1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciFlatArrayKlass.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "ci/ciSymbols.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/accessDecorators.hpp"
36 #include "oops/klass.inline.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "opto/addnode.hpp"
39 #include "opto/arraycopynode.hpp"
40 #include "opto/c2compiler.hpp"
41 #include "opto/castnode.hpp"
42 #include "opto/cfgnode.hpp"
43 #include "opto/convertnode.hpp"
44 #include "opto/countbitsnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/library_call.hpp"
47 #include "opto/mathexactnode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "opto/vectornode.hpp"
56 #include "prims/jvmtiExport.hpp"
57 #include "prims/jvmtiThreadState.hpp"
58 #include "prims/unsafe.hpp"
59 #include "runtime/jniHandles.inline.hpp"
60 #include "runtime/objectMonitor.hpp"
61 #include "runtime/sharedRuntime.hpp"
62 #include "runtime/stubRoutines.hpp"
63 #include "utilities/macros.hpp"
64 #include "utilities/powerOfTwo.hpp"
65
66 //---------------------------make_vm_intrinsic----------------------------
67 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
68 vmIntrinsicID id = m->intrinsic_id();
69 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
70
71 if (!m->is_loaded()) {
72 // Do not attempt to inline unloaded methods.
73 return nullptr;
74 }
75
76 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
77 bool is_available = false;
78
79 {
80 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
81 // the compiler must transition to '_thread_in_vm' state because both
82 // methods access VM-internal data.
83 VM_ENTRY_MARK;
84 methodHandle mh(THREAD, m->get_Method());
85 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
86 if (is_available && is_virtual) {
87 is_available = vmIntrinsics::does_virtual_dispatch(id);
88 }
89 }
90
91 if (is_available) {
92 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
93 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
94 return new LibraryIntrinsic(m, is_virtual,
95 vmIntrinsics::predicates_needed(id),
96 vmIntrinsics::does_virtual_dispatch(id),
97 id);
98 } else {
99 return nullptr;
100 }
101 }
102
103 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
104 LibraryCallKit kit(jvms, this);
105 Compile* C = kit.C;
106 int nodes = C->unique();
107 #ifndef PRODUCT
108 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
109 char buf[1000];
110 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
111 tty->print_cr("Intrinsic %s", str);
112 }
113 #endif
114 ciMethod* callee = kit.callee();
115 const int bci = kit.bci();
116 #ifdef ASSERT
117 Node* ctrl = kit.control();
118 #endif
119 // Try to inline the intrinsic.
120 if (callee->check_intrinsic_candidate() &&
121 kit.try_to_inline(_last_predicate)) {
122 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
123 : "(intrinsic)";
124 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
125 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dtanh:
250 case vmIntrinsics::_dabs:
251 case vmIntrinsics::_fabs:
252 case vmIntrinsics::_iabs:
253 case vmIntrinsics::_labs:
254 case vmIntrinsics::_datan2:
255 case vmIntrinsics::_dsqrt:
256 case vmIntrinsics::_dsqrt_strict:
257 case vmIntrinsics::_dexp:
258 case vmIntrinsics::_dlog:
259 case vmIntrinsics::_dlog10:
260 case vmIntrinsics::_dpow:
261 case vmIntrinsics::_dcopySign:
262 case vmIntrinsics::_fcopySign:
263 case vmIntrinsics::_dsignum:
264 case vmIntrinsics::_roundF:
265 case vmIntrinsics::_roundD:
266 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
267
268 case vmIntrinsics::_notify:
269 case vmIntrinsics::_notifyAll:
270 return inline_notify(intrinsic_id());
271
272 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
273 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
274 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
275 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
276 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
277 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
278 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
279 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
280 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
281 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
282 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
283 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
284 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
285 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
286
287 case vmIntrinsics::_arraycopy: return inline_arraycopy();
288
289 case vmIntrinsics::_arraySort: return inline_array_sort();
290 case vmIntrinsics::_arrayPartition: return inline_array_partition();
291
292 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
293 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
294 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
295 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
296
297 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
298 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
299 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
300 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
301 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
302 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
303 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
304 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
305
306 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
307
308 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
309
310 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
311 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
312 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
313 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
314
315 case vmIntrinsics::_compressStringC:
316 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
317 case vmIntrinsics::_inflateStringC:
318 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
319
320 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
321 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
331 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
332
333 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
334 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
335 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
336 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
337 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
338 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
339 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
340 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
341 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
342 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
343
344 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
345 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
346 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
347 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
348 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
349 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
350 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
351 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
352 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
353
354 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
355 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
356 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
357 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
358 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
359 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
360 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
361 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
362 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
363
364 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
365 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
366 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
367 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
368
369 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
370 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
371 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
372 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
373
374 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
375 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
376 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
377 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
378 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
379 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
380 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
381 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
382 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
383
384 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
385 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
386 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
387 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
388 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
389 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
390 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
391 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
392 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
393
394 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
395 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
396 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
397 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
398 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
399 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
400 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
401 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
402 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
403
404 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
405 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
406 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
407 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
408 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
409 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
410 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
411 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
412 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
413
414 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
417 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
419
420 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
421 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
422 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
423 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
424 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
440
441 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
442 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
443 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
444 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
456
457 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
459 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
460 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
461
462 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
465 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
467
468 case vmIntrinsics::_loadFence:
469 case vmIntrinsics::_storeFence:
470 case vmIntrinsics::_storeStoreFence:
471 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
472
473 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
474
475 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
476 case vmIntrinsics::_currentThread: return inline_native_currentThread();
477 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
478
479 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
480 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
481
482 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
483 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
484
485 #if INCLUDE_JVMTI
486 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
487 "notifyJvmtiStart", true, false);
488 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
489 "notifyJvmtiEnd", false, true);
