = argument(2);
9387 Node* ct = argument(3);
9388 Node* ctOfs = argument(4);
9389 Node* out = argument(5);
9390 Node* outOfs = argument(6);
9391 Node* gctr_object = argument(7);
9392 Node* ghash_object = argument(8);
9393
9394 // (1) in, ct and out are arrays.
9395 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9396 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9397 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9398 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9399 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9400 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9401
9402 // checks are the responsibility of the caller
9403 Node* in_start = in;
9404 Node* ct_start = ct;
9405 Node* out_start = out;
9406 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9407 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9408 in_start = array_element_address(in, inOfs, T_BYTE);
9409 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9410 out_start = array_element_address(out, outOfs, T_BYTE);
9411 }
9412
9413 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9414 // (because of the predicated logic executed earlier).
9415 // so we cast it here safely.
9416 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9417 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9418 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9419 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9420 Node* state = load_field_from_object(ghash_object, "state", "[J");
9421
9422 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9423 return false;
9424 }
9425 // cast it to what we know it will be at runtime
9426 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9427 assert(tinst != nullptr, "GCTR obj is null");
9428 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9429 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9430 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9431 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9432 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9433 const TypeOopPtr* xtype = aklass->as_instance_type();
9434 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9435 aescrypt_object = _gvn.transform(aescrypt_object);
9436 // we need to get the start of the aescrypt_object's expanded key array
9437 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9438 if (k_start == nullptr) return false;
9439 // similarly, get the start address of the r vector
9440 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9441 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9442 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9443
9444
9445 // Call the stub, passing params
9446 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9447 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9448 stubAddr, stubName, TypePtr::BOTTOM,
9449 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9450
9451 // return cipher length (int)
9452 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9453 set_result(retvalue);
9454
9455 return true;
9456 }
9457
9458 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9459 // Return node representing slow path of predicate check.
9460 // the pseudo code we want to emulate with this predicate is:
9461 // for encryption:
9462 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9463 // for decryption:
9464 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9465 // note cipher==plain is more conservative than the original java code but that's OK
9466 //
9467
9468 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9469 // The receiver was checked for null already.
9470 Node* objGCTR = argument(7);
9471 // Load embeddedCipher field of GCTR object.
9472 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9473 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9474
9475 // get AESCrypt klass for instanceOf check
9476 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9477 // will have same classloader as CipherBlockChaining object
9478 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9479 assert(tinst != nullptr, "GCTR obj is null");
9480 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9481
9482 // we want to do an instanceof comparison against the AESCrypt class
9483 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9484 if (!klass_AESCrypt->is_loaded()) {
9485 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9486 Node* ctrl = control();
9487 set_control(top()); // no regular fast path
9488 return ctrl;
9489 }
9490
9491 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9492 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9493 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9494 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9495 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9496
9497 return instof_false; // even if it is null
9498 }
9499
9500 //------------------------------get_state_from_digest_object-----------------------
9501 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9502 const char* state_type;
9503 switch (elem_type) {
9504 case T_BYTE: state_type = "[B"; break;
9505 case T_INT: state_type = "[I"; break;
9506 case T_LONG: state_type = "[J"; break;
9507 default: ShouldNotReachHere();
9508 }
9509 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9510 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9511 if (digest_state == nullptr) return (Node *) nullptr;
9512
9513 // now have the array, need to get the start address of the state array
9514 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9515 return state;
9516 }
9517
9518 //------------------------------get_block_size_from_sha3_object----------------------------------
9519 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9520 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9521 assert (block_size != nullptr, "sanity");
9522 return block_size;
9523 }
9524
9525 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9526 // Return node representing slow path of predicate check.
9527 // the pseudo code we want to emulate with this predicate is:
9528 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9529 //
9530 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9531 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9532 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9533 assert((uint)predicate < 5, "sanity");
9534
9535 // The receiver was checked for null already.