490 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
491 "notifyJvmtiMount", false, false);
492 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
493 "notifyJvmtiUnmount", false, false);
494 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
495 #endif
496
497 #ifdef JFR_HAVE_INTRINSICS
498 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
499 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
500 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
501 #endif
502 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
503 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
504 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
505 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
506 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
507 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
508 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
509 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
510 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
511 case vmIntrinsics::_getLength: return inline_native_getLength();
512 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
513 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
514 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
515 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
516 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
517 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
518 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
519
520 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
521 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
522 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
523 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
524 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
525
526 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
527
528 case vmIntrinsics::_isInstance:
529 case vmIntrinsics::_isHidden:
530 case vmIntrinsics::_getSuperclass:
531 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
532
533 case vmIntrinsics::_floatToRawIntBits:
534 case vmIntrinsics::_floatToIntBits:
535 case vmIntrinsics::_intBitsToFloat:
536 case vmIntrinsics::_doubleToRawLongBits:
537 case vmIntrinsics::_doubleToLongBits:
538 case vmIntrinsics::_longBitsToDouble:
539 case vmIntrinsics::_floatToFloat16:
540 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
541 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
542 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
543 case vmIntrinsics::_floatIsFinite:
544 case vmIntrinsics::_floatIsInfinite:
545 case vmIntrinsics::_doubleIsFinite:
546 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
547
548 case vmIntrinsics::_numberOfLeadingZeros_i:
549 case vmIntrinsics::_numberOfLeadingZeros_l:
550 case vmIntrinsics::_numberOfTrailingZeros_i:
551 case vmIntrinsics::_numberOfTrailingZeros_l:
552 case vmIntrinsics::_bitCount_i:
553 case vmIntrinsics::_bitCount_l:
554 case vmIntrinsics::_reverse_i:
555 case vmIntrinsics::_reverse_l:
556 case vmIntrinsics::_reverseBytes_i:
557 case vmIntrinsics::_reverseBytes_l:
558 case vmIntrinsics::_reverseBytes_s:
559 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
560
561 case vmIntrinsics::_compress_i:
562 case vmIntrinsics::_compress_l:
563 case vmIntrinsics::_expand_i:
564 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
565
566 case vmIntrinsics::_compareUnsigned_i:
567 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
568
569 case vmIntrinsics::_divideUnsigned_i:
570 case vmIntrinsics::_divideUnsigned_l:
571 case vmIntrinsics::_remainderUnsigned_i:
572 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
573
574 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
575
576 case vmIntrinsics::_Reference_get: return inline_reference_get();
577 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
578 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
579 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
580 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
581
582 case vmIntrinsics::_Class_cast: return inline_Class_cast();
583
584 case vmIntrinsics::_aescrypt_encryptBlock:
585 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
586
587 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
588 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
589 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
590
591 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
592 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
593 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
594
595 case vmIntrinsics::_counterMode_AESCrypt:
596 return inline_counterMode_AESCrypt(intrinsic_id());
597
598 case vmIntrinsics::_galoisCounterMode_AESCrypt:
599 return inline_galoisCounterMode_AESCrypt();
600
601 case vmIntrinsics::_md5_implCompress:
602 case vmIntrinsics::_sha_implCompress:
603 case vmIntrinsics::_sha2_implCompress:
604 case vmIntrinsics::_sha5_implCompress:
605 case vmIntrinsics::_sha3_implCompress:
606 return inline_digestBase_implCompress(intrinsic_id());
607 case vmIntrinsics::_double_keccak:
608 return inline_double_keccak();
609
610 case vmIntrinsics::_digestBase_implCompressMB:
611 return inline_digestBase_implCompressMB(predicate);
612
613 case vmIntrinsics::_multiplyToLen:
614 return inline_multiplyToLen();
615
616 case vmIntrinsics::_squareToLen:
617 return inline_squareToLen();
618
619 case vmIntrinsics::_mulAdd:
620 return inline_mulAdd();
621
622 case vmIntrinsics::_montgomeryMultiply:
623 return inline_montgomeryMultiply();
624 case vmIntrinsics::_montgomerySquare:
625 return inline_montgomerySquare();
626
627 case vmIntrinsics::_bigIntegerRightShiftWorker:
628 return inline_bigIntegerShift(true);
629 case vmIntrinsics::_bigIntegerLeftShiftWorker:
630 return inline_bigIntegerShift(false);
631
632 case vmIntrinsics::_vectorizedMismatch:
633 return inline_vectorizedMismatch();
634
635 case vmIntrinsics::_ghash_processBlocks:
636 return inline_ghash_processBlocks();
637 case vmIntrinsics::_chacha20Block:
638 return inline_chacha20Block();
639 case vmIntrinsics::_kyberNtt:
640 return inline_kyberNtt();
641 case vmIntrinsics::_kyberInverseNtt:
642 return inline_kyberInverseNtt();
643 case vmIntrinsics::_kyberNttMult:
644 return inline_kyberNttMult();
645 case vmIntrinsics::_kyberAddPoly_2:
646 return inline_kyberAddPoly_2();
647 case vmIntrinsics::_kyberAddPoly_3:
648 return inline_kyberAddPoly_3();
649 case vmIntrinsics::_kyber12To16:
650 return inline_kyber12To16();
651 case vmIntrinsics::_kyberBarrettReduce:
652 return inline_kyberBarrettReduce();
653 case vmIntrinsics::_dilithiumAlmostNtt:
654 return inline_dilithiumAlmostNtt();
655 case vmIntrinsics::_dilithiumAlmostInverseNtt:
656 return inline_dilithiumAlmostInverseNtt();
657 case vmIntrinsics::_dilithiumNttMult:
658 return inline_dilithiumNttMult();
659 case vmIntrinsics::_dilithiumMontMulByConstant:
660 return inline_dilithiumMontMulByConstant();
661 case vmIntrinsics::_dilithiumDecomposePoly:
662 return inline_dilithiumDecomposePoly();
663 case vmIntrinsics::_base64_encodeBlock:
664 return inline_base64_encodeBlock();
665 case vmIntrinsics::_base64_decodeBlock:
666 return inline_base64_decodeBlock();
667 case vmIntrinsics::_poly1305_processBlocks:
668 return inline_poly1305_processBlocks();
669 case vmIntrinsics::_intpoly_montgomeryMult_P256:
670 return inline_intpoly_montgomeryMult_P256();
671 case vmIntrinsics::_intpoly_assign:
672 return inline_intpoly_assign();
673 case vmIntrinsics::_encodeISOArray:
674 case vmIntrinsics::_encodeByteISOArray:
675 return inline_encodeISOArray(false);
676 case vmIntrinsics::_encodeAsciiArray:
677 return inline_encodeISOArray(true);
678
679 case vmIntrinsics::_updateCRC32:
680 return inline_updateCRC32();
681 case vmIntrinsics::_updateBytesCRC32:
682 return inline_updateBytesCRC32();
683 case vmIntrinsics::_updateByteBufferCRC32:
684 return inline_updateByteBufferCRC32();
685
686 case vmIntrinsics::_updateBytesCRC32C:
687 return inline_updateBytesCRC32C();
688 case vmIntrinsics::_updateDirectByteBufferCRC32C:
689 return inline_updateDirectByteBufferCRC32C();
690
691 case vmIntrinsics::_updateBytesAdler32:
692 return inline_updateBytesAdler32();
693 case vmIntrinsics::_updateByteBufferAdler32:
694 return inline_updateByteBufferAdler32();
695
696 case vmIntrinsics::_profileBoolean:
697 return inline_profileBoolean();
698 case vmIntrinsics::_isCompileConstant:
699 return inline_isCompileConstant();
700
701 case vmIntrinsics::_countPositives:
702 return inline_countPositives();
703
704 case vmIntrinsics::_fmaD:
705 case vmIntrinsics::_fmaF:
706 return inline_fma(intrinsic_id());
707
708 case vmIntrinsics::_isDigit:
709 case vmIntrinsics::_isLowerCase:
710 case vmIntrinsics::_isUpperCase:
711 case vmIntrinsics::_isWhitespace:
712 return inline_character_compare(intrinsic_id());
713
714 case vmIntrinsics::_min:
715 case vmIntrinsics::_max:
716 case vmIntrinsics::_min_strict:
717 case vmIntrinsics::_max_strict:
718 case vmIntrinsics::_minL:
719 case vmIntrinsics::_maxL:
720 case vmIntrinsics::_minF:
721 case vmIntrinsics::_maxF:
722 case vmIntrinsics::_minD:
723 case vmIntrinsics::_maxD:
724 case vmIntrinsics::_minF_strict:
725 case vmIntrinsics::_maxF_strict:
726 case vmIntrinsics::_minD_strict:
727 case vmIntrinsics::_maxD_strict:
728 return inline_min_max(intrinsic_id());
729
730 case vmIntrinsics::_VectorUnaryOp:
731 return inline_vector_nary_operation(1);
732 case vmIntrinsics::_VectorBinaryOp:
733 return inline_vector_nary_operation(2);
734 case vmIntrinsics::_VectorTernaryOp:
735 return inline_vector_nary_operation(3);
736 case vmIntrinsics::_VectorFromBitsCoerced:
737 return inline_vector_frombits_coerced();
738 case vmIntrinsics::_VectorMaskOp:
739 return inline_vector_mask_operation();
740 case vmIntrinsics::_VectorLoadOp:
741 return inline_vector_mem_operation(/*is_store=*/false);
742 case vmIntrinsics::_VectorLoadMaskedOp:
743 return inline_vector_mem_masked_operation(/*is_store*/false);
744 case vmIntrinsics::_VectorStoreOp:
745 return inline_vector_mem_operation(/*is_store=*/true);
746 case vmIntrinsics::_VectorStoreMaskedOp:
747 return inline_vector_mem_masked_operation(/*is_store=*/true);
748 case vmIntrinsics::_VectorGatherOp:
749 return inline_vector_gather_scatter(/*is_scatter*/ false);
750 case vmIntrinsics::_VectorScatterOp:
751 return inline_vector_gather_scatter(/*is_scatter*/ true);
752 case vmIntrinsics::_VectorReductionCoerced:
753 return inline_vector_reduction();
754 case vmIntrinsics::_VectorTest:
755 return inline_vector_test();
756 case vmIntrinsics::_VectorBlend:
757 return inline_vector_blend();
758 case vmIntrinsics::_VectorRearrange:
759 return inline_vector_rearrange();
760 case vmIntrinsics::_VectorSelectFrom:
761 return inline_vector_select_from();
762 case vmIntrinsics::_VectorCompare:
763 return inline_vector_compare();
764 case vmIntrinsics::_VectorBroadcastInt:
765 return inline_vector_broadcast_int();
766 case vmIntrinsics::_VectorConvert:
767 return inline_vector_convert();
768 case vmIntrinsics::_VectorInsert:
769 return inline_vector_insert();
770 case vmIntrinsics::_VectorExtract:
771 return inline_vector_extract();
772 case vmIntrinsics::_VectorCompressExpand:
773 return inline_vector_compress_expand();
774 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
775 return inline_vector_select_from_two_vectors();
776 case vmIntrinsics::_IndexVector:
777 return inline_index_vector();
778 case vmIntrinsics::_IndexPartiallyInUpperRange:
779 return inline_index_partially_in_upper_range();
780
781 case vmIntrinsics::_getObjectSize:
782 return inline_getObjectSize();
783
784 case vmIntrinsics::_blackhole:
785 return inline_blackhole();
786
787 default:
788 // If you get here, it may be that someone has added a new intrinsic
789 // to the list in vmIntrinsics.hpp without implementing it here.
790 #ifndef PRODUCT
791 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
792 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
793 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
794 }
795 #endif
796 return false;
797 }
798 }
799
800 Node* LibraryCallKit::try_to_predicate(int predicate) {
801 if (!jvms()->has_method()) {
802 // Root JVMState has a null method.
803 assert(map()->memory()->Opcode() == Op_Parm, "");
804 // Insert the memory aliasing node
805 set_all_memory(reset_memory());
806 }
807 assert(merged_memory(), "");
808
809 switch (intrinsic_id()) {
810 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
811 return inline_cipherBlockChaining_AESCrypt_predicate(false);
812 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
813 return inline_cipherBlockChaining_AESCrypt_predicate(true);
814 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
815 return inline_electronicCodeBook_AESCrypt_predicate(false);
816 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
817 return inline_electronicCodeBook_AESCrypt_predicate(true);
818 case vmIntrinsics::_counterMode_AESCrypt:
819 return inline_counterMode_AESCrypt_predicate();
820 case vmIntrinsics::_digestBase_implCompressMB:
821 return inline_digestBase_implCompressMB_predicate(predicate);
822 case vmIntrinsics::_galoisCounterMode_AESCrypt:
823 return inline_galoisCounterMode_AESCrypt_predicate();
824
825 default:
826 // If you get here, it may be that someone has added a new intrinsic
827 // to the list in vmIntrinsics.hpp without implementing it here.
828 #ifndef PRODUCT
829 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
830 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
831 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
832 }
833 #endif
834 Node* slow_ctl = control();
835 set_control(top()); // No fast path intrinsic
836 return slow_ctl;
837 }
838 }
839
840 //------------------------------set_result-------------------------------
841 // Helper function for finishing intrinsics.
842 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
843 record_for_igvn(region);
844 set_control(_gvn.transform(region));
845 set_result( _gvn.transform(value));
846 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
847 }
848
849 //------------------------------generate_guard---------------------------
850 // Helper function for generating guarded fast-slow graph structures.
851 // The given 'test', if true, guards a slow path. If the test fails
852 // then a fast path can be taken. (We generally hope it fails.)
853 // In all cases, GraphKit::control() is updated to the fast path.
854 // The returned value represents the control for the slow path.
855 // The return value is never 'top'; it is either a valid control
856 // or null if it is obvious that the slow path can never be taken.
857 // Also, if region and the slow control are not null, the slow edge
858 // is appended to the region.
859 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
860 if (stopped()) {
861 // Already short circuited.
862 return nullptr;
863 }
864
865 // Build an if node and its projections.
866 // If test is true we take the slow path, which we assume is uncommon.
867 if (_gvn.type(test) == TypeInt::ZERO) {
868 // The slow branch is never taken. No need to build this guard.
869 return nullptr;
870 }
871
872 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
873
874 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
875 if (if_slow == top()) {
876 // The slow branch is never taken. No need to build this guard.
877 return nullptr;
878 }
879
880 if (region != nullptr)
881 region->add_req(if_slow);
882
883 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
884 set_control(if_fast);
885
886 return if_slow;
887 }
888
889 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
890 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
891 }
892 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
893 return generate_guard(test, region, PROB_FAIR);
894 }
895
896 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
897 Node* *pos_index) {
898 if (stopped())
899 return nullptr; // already stopped
900 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
901 return nullptr; // index is already adequately typed
902 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
903 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
904 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
905 if (is_neg != nullptr && pos_index != nullptr) {
906 // Emulate effect of Parse::adjust_map_after_if.
907 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
908 (*pos_index) = _gvn.transform(ccast);
909 }
910 return is_neg;
911 }
912
913 // Make sure that 'position' is a valid limit index, in [0..length].
914 // There are two equivalent plans for checking this:
915 // A. (offset + copyLength) unsigned<= arrayLength
916 // B. offset <= (arrayLength - copyLength)
917 // We require that all of the values above, except for the sum and
918 // difference, are already known to be non-negative.
919 // Plan A is robust in the face of overflow, if offset and copyLength
920 // are both hugely positive.
921 //
922 // Plan B is less direct and intuitive, but it does not overflow at
923 // all, since the difference of two non-negatives is always
924 // representable. Whenever Java methods must perform the equivalent
925 // check they generally use Plan B instead of Plan A.
926 // For the moment we use Plan A.
927 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
928 Node* subseq_length,
929 Node* array_length,
930 RegionNode* region) {
931 if (stopped())
932 return nullptr; // already stopped
933 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
934 if (zero_offset && subseq_length->eqv_uncast(array_length))
935 return nullptr; // common case of whole-array copy
936 Node* last = subseq_length;
937 if (!zero_offset) // last += offset
938 last = _gvn.transform(new AddINode(last, offset));
939 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
940 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
941 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
942 return is_over;
943 }
944
945 // Emit range checks for the given String.value byte array
946 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
947 if (stopped()) {
948 return; // already stopped
949 }
950 RegionNode* bailout = new RegionNode(1);
951 record_for_igvn(bailout);
952 if (char_count) {
953 // Convert char count to byte count
954 count = _gvn.transform(new LShiftINode(count, intcon(1)));
955 }
956
957 // Offset and count must not be negative
958 generate_negative_guard(offset, bailout);
959 generate_negative_guard(count, bailout);
960 // Offset + count must not exceed length of array
961 generate_limit_guard(offset, count, load_array_length(array), bailout);
962
963 if (bailout->req() > 1) {
964 PreserveJVMState pjvms(this);
965 set_control(_gvn.transform(bailout));
966 uncommon_trap(Deoptimization::Reason_intrinsic,
967 Deoptimization::Action_maybe_recompile);
968 }
969 }
970
971 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
972 bool is_immutable) {
973 ciKlass* thread_klass = env()->Thread_klass();
974 const Type* thread_type
975 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
976
977 Node* thread = _gvn.transform(new ThreadLocalNode());
978 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
979 tls_output = thread;
980
981 Node* thread_obj_handle
982 = (is_immutable
983 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
984 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
985 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
986 thread_obj_handle = _gvn.transform(thread_obj_handle);
987
988 DecoratorSet decorators = IN_NATIVE;
989 if (is_immutable) {
990 decorators |= C2_IMMUTABLE_MEMORY;
991 }
992 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
993 }
994
995 //--------------------------generate_current_thread--------------------
996 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
997 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
998 /*is_immutable*/false);
999 }
1000
1001 //--------------------------generate_virtual_thread--------------------
1002 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1003 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1004 !C->method()->changes_current_thread());
1005 }
1006
1007 //------------------------------make_string_method_node------------------------
1008 // Helper method for String intrinsic functions. This version is called with
1009 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1010 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1011 // containing the lengths of str1 and str2.
1012 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1013 Node* result = nullptr;
1014 switch (opcode) {
1015 case Op_StrIndexOf:
1016 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1017 str1_start, cnt1, str2_start, cnt2, ae);
1018 break;
1019 case Op_StrComp:
1020 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1021 str1_start, cnt1, str2_start, cnt2, ae);
1022 break;
1023 case Op_StrEquals:
1024 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1025 // Use the constant length if there is one because optimized match rule may exist.
1026 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1027 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1028 break;
1029 default:
1030 ShouldNotReachHere();
1031 return nullptr;
1032 }
1033
1034 // All these intrinsics have checks.
1035 C->set_has_split_ifs(true); // Has chance for split-if optimization
1036 clear_upper_avx();
1037
1038 return _gvn.transform(result);
1039 }
1040
1041 //------------------------------inline_string_compareTo------------------------
1042 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1043 Node* arg1 = argument(0);
1044 Node* arg2 = argument(1);
1045
1046 arg1 = must_be_not_null(arg1, true);
1047 arg2 = must_be_not_null(arg2, true);
1048
1049 // Get start addr and length of first argument
1050 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1051 Node* arg1_cnt = load_array_length(arg1);
1052
1053 // Get start addr and length of second argument
1054 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1055 Node* arg2_cnt = load_array_length(arg2);
1056
1057 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1058 set_result(result);
1059 return true;
1060 }
1061
1062 //------------------------------inline_string_equals------------------------
1063 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1064 Node* arg1 = argument(0);
1065 Node* arg2 = argument(1);
1066
1067 // paths (plus control) merge
1068 RegionNode* region = new RegionNode(3);
1069 Node* phi = new PhiNode(region, TypeInt::BOOL);
1070
1071 if (!stopped()) {
1072
1073 arg1 = must_be_not_null(arg1, true);
1074 arg2 = must_be_not_null(arg2, true);
1075
1076 // Get start addr and length of first argument
1077 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1078 Node* arg1_cnt = load_array_length(arg1);
1079
1080 // Get start addr and length of second argument
1081 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1082 Node* arg2_cnt = load_array_length(arg2);
1083
1084 // Check for arg1_cnt != arg2_cnt
1085 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1086 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1087 Node* if_ne = generate_slow_guard(bol, nullptr);
1088 if (if_ne != nullptr) {
1089 phi->init_req(2, intcon(0));
1090 region->init_req(2, if_ne);
1091 }
1092
1093 // Check for count == 0 is done by assembler code for StrEquals.