9536 Node* digestBaseObj = argument(0);
9537
9538 // get DigestBase klass for instanceOf check
9539 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9540 assert(tinst != nullptr, "digestBaseObj is null");
9541 assert(tinst->is_loaded(), "DigestBase is not loaded");
9542
9543 const char* klass_name = nullptr;
9544 switch (predicate) {
9545 case 0:
9546 if (UseMD5Intrinsics) {
9547 // we want to do an instanceof comparison against the MD5 class
9548 klass_name = "sun/security/provider/MD5";
9549 }
9550 break;
9551 case 1:
9552 if (UseSHA1Intrinsics) {
9553 // we want to do an instanceof comparison against the SHA class
9554 klass_name = "sun/security/provider/SHA";
9555 }
9556 break;
9557 case 2:
9558 if (UseSHA256Intrinsics) {
9559 // we want to do an instanceof comparison against the SHA2 class
9560 klass_name = "sun/security/provider/SHA2";
9561 }
9562 break;
9563 case 3:
9564 if (UseSHA512Intrinsics) {
9565 // we want to do an instanceof comparison against the SHA5 class
9566 klass_name = "sun/security/provider/SHA5";
9567 }
9568 break;
9569 case 4:
9570 if (UseSHA3Intrinsics) {
9571 // we want to do an instanceof comparison against the SHA3 class
9572 klass_name = "sun/security/provider/SHA3";
9573 }
9574 break;
9575 default:
9576 fatal("unknown SHA intrinsic predicate: %d", predicate);
9577 }
9578
9579 ciKlass* klass = nullptr;
9580 if (klass_name != nullptr) {
9581 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9582 }
9583 if ((klass == nullptr) || !klass->is_loaded()) {
9584 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9585 Node* ctrl = control();
9586 set_control(top()); // no intrinsic path
9587 return ctrl;
9588 }
9589 ciInstanceKlass* instklass = klass->as_instance_klass();
9590
9591 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9592 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9593 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9594 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9595
9596 return instof_false; // even if it is null
9597 }
9598
9599 //-------------inline_fma-----------------------------------
9600 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9601 Node *a = nullptr;
9602 Node *b = nullptr;
9603 Node *c = nullptr;
9604 Node* result = nullptr;
9605 switch (id) {
9606 case vmIntrinsics::_fmaD:
9607 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9608 // no receiver since it is static method
9609 a = argument(0);
9610 b = argument(2);
9611 c = argument(4);
9612 result = _gvn.transform(new FmaDNode(a, b, c));
9613 break;
9614 case vmIntrinsics::_fmaF:
9615 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9616 a = argument(0);
9617 b = argument(1);
9618 c = argument(2);
9619 result = _gvn.transform(new FmaFNode(a, b, c));
9620 break;
9621 default:
9622 fatal_unexpected_iid(id); break;
9623 }
9624 set_result(result);
9625 return true;
9626 }
9627
9628 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9629 // argument(0) is receiver
9630 Node* codePoint = argument(1);
9631 Node* n = nullptr;
9632
9633 switch (id) {
9634 case vmIntrinsics::_isDigit :
9635 n = new DigitNode(control(), codePoint);
9636 break;
9637 case vmIntrinsics::_isLowerCase :
9638 n = new LowerCaseNode(control(), codePoint);
9639 break;
9640 case vmIntrinsics::_isUpperCase :
9641 n = new UpperCaseNode(control(), codePoint);
9642 break;
9643 case vmIntrinsics::_isWhitespace :
9644 n = new WhitespaceNode(control(), codePoint);
9645 break;
9646 default:
9647 fatal_unexpected_iid(id);
9648 }
9649
9650 set_result(_gvn.transform(n));
9651 return true;
9652 }
9653
9654 bool LibraryCallKit::inline_profileBoolean() {
9655 Node* counts = argument(1);
9656 const TypeAryPtr* ary = nullptr;
9657 ciArray* aobj = nullptr;
9658 if (counts->is_Con()
9659 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9660 && (aobj = ary->const_oop()->as_array()) != nullptr
9661 && (aobj->length() == 2)) {
9662 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9663 jint false_cnt = aobj->element_value(0).as_int();
9664 jint true_cnt = aobj->element_value(1).as_int();
9665
9666 if (C->log() != nullptr) {
9667 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9668 false_cnt, true_cnt);
9669 }
9670
9671 if (false_cnt + true_cnt == 0) {
9672 // According to profile, never executed.
9673 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9674 Deoptimization::Action_reinterpret);
9675 return true;
9676 }
9677
9678 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9679 // is a number of each value occurrences.
9680 Node* result = argument(0);
9681 if (false_cnt == 0 || true_cnt == 0) {
9682 // According to profile, one value has been never seen.