1094
1095 if (!stopped()) {
1096 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1097 phi->init_req(1, equals);
1098 region->init_req(1, control());
1099 }
1100 }
1101
1102 // post merge
1103 set_control(_gvn.transform(region));
1104 record_for_igvn(region);
1105
1106 set_result(_gvn.transform(phi));
1107 return true;
1108 }
1109
1110 //------------------------------inline_array_equals----------------------------
1111 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1112 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1113 Node* arg1 = argument(0);
1114 Node* arg2 = argument(1);
1115
1116 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1117 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1118 clear_upper_avx();
1119
1120 return true;
1121 }
1122
1123
1124 //------------------------------inline_countPositives------------------------------
1125 bool LibraryCallKit::inline_countPositives() {
1126 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1127 return false;
1128 }
1129
1130 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1131 // no receiver since it is static method
1132 Node* ba = argument(0);
1133 Node* offset = argument(1);
1134 Node* len = argument(2);
1135
1136 ba = must_be_not_null(ba, true);
1137
1138 // Range checks
1139 generate_string_range_check(ba, offset, len, false);
1140 if (stopped()) {
1141 return true;
1142 }
1143 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1144 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1145 set_result(_gvn.transform(result));
1146 clear_upper_avx();
1147 return true;
1148 }
1149
1150 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1151 Node* index = argument(0);
1152 Node* length = bt == T_INT ? argument(1) : argument(2);
1153 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1154 return false;
1155 }
1156
1157 // check that length is positive
1158 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1159 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1160
1161 {
1162 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1163 uncommon_trap(Deoptimization::Reason_intrinsic,
1164 Deoptimization::Action_make_not_entrant);
1165 }
1166
1167 if (stopped()) {
1168 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1169 return true;
1170 }
1171
1172 // length is now known positive, add a cast node to make this explicit
1173 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1174 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1175 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1176 ConstraintCastNode::RegularDependency, bt);
1177 casted_length = _gvn.transform(casted_length);
1178 replace_in_map(length, casted_length);
1179 length = casted_length;
1180
1181 // Use an unsigned comparison for the range check itself
1182 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1183 BoolTest::mask btest = BoolTest::lt;
1184 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1185 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1186 _gvn.set_type(rc, rc->Value(&_gvn));
1187 if (!rc_bool->is_Con()) {
1188 record_for_igvn(rc);
1189 }
1190 set_control(_gvn.transform(new IfTrueNode(rc)));
1191 {
1192 PreserveJVMState pjvms(this);
1193 set_control(_gvn.transform(new IfFalseNode(rc)));
1194 uncommon_trap(Deoptimization::Reason_range_check,
1195 Deoptimization::Action_make_not_entrant);
1196 }
1197
1198 if (stopped()) {
1199 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1200 return true;
1201 }
1202
1203 // index is now known to be >= 0 and < length, cast it
1204 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1205 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1206 ConstraintCastNode::RegularDependency, bt);
1207 result = _gvn.transform(result);
1208 set_result(result);
1209 replace_in_map(index, result);
1210 return true;
1211 }
1212
1213 //------------------------------inline_string_indexOf------------------------
1214 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1215 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1216 return false;
1217 }
1218 Node* src = argument(0);
1219 Node* tgt = argument(1);
1220
1221 // Make the merge point
1222 RegionNode* result_rgn = new RegionNode(4);
1223 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1224
1225 src = must_be_not_null(src, true);
1226 tgt = must_be_not_null(tgt, true);
1227
1228 // Get start addr and length of source string
1229 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1230 Node* src_count = load_array_length(src);
1231
1232 // Get start addr and length of substring
1233 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1234 Node* tgt_count = load_array_length(tgt);
1235
1236 Node* result = nullptr;
1237 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1238
1239 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1240 // Divide src size by 2 if String is UTF16 encoded
1241 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1242 }
1243 if (ae == StrIntrinsicNode::UU) {
1244 // Divide substring size by 2 if String is UTF16 encoded
1245 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1246 }
1247
1248 if (call_opt_stub) {
1249 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1250 StubRoutines::_string_indexof_array[ae],
1251 "stringIndexOf", TypePtr::BOTTOM, src_start,
1252 src_count, tgt_start, tgt_count);
1253 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1254 } else {
1255 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1256 result_rgn, result_phi, ae);
1257 }
1258 if (result != nullptr) {
1259 result_phi->init_req(3, result);
1260 result_rgn->init_req(3, control());
1261 }
1262 set_control(_gvn.transform(result_rgn));
1263 record_for_igvn(result_rgn);
1264 set_result(_gvn.transform(result_phi));
1265
1266 return true;
1267 }
1268
1269 //-----------------------------inline_string_indexOfI-----------------------
1270 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1271 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1272 return false;
1273 }
1274 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1275 return false;
1276 }
1277
1278 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1279 Node* src = argument(0); // byte[]
1280 Node* src_count = argument(1); // char count
1281 Node* tgt = argument(2); // byte[]
1282 Node* tgt_count = argument(3); // char count
1283 Node* from_index = argument(4); // char index
1284
1285 src = must_be_not_null(src, true);
1286 tgt = must_be_not_null(tgt, true);
1287
1288 // Multiply byte array index by 2 if String is UTF16 encoded
1289 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1290 src_count = _gvn.transform(new SubINode(src_count, from_index));
1291 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1292 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1293
1294 // Range checks
1295 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1296 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1297 if (stopped()) {
1298 return true;
1299 }
1300
1301 RegionNode* region = new RegionNode(5);
1302 Node* phi = new PhiNode(region, TypeInt::INT);
1303 Node* result = nullptr;
1304
1305 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1306
1307 if (call_opt_stub) {
1308 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1309 StubRoutines::_string_indexof_array[ae],
1310 "stringIndexOf", TypePtr::BOTTOM, src_start,
1311 src_count, tgt_start, tgt_count);
1312 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1313 } else {
1314 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1315 region, phi, ae);
1316 }
1317 if (result != nullptr) {
1318 // The result is index relative to from_index if substring was found, -1 otherwise.
1319 // Generate code which will fold into cmove.
1320 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1321 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1322
1323 Node* if_lt = generate_slow_guard(bol, nullptr);
1324 if (if_lt != nullptr) {
1325 // result == -1
1326 phi->init_req(3, result);
1327 region->init_req(3, if_lt);
1328 }
1329 if (!stopped()) {
1330 result = _gvn.transform(new AddINode(result, from_index));
1331 phi->init_req(4, result);
1332 region->init_req(4, control());
1333 }
1334 }
1335
1336 set_control(_gvn.transform(region));
1337 record_for_igvn(region);
1338 set_result(_gvn.transform(phi));
1339 clear_upper_avx();
1340
1341 return true;
1342 }
1343
1344 // Create StrIndexOfNode with fast path checks
1345 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1346 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1347 // Check for substr count > string count
1348 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1349 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1350 Node* if_gt = generate_slow_guard(bol, nullptr);
1351 if (if_gt != nullptr) {
1352 phi->init_req(1, intcon(-1));
1353 region->init_req(1, if_gt);
1354 }
1355 if (!stopped()) {
1356 // Check for substr count == 0
1357 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1358 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1359 Node* if_zero = generate_slow_guard(bol, nullptr);
1360 if (if_zero != nullptr) {
1361 phi->init_req(2, intcon(0));
1362 region->init_req(2, if_zero);
1363 }
1364 }
1365 if (!stopped()) {
1366 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1367 }
1368 return nullptr;
1369 }
1370
1371 //-----------------------------inline_string_indexOfChar-----------------------
1372 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1373 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1374 return false;
1375 }
1376 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1377 return false;
1378 }
1379 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1380 Node* src = argument(0); // byte[]
1381 Node* int_ch = argument(1);
1382 Node* from_index = argument(2);
1383 Node* max = argument(3);
1384
1385 src = must_be_not_null(src, true);
1386
1387 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1388 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1389 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1390
1391 // Range checks
1392 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1393
1394 // Check for int_ch >= 0
1395 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1396 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1397 {
1398 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1399 uncommon_trap(Deoptimization::Reason_intrinsic,
1400 Deoptimization::Action_maybe_recompile);
1401 }
1402 if (stopped()) {
1403 return true;
1404 }
1405
1406 RegionNode* region = new RegionNode(3);
1407 Node* phi = new PhiNode(region, TypeInt::INT);
1408
1409 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1410 C->set_has_split_ifs(true); // Has chance for split-if optimization
1411 _gvn.transform(result);
1412
1413 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1414 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1415
1416 Node* if_lt = generate_slow_guard(bol, nullptr);
1417 if (if_lt != nullptr) {
1418 // result == -1
1419 phi->init_req(2, result);
1420 region->init_req(2, if_lt);
1421 }
1422 if (!stopped()) {
1423 result = _gvn.transform(new AddINode(result, from_index));
1424 phi->init_req(1, result);
1425 region->init_req(1, control());
1426 }
1427 set_control(_gvn.transform(region));
1428 record_for_igvn(region);
1429 set_result(_gvn.transform(phi));
1430 clear_upper_avx();
1431
1432 return true;
1433 }
1434 //---------------------------inline_string_copy---------------------
1435 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1436 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1437 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1438 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1439 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1440 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1441 bool LibraryCallKit::inline_string_copy(bool compress) {
1442 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1443 return false;
1444 }
1445 int nargs = 5; // 2 oops, 3 ints
1446 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1447
1448 Node* src = argument(0);
1449 Node* src_offset = argument(1);
1450 Node* dst = argument(2);
1451 Node* dst_offset = argument(3);
1452 Node* length = argument(4);
1453
1454 // Check for allocation before we add nodes that would confuse
1455 // tightly_coupled_allocation()
1456 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1457
1458 // Figure out the size and type of the elements we will be copying.
1459 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1460 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1461 if (src_type == nullptr || dst_type == nullptr) {
1462 return false;
1463 }
1464 BasicType src_elem = src_type->elem()->array_element_basic_type();
1465 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1466 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1467 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1468 "Unsupported array types for inline_string_copy");
1469
1470 src = must_be_not_null(src, true);
1471 dst = must_be_not_null(dst, true);
1472
1473 // Convert char[] offsets to byte[] offsets
1474 bool convert_src = (compress && src_elem == T_BYTE);
1475 bool convert_dst = (!compress && dst_elem == T_BYTE);
1476 if (convert_src) {
1477 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1478 } else if (convert_dst) {
1479 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1480 }
1481
1482 // Range checks
1483 generate_string_range_check(src, src_offset, length, convert_src);
1484 generate_string_range_check(dst, dst_offset, length, convert_dst);
1485 if (stopped()) {
1486 return true;
1487 }
1488
1489 Node* src_start = array_element_address(src, src_offset, src_elem);
1490 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1491 // 'src_start' points to src array + scaled offset
1492 // 'dst_start' points to dst array + scaled offset
1493 Node* count = nullptr;
1494 if (compress) {
1495 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1496 } else {
1497 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1498 }
1499
1500 if (alloc != nullptr) {
1501 if (alloc->maybe_set_complete(&_gvn)) {
1502 // "You break it, you buy it."
1503 InitializeNode* init = alloc->initialization();
1504 assert(init->is_complete(), "we just did this");
1505 init->set_complete_with_arraycopy();
1506 assert(dst->is_CheckCastPP(), "sanity");
1507 assert(dst->in(0)->in(0) == init, "dest pinned");
1508 }
1509 // Do not let stores that initialize this object be reordered with
1510 // a subsequent store that would make this object accessible by
1511 // other threads.
1512 // Record what AllocateNode this StoreStore protects so that
1513 // escape analysis can go from the MemBarStoreStoreNode to the
1514 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1515 // based on the escape status of the AllocateNode.
1516 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1517 }
1518 if (compress) {
1519 set_result(_gvn.transform(count));
1520 }
1521 clear_upper_avx();
1522
1523 return true;
1524 }
1525
1526 #ifdef _LP64
1527 #define XTOP ,top() /*additional argument*/
1528 #else //_LP64
1529 #define XTOP /*no additional argument*/
1530 #endif //_LP64
1531
1532 //------------------------inline_string_toBytesU--------------------------
1533 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1534 bool LibraryCallKit::inline_string_toBytesU() {
1535 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1536 return false;
1537 }
1538 // Get the arguments.
1539 Node* value = argument(0);
1540 Node* offset = argument(1);
1541 Node* length = argument(2);
1542
1543 Node* newcopy = nullptr;
1544
1545 // Set the original stack and the reexecute bit for the interpreter to reexecute
1546 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1547 { PreserveReexecuteState preexecs(this);
1548 jvms()->set_should_reexecute(true);
1549
1550 // Check if a null path was taken unconditionally.
1551 value = null_check(value);
1552
1553 RegionNode* bailout = new RegionNode(1);
1554 record_for_igvn(bailout);
1555
1556 // Range checks
1557 generate_negative_guard(offset, bailout);
1558 generate_negative_guard(length, bailout);
1559 generate_limit_guard(offset, length, load_array_length(value), bailout);
1560 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1561 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1562
1563 if (bailout->req() > 1) {
1564 PreserveJVMState pjvms(this);
1565 set_control(_gvn.transform(bailout));
1566 uncommon_trap(Deoptimization::Reason_intrinsic,
1567 Deoptimization::Action_maybe_recompile);
1568 }
1569 if (stopped()) {
1570 return true;
1571 }
1572
1573 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1574 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1575 newcopy = new_array(klass_node, size, 0); // no arguments to push
1576 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1577 guarantee(alloc != nullptr, "created above");
1578
1579 // Calculate starting addresses.
1580 Node* src_start = array_element_address(value, offset, T_CHAR);
1581 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1582
1583 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1584 const TypeInt* toffset = gvn().type(offset)->is_int();
1585 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1586
1587 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1588 const char* copyfunc_name = "arraycopy";
1589 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1590 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1591 OptoRuntime::fast_arraycopy_Type(),
1592 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1593 src_start, dst_start, ConvI2X(length) XTOP);
1594 // Do not let reads from the cloned object float above the arraycopy.
1595 if (alloc->maybe_set_complete(&_gvn)) {
1596 // "You break it, you buy it."
1597 InitializeNode* init = alloc->initialization();
1598 assert(init->is_complete(), "we just did this");
1599 init->set_complete_with_arraycopy();
1600 assert(newcopy->is_CheckCastPP(), "sanity");
1601 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1602 }
1603 // Do not let stores that initialize this object be reordered with
1604 // a subsequent store that would make this object accessible by
1605 // other threads.
1606 // Record what AllocateNode this StoreStore protects so that
1607 // escape analysis can go from the MemBarStoreStoreNode to the
1608 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1609 // based on the escape status of the AllocateNode.
1610 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1611 } // original reexecute is set back here
1612
1613 C->set_has_split_ifs(true); // Has chance for split-if optimization
1614 if (!stopped()) {
1615 set_result(newcopy);
1616 }
1617 clear_upper_avx();
1618
1619 return true;
1620 }
1621
1622 //------------------------inline_string_getCharsU--------------------------
1623 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1624 bool LibraryCallKit::inline_string_getCharsU() {
1625 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1626 return false;
1627 }
1628
1629 // Get the arguments.
1630 Node* src = argument(0);
1631 Node* src_begin = argument(1);
1632 Node* src_end = argument(2); // exclusive offset (i < src_end)
1633 Node* dst = argument(3);
1634 Node* dst_begin = argument(4);
1635
1636 // Check for allocation before we add nodes that would confuse
1637 // tightly_coupled_allocation()
1638 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1639
1640 // Check if a null path was taken unconditionally.
1641 src = null_check(src);
1642 dst = null_check(dst);
1643 if (stopped()) {
1644 return true;
1645 }
1646
1647 // Get length and convert char[] offset to byte[] offset
1648 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1649 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1650
1651 // Range checks
1652 generate_string_range_check(src, src_begin, length, true);
1653 generate_string_range_check(dst, dst_begin, length, false);
1654 if (stopped()) {
1655 return true;
1656 }
1657
1658 if (!stopped()) {
1659 // Calculate starting addresses.
1660 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1661 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1662
1663 // Check if array addresses are aligned to HeapWordSize
1664 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1665 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1666 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1667 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1668
1669 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1670 const char* copyfunc_name = "arraycopy";
1671 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1672 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1673 OptoRuntime::fast_arraycopy_Type(),
1674 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1675 src_start, dst_start, ConvI2X(length) XTOP);
1676 // Do not let reads from the cloned object float above the arraycopy.
1677 if (alloc != nullptr) {
1678 if (alloc->maybe_set_complete(&_gvn)) {
1679 // "You break it, you buy it."
1680 InitializeNode* init = alloc->initialization();
1681 assert(init->is_complete(), "we just did this");
1682 init->set_complete_with_arraycopy();
1683 assert(dst->is_CheckCastPP(), "sanity");
1684 assert(dst->in(0)->in(0) == init, "dest pinned");
1685 }
1686 // Do not let stores that initialize this object be reordered with
1687 // a subsequent store that would make this object accessible by
1688 // other threads.
1689 // Record what AllocateNode this StoreStore protects so that
1690 // escape analysis can go from the MemBarStoreStoreNode to the
1691 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1692 // based on the escape status of the AllocateNode.
1693 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1694 } else {
1695 insert_mem_bar(Op_MemBarCPUOrder);
1696 }
1697 }
1698
1699 C->set_has_split_ifs(true); // Has chance for split-if optimization
1700 return true;
1701 }
1702
1703 //----------------------inline_string_char_access----------------------------
1704 // Store/Load char to/from byte[] array.
1705 // static void StringUTF16.putChar(byte[] val, int index, int c)
1706 // static char StringUTF16.getChar(byte[] val, int index)
1707 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1708 Node* value = argument(0);
1709 Node* index = argument(1);
1710 Node* ch = is_store ? argument(2) : nullptr;
1711
1712 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1713 // correctly requires matched array shapes.
1714 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1715 "sanity: byte[] and char[] bases agree");
1716 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1717 "sanity: byte[] and char[] scales agree");
1718
1719 // Bail when getChar over constants is requested: constant folding would
1720 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1721 // Java method would constant fold nicely instead.