9683 int expected_val = (false_cnt == 0) ? 1 : 0;
9684
9685 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9686 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9687
9688 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9689 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9690 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9691
9692 { // Slow path: uncommon trap for never seen value and then reexecute
9693 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9694 // the value has been seen at least once.
9695 PreserveJVMState pjvms(this);
9696 PreserveReexecuteState preexecs(this);
9697 jvms()->set_should_reexecute(true);
9698
9699 set_control(slow_path);
9700 set_i_o(i_o());
9701
9702 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9703 Deoptimization::Action_reinterpret);
9704 }
9705 // The guard for never seen value enables sharpening of the result and
9706 // returning a constant. It allows to eliminate branches on the same value
9707 // later on.
9708 set_control(fast_path);
9709 result = intcon(expected_val);
9710 }
9711 // Stop profiling.
9712 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9713 // By replacing method body with profile data (represented as ProfileBooleanNode
9714 // on IR level) we effectively disable profiling.
9715 // It enables full speed execution once optimized code is generated.
9716 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9717 C->record_for_igvn(profile);
9718 set_result(profile);
9719 return true;
9720 } else {
9721 // Continue profiling.
9722 // Profile data isn't available at the moment. So, execute method's bytecode version.
9723 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9724 // is compiled and counters aren't available since corresponding MethodHandle
9725 // isn't a compile-time constant.
9726 return false;
9727 }
9728 }
9729
9730 bool LibraryCallKit::inline_isCompileConstant() {
9731 Node* n = argument(0);
9732 set_result(n->is_Con() ? intcon(1) : intcon(0));
9733 return true;
9734 }
9735
9736 //------------------------------- inline_getObjectSize --------------------------------------
9737 //
9738 // Calculate the runtime size of the object/array.
9739 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9740 //
9741 bool LibraryCallKit::inline_getObjectSize() {
9742 Node* obj = argument(3);
9743 Node* klass_node = load_object_klass(obj);
9744
9745 jint layout_con = Klass::_lh_neutral_value;
9746 Node* layout_val = get_layout_helper(klass_node, layout_con);
9747 int layout_is_con = (layout_val == nullptr);
9748
9749 if (layout_is_con) {
9750 // Layout helper is constant, can figure out things at compile time.
9751
9752 if (Klass::layout_helper_is_instance(layout_con)) {
9753 // Instance case: layout_con contains the size itself.
9754 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9755 set_result(size);
9756 } else {
9757 // Array case: size is round(header + element_size*arraylength).
9758 // Since arraylength is different for every array instance, we have to
9759 // compute the whole thing at runtime.
9760
9761 Node* arr_length = load_array_length(obj);
9762
9763 int round_mask = MinObjAlignmentInBytes - 1;
9764 int hsize = Klass::layout_helper_header_size(layout_con);
9765 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9766
9767 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9768 round_mask = 0; // strength-reduce it if it goes away completely
9769 }
9770 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9771 Node* header_size = intcon(hsize + round_mask);
9772
9773 Node* lengthx = ConvI2X(arr_length);
9774 Node* headerx = ConvI2X(header_size);
9775
9776 Node* abody = lengthx;
9777 if (eshift != 0) {
9778 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9779 }
9780 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9781 if (round_mask != 0) {
9782 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9783 }
9784 size = ConvX2L(size);
9785 set_result(size);
9786 }
9787 } else {
9788 // Layout helper is not constant, need to test for array-ness at runtime.
9789
9790 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9791 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9792 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9793 record_for_igvn(result_reg);
9794
9795 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9796 if (array_ctl != nullptr) {
9797 // Array case: size is round(header + element_size*arraylength).
9798 // Since arraylength is different for every array instance, we have to
9799 // compute the whole thing at runtime.
9800
9801 PreserveJVMState pjvms(this);
9802 set_control(array_ctl);
9803 Node* arr_length = load_array_length(obj);
9804
9805 int round_mask = MinObjAlignmentInBytes - 1;
9806 Node* mask = intcon(round_mask);
9807
9808 Node* hss = intcon(Klass::_lh_header_size_shift);
9809 Node* hsm = intcon(Klass::_lh_header_size_mask);
9810 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9811 header_size = _gvn.transform(new AndINode(header_size, hsm));
9812 header_size = _gvn.transform(new AddINode(header_size, mask));
9813
9814 // There is no need to mask or shift this value.
9815 // The semantics of LShiftINode include an implicit mask to 0x1F.