1722 if (!is_store && value->is_Con() && index->is_Con()) {
1723 return false;
1724 }
1725
1726 // Save state and restore on bailout
1727 uint old_sp = sp();
1728 SafePointNode* old_map = clone_map();
1729
1730 value = must_be_not_null(value, true);
1731
1732 Node* adr = array_element_address(value, index, T_CHAR);
1733 if (adr->is_top()) {
1734 set_map(old_map);
1735 set_sp(old_sp);
1736 return false;
1737 }
1738 destruct_map_clone(old_map);
1739 if (is_store) {
1740 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1741 } else {
1742 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);
1743 set_result(ch);
1744 }
1745 return true;
1746 }
1747
1748
1749 //------------------------------inline_math-----------------------------------
1750 // public static double Math.abs(double)
1751 // public static double Math.sqrt(double)
1752 // public static double Math.log(double)
1753 // public static double Math.log10(double)
1754 // public static double Math.round(double)
1755 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1756 Node* arg = argument(0);
1757 Node* n = nullptr;
1758 switch (id) {
1759 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1760 case vmIntrinsics::_dsqrt:
1761 case vmIntrinsics::_dsqrt_strict:
1762 n = new SqrtDNode(C, control(), arg); break;
1763 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1764 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1765 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1766 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1767 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1768 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1769 default: fatal_unexpected_iid(id); break;
1770 }
1771 set_result(_gvn.transform(n));
1772 return true;
1773 }
1774
1775 //------------------------------inline_math-----------------------------------
1776 // public static float Math.abs(float)
1777 // public static int Math.abs(int)
1778 // public static long Math.abs(long)
1779 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1780 Node* arg = argument(0);
1781 Node* n = nullptr;
1782 switch (id) {
1783 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1784 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1785 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1786 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1787 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1788 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1789 default: fatal_unexpected_iid(id); break;
1790 }
1791 set_result(_gvn.transform(n));
1792 return true;
1793 }
1794
1795 //------------------------------runtime_math-----------------------------
1796 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1797 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1798 "must be (DD)D or (D)D type");
1799
1800 // Inputs
1801 Node* a = argument(0);
1802 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1803
1804 const TypePtr* no_memory_effects = nullptr;
1805 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1806 no_memory_effects,
1807 a, top(), b, b ? top() : nullptr);
1808 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1809 #ifdef ASSERT
1810 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1811 assert(value_top == top(), "second value must be top");
1812 #endif
1813
1814 set_result(value);
1815 return true;
1816 }
1817
1818 //------------------------------inline_math_pow-----------------------------
1819 bool LibraryCallKit::inline_math_pow() {
1820 Node* exp = argument(2);
1821 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1822 if (d != nullptr) {
1823 if (d->getd() == 2.0) {
1824 // Special case: pow(x, 2.0) => x * x
1825 Node* base = argument(0);
1826 set_result(_gvn.transform(new MulDNode(base, base)));
1827 return true;
1828 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1829 // Special case: pow(x, 0.5) => sqrt(x)
1830 Node* base = argument(0);
1831 Node* zero = _gvn.zerocon(T_DOUBLE);
1832
1833 RegionNode* region = new RegionNode(3);
1834 Node* phi = new PhiNode(region, Type::DOUBLE);
1835
1836 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1837 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1838 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1839 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1840 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1841
1842 Node* if_pow = generate_slow_guard(test, nullptr);
1843 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1844 phi->init_req(1, value_sqrt);
1845 region->init_req(1, control());
1846
1847 if (if_pow != nullptr) {
1848 set_control(if_pow);
1849 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1850 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1851 const TypePtr* no_memory_effects = nullptr;
1852 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1853 no_memory_effects, base, top(), exp, top());
1854 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1855 #ifdef ASSERT
1856 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1857 assert(value_top == top(), "second value must be top");
1858 #endif
1859 phi->init_req(2, value_pow);
1860 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1861 }
1862
1863 C->set_has_split_ifs(true); // Has chance for split-if optimization
1864 set_control(_gvn.transform(region));
1865 record_for_igvn(region);
1866 set_result(_gvn.transform(phi));
1867
1868 return true;
1869 }
1870 }
1871
1872 return StubRoutines::dpow() != nullptr ?
1873 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1874 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1875 }
1876
1877 //------------------------------inline_math_native-----------------------------
1878 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1879 switch (id) {
1880 case vmIntrinsics::_dsin:
1881 return StubRoutines::dsin() != nullptr ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1883 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1884 case vmIntrinsics::_dcos:
1885 return StubRoutines::dcos() != nullptr ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1887 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1888 case vmIntrinsics::_dtan:
1889 return StubRoutines::dtan() != nullptr ?
1890 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1891 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1892 case vmIntrinsics::_dtanh:
1893 return StubRoutines::dtanh() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1895 case vmIntrinsics::_dexp:
1896 return StubRoutines::dexp() != nullptr ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1898 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1899 case vmIntrinsics::_dlog:
1900 return StubRoutines::dlog() != nullptr ?
1901 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1902 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1903 case vmIntrinsics::_dlog10:
1904 return StubRoutines::dlog10() != nullptr ?
1905 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1906 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1907
1908 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1909 case vmIntrinsics::_ceil:
1910 case vmIntrinsics::_floor:
1911 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1912
1913 case vmIntrinsics::_dsqrt:
1914 case vmIntrinsics::_dsqrt_strict:
1915 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1916 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1917 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1918 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1919 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1920
1921 case vmIntrinsics::_dpow: return inline_math_pow();
1922 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1923 case vmIntrinsics::_fcopySign: return inline_math(id);
1924 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1925 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1926 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1927
1928 // These intrinsics are not yet correctly implemented
1929 case vmIntrinsics::_datan2:
1930 return false;
1931
1932 default:
1933 fatal_unexpected_iid(id);
1934 return false;
1935 }
1936 }
1937
1938 //----------------------------inline_notify-----------------------------------*
1939 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1940 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1941 address func;
1942 if (id == vmIntrinsics::_notify) {
1943 func = OptoRuntime::monitor_notify_Java();
1944 } else {
1945 func = OptoRuntime::monitor_notifyAll_Java();
1946 }
1947 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1948 make_slow_call_ex(call, env()->Throwable_klass(), false);
1949 return true;
1950 }
1951
1952
1953 //----------------------------inline_min_max-----------------------------------
1954 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1955 Node* a = nullptr;
1956 Node* b = nullptr;
1957 Node* n = nullptr;
1958 switch (id) {
1959 case vmIntrinsics::_min:
1960 case vmIntrinsics::_max:
1961 case vmIntrinsics::_minF:
1962 case vmIntrinsics::_maxF:
1963 case vmIntrinsics::_minF_strict:
1964 case vmIntrinsics::_maxF_strict:
1965 case vmIntrinsics::_min_strict:
1966 case vmIntrinsics::_max_strict:
1967 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1968 a = argument(0);
1969 b = argument(1);
1970 break;
1971 case vmIntrinsics::_minD:
1972 case vmIntrinsics::_maxD:
1973 case vmIntrinsics::_minD_strict:
1974 case vmIntrinsics::_maxD_strict:
1975 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1976 a = argument(0);
1977 b = argument(2);
1978 break;
1979 case vmIntrinsics::_minL:
1980 case vmIntrinsics::_maxL:
1981 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1982 a = argument(0);
1983 b = argument(2);
1984 break;
1985 default:
1986 fatal_unexpected_iid(id);
1987 break;
1988 }
1989
1990 switch (id) {
1991 case vmIntrinsics::_min:
1992 case vmIntrinsics::_min_strict:
1993 n = new MinINode(a, b);
1994 break;
1995 case vmIntrinsics::_max:
1996 case vmIntrinsics::_max_strict:
1997 n = new MaxINode(a, b);
1998 break;
1999 case vmIntrinsics::_minF:
2000 case vmIntrinsics::_minF_strict:
2001 n = new MinFNode(a, b);
2002 break;
2003 case vmIntrinsics::_maxF:
2004 case vmIntrinsics::_maxF_strict:
2005 n = new MaxFNode(a, b);
2006 break;
2007 case vmIntrinsics::_minD:
2008 case vmIntrinsics::_minD_strict:
2009 n = new MinDNode(a, b);
2010 break;
2011 case vmIntrinsics::_maxD:
2012 case vmIntrinsics::_maxD_strict:
2013 n = new MaxDNode(a, b);
2014 break;
2015 case vmIntrinsics::_minL:
2016 n = new MinLNode(_gvn.C, a, b);
2017 break;
2018 case vmIntrinsics::_maxL:
2019 n = new MaxLNode(_gvn.C, a, b);
2020 break;
2021 default:
2022 fatal_unexpected_iid(id);
2023 break;
2024 }
2025
2026 set_result(_gvn.transform(n));
2027 return true;
2028 }
2029
2030 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2031 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2032 env()->ArithmeticException_instance())) {
2033 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2034 // so let's bail out intrinsic rather than risking deopting again.
2035 return false;
2036 }
2037
2038 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2039 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2040 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2041 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2042
2043 {
2044 PreserveJVMState pjvms(this);
2045 PreserveReexecuteState preexecs(this);
2046 jvms()->set_should_reexecute(true);
2047
2048 set_control(slow_path);
2049 set_i_o(i_o());
2050
2051 builtin_throw(Deoptimization::Reason_intrinsic,
2052 env()->ArithmeticException_instance(),
2053 /*allow_too_many_traps*/ false);
2054 }
2055
2056 set_control(fast_path);
2057 set_result(math);
2058 return true;
2059 }
2060
2061 template <typename OverflowOp>
2062 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2063 typedef typename OverflowOp::MathOp MathOp;
2064
2065 MathOp* mathOp = new MathOp(arg1, arg2);
2066 Node* operation = _gvn.transform( mathOp );
2067 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2068 return inline_math_mathExact(operation, ofcheck);
2069 }
2070
2071 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2072 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2073 }
2074
2075 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2076 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2077 }
2078
2079 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2080 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2081 }
2082
2083 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2084 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2085 }
2086
2087 bool LibraryCallKit::inline_math_negateExactI() {
2088 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2089 }
2090
2091 bool LibraryCallKit::inline_math_negateExactL() {
2092 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2093 }
2094
2095 bool LibraryCallKit::inline_math_multiplyExactI() {
2096 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2097 }
2098
2099 bool LibraryCallKit::inline_math_multiplyExactL() {
2100 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2101 }
2102
2103 bool LibraryCallKit::inline_math_multiplyHigh() {
2104 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2105 return true;
2106 }
2107
2108 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2109 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2110 return true;
2111 }
2112
2113 inline int
2114 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2115 const TypePtr* base_type = TypePtr::NULL_PTR;
2116 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2117 if (base_type == nullptr) {
2118 // Unknown type.
2119 return Type::AnyPtr;
2120 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2121 // Since this is a null+long form, we have to switch to a rawptr.
2122 base = _gvn.transform(new CastX2PNode(offset));
2123 offset = MakeConX(0);
2124 return Type::RawPtr;
2125 } else if (base_type->base() == Type::RawPtr) {
2126 return Type::RawPtr;
2127 } else if (base_type->isa_oopptr()) {
2128 // Base is never null => always a heap address.
2129 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2130 return Type::OopPtr;
2131 }
2132 // Offset is small => always a heap address.
2133 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2134 if (offset_type != nullptr &&
2135 base_type->offset() == 0 && // (should always be?)
2136 offset_type->_lo >= 0 &&
2137 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2138 return Type::OopPtr;
2139 } else if (type == T_OBJECT) {
2140 // off heap access to an oop doesn't make any sense. Has to be on
2141 // heap.
2142 return Type::OopPtr;
2143 }
2144 // Otherwise, it might either be oop+off or null+addr.
2145 return Type::AnyPtr;
2146 } else {
2147 // No information:
2148 return Type::AnyPtr;
2149 }
2150 }
2151
2152 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2153 Node* uncasted_base = base;
2154 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2155 if (kind == Type::RawPtr) {
2156 return basic_plus_adr(top(), uncasted_base, offset);
2157 } else if (kind == Type::AnyPtr) {
2158 assert(base == uncasted_base, "unexpected base change");
2159 if (can_cast) {
2160 if (!_gvn.type(base)->speculative_maybe_null() &&
2161 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2162 // According to profiling, this access is always on
2163 // heap. Casting the base to not null and thus avoiding membars
2164 // around the access should allow better optimizations
2165 Node* null_ctl = top();
2166 base = null_check_oop(base, &null_ctl, true, true, true);
2167 assert(null_ctl->is_top(), "no null control here");
2168 return basic_plus_adr(base, offset);
2169 } else if (_gvn.type(base)->speculative_always_null() &&
2170 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2171 // According to profiling, this access is always off
2172 // heap.
2173 base = null_assert(base);
2174 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2175 offset = MakeConX(0);
2176 return basic_plus_adr(top(), raw_base, offset);
2177 }
2178 }
2179 // We don't know if it's an on heap or off heap access. Fall back
2180 // to raw memory access.
2181 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2182 return basic_plus_adr(top(), raw, offset);
2183 } else {
2184 assert(base == uncasted_base, "unexpected base change");
2185 // We know it's an on heap access so base can't be null
2186 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2187 base = must_be_not_null(base, true);
2188 }
2189 return basic_plus_adr(base, offset);
2190 }
2191 }
2192
2193 //--------------------------inline_number_methods-----------------------------
2194 // inline int Integer.numberOfLeadingZeros(int)
2195 // inline int Long.numberOfLeadingZeros(long)
2196 //
2197 // inline int Integer.numberOfTrailingZeros(int)
2198 // inline int Long.numberOfTrailingZeros(long)
2199 //
2200 // inline int Integer.bitCount(int)
2201 // inline int Long.bitCount(long)
2202 //
2203 // inline char Character.reverseBytes(char)
2204 // inline short Short.reverseBytes(short)
2205 // inline int Integer.reverseBytes(int)
2206 // inline long Long.reverseBytes(long)
2207 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2208 Node* arg = argument(0);
2209 Node* n = nullptr;
2210 switch (id) {
2211 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2212 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2213 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2214 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2215 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2216 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2217 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2218 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2219 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2220 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2221 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2222 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2223 default: fatal_unexpected_iid(id); break;
2224 }
2225 set_result(_gvn.transform(n));
2226 return true;
2227 }
2228
2229 //--------------------------inline_bitshuffle_methods-----------------------------
2230 // inline int Integer.compress(int, int)
2231 // inline int Integer.expand(int, int)
2232 // inline long Long.compress(long, long)
2233 // inline long Long.expand(long, long)
2234 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2235 Node* n = nullptr;
2236 switch (id) {
2237 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2238 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2239 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2240 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2241 default: fatal_unexpected_iid(id); break;
2242 }
2243 set_result(_gvn.transform(n));
2244 return true;
2245 }
2246
2247 //--------------------------inline_number_methods-----------------------------
2248 // inline int Integer.compareUnsigned(int, int)
2249 // inline int Long.compareUnsigned(long, long)
2250 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2251 Node* arg1 = argument(0);
2252 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2253 Node* n = nullptr;
2254 switch (id) {
2255 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2256 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2257 default: fatal_unexpected_iid(id); break;
2258 }
2259 set_result(_gvn.transform(n));
2260 return true;
2261 }
2262
2263 //--------------------------inline_unsigned_divmod_methods-----------------------------
2264 // inline int Integer.divideUnsigned(int, int)
2265 // inline int Integer.remainderUnsigned(int, int)
2266 // inline long Long.divideUnsigned(long, long)
2267 // inline long Long.remainderUnsigned(long, long)
2268 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2269 Node* n = nullptr;
2270 switch (id) {
2271 case vmIntrinsics::_divideUnsigned_i: {
2272 zero_check_int(argument(1));
2273 // Compile-time detect of null-exception
2274 if (stopped()) {
2275 return true; // keep the graph constructed so far
2276 }
2277 n = new UDivINode(control(), argument(0), argument(1));
2278 break;
2279 }
2280 case vmIntrinsics::_divideUnsigned_l: {
2281 zero_check_long(argument(2));
2282 // Compile-time detect of null-exception
2283 if (stopped()) {
2284 return true; // keep the graph constructed so far
2285 }
2286 n = new UDivLNode(control(), argument(0), argument(2));
2287 break;
2288 }
2289 case vmIntrinsics::_remainderUnsigned_i: {
2290 zero_check_int(argument(1));
2291 // Compile-time detect of null-exception
2292 if (stopped()) {
2293 return true; // keep the graph constructed so far
2294 }
2295 n = new UModINode(control(), argument(0), argument(1));
2296 break;
2297 }
2298 case vmIntrinsics::_remainderUnsigned_l: {
2299 zero_check_long(argument(2));
2300 // Compile-time detect of null-exception
2301 if (stopped()) {
2302 return true; // keep the graph constructed so far
2303 }
2304 n = new UModLNode(control(), argument(0), argument(2));
2305 break;
2306 }
2307 default: fatal_unexpected_iid(id); break;
2308 }
2309 set_result(_gvn.transform(n));
2310 return true;
2311 }
2312
2313 //----------------------------inline_unsafe_access----------------------------
2314
2315 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2316 // Attempt to infer a sharper value type from the offset and base type.
2317 ciKlass* sharpened_klass = nullptr;
2318 bool null_free = false;
2319
2320 // See if it is an instance field, with an object type.
2321 if (alias_type->field() != nullptr) {
2322 if (alias_type->field()->type()->is_klass()) {
2323 sharpened_klass = alias_type->field()->type()->as_klass();
2324 null_free = alias_type->field()->is_null_free();
2325 }
2326 }
2327
2328 const TypeOopPtr* result = nullptr;
2329 // See if it is a narrow oop array.
2330 if (adr_type->isa_aryptr()) {
2331 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2332 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2333 null_free = adr_type->is_aryptr()->is_null_free();
2334 if (elem_type != nullptr && elem_type->is_loaded()) {
2335 // Sharpen the value type.
2336 result = elem_type;
2337 }
2338 }
2339 }
2340
2341 // The sharpened class might be unloaded if there is no class loader
2342 // contraint in place.
2343 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2344 // Sharpen the value type.