9816 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9817 Node* elem_shift = layout_val;
9818
9819 Node* lengthx = ConvI2X(arr_length);
9820 Node* headerx = ConvI2X(header_size);
9821
9822 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9823 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9824 if (round_mask != 0) {
9825 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9826 }
9827 size = ConvX2L(size);
9828
9829 result_reg->init_req(_array_path, control());
9830 result_val->init_req(_array_path, size);
9831 }
9832
9833 if (!stopped()) {
9834 // Instance case: the layout helper gives us instance size almost directly,
9835 // but we need to mask out the _lh_instance_slow_path_bit.
9836 Node* size = ConvI2X(layout_val);
9837 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9838 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9839 size = _gvn.transform(new AndXNode(size, mask));
9840 size = ConvX2L(size);
9841
9842 result_reg->init_req(_instance_path, control());
9843 result_val->init_req(_instance_path, size);
9844 }
9845
9846 set_result(result_reg, result_val);
9847 }
9848
9849 return true;
9850 }
9851
9852 //------------------------------- inline_blackhole --------------------------------------
9853 //
9854 // Make sure all arguments to this node are alive.
9855 // This matches methods that were requested to be blackholed through compile commands.
9856 //
9857 bool LibraryCallKit::inline_blackhole() {
9858 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9859 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9860 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9861
9862 // Blackhole node pinches only the control, not memory. This allows
9863 // the blackhole to be pinned in the loop that computes blackholed
9864 // values, but have no other side effects, like breaking the optimizations
9865 // across the blackhole.
9866
9867 Node* bh = _gvn.transform(new BlackholeNode(control()));
9868 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9869
9870 // Bind call arguments as blackhole arguments to keep them alive
9871 uint nargs = callee()->arg_size();
9872 for (uint i = 0; i < nargs; i++) {
9873 bh->add_req(argument(i));
9874 }
9875
9876 return true;
9877 }
9878
9879 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9880 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9881 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9882 return nullptr; // box klass is not Float16
9883 }
9884
9885 // Null check; get notnull casted pointer
9886 Node* null_ctl = top();
9887 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9888 // If not_null_box is dead, only null-path is taken
9889 if (stopped()) {
9890 set_control(null_ctl);
9891 return nullptr;
9892 }
9893 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9894 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9895 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9896 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9897 }
9898
9899 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9900 PreserveReexecuteState preexecs(this);
9901 jvms()->set_should_reexecute(true);
9902
9903 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9904 Node* klass_node = makecon(klass_type);
9905 Node* box = new_instance(klass_node);
9906
9907 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9908 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9909
9910 Node* field_store = _gvn.transform(access_store_at(box,
9911 value_field,
9912 value_adr_type,
9913 value,
9914 TypeInt::SHORT,
9915 T_SHORT,
9916 IN_HEAP));
9917 set_memory(field_store, value_adr_type);
9918 return box;
9919 }
9920
9921 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9922 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9923 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9924 return false;
9925 }
9926
9927 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9928 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9929 return false;
9930 }
9931
9932 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9933 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9934 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9935 ciSymbols::short_signature(),
9936 false);
9937 assert(field != nullptr, "");
9938
9939 // Transformed nodes
9940 Node* fld1 = nullptr;
9941 Node* fld2 = nullptr;
9942 Node* fld3 = nullptr;
9943 switch(num_args) {
9944 case 3:
9945 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9946 if (fld3 == nullptr) {
9947 return false;
9948 }
9949 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9950 // fall-through
9951 case 2:
9952 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9953 if (fld2 == nullptr) {
9954 return false;
9955 }
9956 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9957 // fall-through
9958 case 1:
9959 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9960 if (fld1 == nullptr) {
9961 return false;
9962 }
9963 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9964 break;
9965 default: fatal("Unsupported number of arguments %d", num_args);
9966 }
9967
9968 Node* result = nullptr;
9969 switch (id) {
9970 // Unary operations
9971 case vmIntrinsics::_sqrt_float16:
9972 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9973 break;
9974 // Ternary operations
9975 case vmIntrinsics::_fma_float16:
9976 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9977 break;
9978 default:
9979 fatal_unexpected_iid(id);
9980 break;
9981 }
9982 result = _gvn.transform(new ReinterpretHF2SNode(result));
9983 set_result(box_fp16_value(float16_box_type, field, result));
9984 return true;
9985 }
9986