2345 result = TypeOopPtr::make_from_klass(sharpened_klass);
2346 if (null_free) {
2347 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2348 }
2349 }
2350 if (result != nullptr) {
2351 #ifndef PRODUCT
2352 if (C->print_intrinsics() || C->print_inlining()) {
2353 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2354 tty->print(" sharpened value: "); result->dump(); tty->cr();
2355 }
2356 #endif
2357 }
2358 return result;
2359 }
2360
2361 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2362 switch (kind) {
2363 case Relaxed:
2364 return MO_UNORDERED;
2365 case Opaque:
2366 return MO_RELAXED;
2367 case Acquire:
2368 return MO_ACQUIRE;
2369 case Release:
2370 return MO_RELEASE;
2371 case Volatile:
2372 return MO_SEQ_CST;
2373 default:
2374 ShouldNotReachHere();
2375 return 0;
2376 }
2377 }
2378
2379 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2380 if (callee()->is_static()) return false; // caller must have the capability!
2381 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2382 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2383 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2384 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2385
2386 if (is_reference_type(type)) {
2387 decorators |= ON_UNKNOWN_OOP_REF;
2388 }
2389
2390 if (unaligned) {
2391 decorators |= C2_UNALIGNED;
2392 }
2393
2394 #ifndef PRODUCT
2395 {
2396 ResourceMark rm;
2397 // Check the signatures.
2398 ciSignature* sig = callee()->signature();
2399 #ifdef ASSERT
2400 if (!is_store) {
2401 // Object getReference(Object base, int/long offset), etc.
2402 BasicType rtype = sig->return_type()->basic_type();
2403 assert(rtype == type, "getter must return the expected value");
2404 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2405 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2406 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2407 } else {
2408 // void putReference(Object base, int/long offset, Object x), etc.
2409 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2410 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2411 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2412 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2413 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2414 assert(vtype == type, "putter must accept the expected value");
2415 }
2416 #endif // ASSERT
2417 }
2418 #endif //PRODUCT
2419
2420 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2421
2422 Node* receiver = argument(0); // type: oop
2423
2424 // Build address expression.
2425 Node* heap_base_oop = top();
2426
2427 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2428 Node* base = argument(1); // type: oop
2429 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2430 Node* offset = argument(2); // type: long
2431 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2432 // to be plain byte offsets, which are also the same as those accepted
2433 // by oopDesc::field_addr.
2434 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2435 "fieldOffset must be byte-scaled");
2436
2437 ciInlineKlass* inline_klass = nullptr;
2438 if (is_flat) {
2439 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2440 if (cls == nullptr || cls->const_oop() == nullptr) {
2441 return false;
2442 }
2443 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2444 if (!mirror_type->is_inlinetype()) {
2445 return false;
2446 }
2447 inline_klass = mirror_type->as_inline_klass();
2448 }
2449
2450 if (base->is_InlineType()) {
2451 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2452 InlineTypeNode* vt = base->as_InlineType();
2453 if (offset->is_Con()) {
2454 long off = find_long_con(offset, 0);
2455 ciInlineKlass* vk = vt->type()->inline_klass();
2456 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2457 return false;
2458 }
2459
2460 ciField* field = vk->get_non_flat_field_by_offset(off);
2461 if (field != nullptr) {
2462 BasicType bt = type2field[field->type()->basic_type()];
2463 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2464 bt = T_OBJECT;
2465 }
2466 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2467 Node* value = vt->field_value_by_offset(off, false);
2468 if (value->is_InlineType()) {
2469 value = value->as_InlineType()->adjust_scalarization_depth(this);
2470 }
2471 set_result(value);
2472 return true;
2473 }
2474 }
2475 }
2476 {
2477 // Re-execute the unsafe access if allocation triggers deoptimization.
2478 PreserveReexecuteState preexecs(this);
2479 jvms()->set_should_reexecute(true);
2480 vt = vt->buffer(this);
2481 }
2482 base = vt->get_oop();
2483 }
2484
2485 // 32-bit machines ignore the high half!
2486 offset = ConvL2X(offset);
2487
2488 // Save state and restore on bailout
2489 uint old_sp = sp();
2490 SafePointNode* old_map = clone_map();
2491
2492 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2493 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2494
2495 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2496 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2497 decorators |= IN_NATIVE; // off-heap primitive access
2498 } else {
2499 set_map(old_map);
2500 set_sp(old_sp);
2501 return false; // off-heap oop accesses are not supported
2502 }
2503 } else {
2504 heap_base_oop = base; // on-heap or mixed access
2505 }
2506
2507 // Can base be null? Otherwise, always on-heap access.
2508 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2509
2510 if (!can_access_non_heap) {
2511 decorators |= IN_HEAP;
2512 }
2513
2514 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2515
2516 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2517 if (adr_type == TypePtr::NULL_PTR) {
2518 set_map(old_map);
2519 set_sp(old_sp);
2520 return false; // off-heap access with zero address
2521 }
2522
2523 // Try to categorize the address.
2524 Compile::AliasType* alias_type = C->alias_type(adr_type);
2525 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2526
2527 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2528 alias_type->adr_type() == TypeAryPtr::RANGE) {
2529 set_map(old_map);
2530 set_sp(old_sp);
2531 return false; // not supported
2532 }
2533
2534 bool mismatched = false;
2535 BasicType bt = T_ILLEGAL;
2536 ciField* field = nullptr;
2537 if (adr_type->isa_instptr()) {
2538 const TypeInstPtr* instptr = adr_type->is_instptr();
2539 ciInstanceKlass* k = instptr->instance_klass();
2540 int off = instptr->offset();
2541 if (instptr->const_oop() != nullptr &&
2542 k == ciEnv::current()->Class_klass() &&
2543 instptr->offset() >= (k->size_helper() * wordSize)) {
2544 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2545 field = k->get_field_by_offset(off, true);
2546 } else {
2547 field = k->get_non_flat_field_by_offset(off);
2548 }
2549 if (field != nullptr) {
2550 bt = type2field[field->type()->basic_type()];
2551 }
2552 if (bt != alias_type->basic_type()) {
2553 // Type mismatch. Is it an access to a nested flat field?
2554 field = k->get_field_by_offset(off, false);
2555 if (field != nullptr) {
2556 bt = type2field[field->type()->basic_type()];
2557 }
2558 }
2559 assert(bt == alias_type->basic_type() || is_flat, "should match");
2560 } else {
2561 bt = alias_type->basic_type();
2562 }
2563
2564 if (bt != T_ILLEGAL) {
2565 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2566 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2567 // Alias type doesn't differentiate between byte[] and boolean[]).
2568 // Use address type to get the element type.
2569 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2570 }
2571 if (is_reference_type(bt, true)) {
2572 // accessing an array field with getReference is not a mismatch
2573 bt = T_OBJECT;
2574 }
2575 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2576 // Don't intrinsify mismatched object accesses
2577 set_map(old_map);
2578 set_sp(old_sp);
2579 return false;
2580 }
2581 mismatched = (bt != type);
2582 } else if (alias_type->adr_type()->isa_oopptr()) {
2583 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2584 }
2585
2586 if (is_flat) {
2587 if (adr_type->isa_instptr()) {
2588 if (field == nullptr || field->type() != inline_klass) {
2589 mismatched = true;
2590 }
2591 } else if (adr_type->isa_aryptr()) {
2592 const Type* elem = adr_type->is_aryptr()->elem();
2593 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2594 mismatched = true;
2595 }
2596 } else {
2597 mismatched = true;
2598 }
2599 if (is_store) {
2600 const Type* val_t = _gvn.type(val);
2601 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2602 set_map(old_map);
2603 set_sp(old_sp);
2604 return false;
2605 }
2606 }
2607 }
2608
2609 destruct_map_clone(old_map);
2610 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2611
2612 if (mismatched) {
2613 decorators |= C2_MISMATCHED;
2614 }
2615
2616 // First guess at the value type.
2617 const Type *value_type = Type::get_const_basic_type(type);
2618
2619 // Figure out the memory ordering.
2620 decorators |= mo_decorator_for_access_kind(kind);
2621
2622 if (!is_store) {
2623 if (type == T_OBJECT && !is_flat) {
2624 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2625 if (tjp != nullptr) {
2626 value_type = tjp;
2627 }
2628 }
2629 }
2630
2631 receiver = null_check(receiver);
2632 if (stopped()) {
2633 return true;
2634 }
2635 // Heap pointers get a null-check from the interpreter,
2636 // as a courtesy. However, this is not guaranteed by Unsafe,
2637 // and it is not possible to fully distinguish unintended nulls
2638 // from intended ones in this API.
2639
2640 if (!is_store) {
2641 Node* p = nullptr;
2642 // Try to constant fold a load from a constant field
2643
2644 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2645 // final or stable field
2646 p = make_constant_from_field(field, heap_base_oop);
2647 }
2648
2649 if (p == nullptr) { // Could not constant fold the load
2650 if (is_flat) {
2651 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, adr_type, false, false, true);
2652 } else {
2653 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2654 const TypeOopPtr* ptr = value_type->make_oopptr();
2655 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2656 // Load a non-flattened inline type from memory
2657 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2658 }
2659 }
2660 // Normalize the value returned by getBoolean in the following cases
2661 if (type == T_BOOLEAN &&
2662 (mismatched ||
2663 heap_base_oop == top() || // - heap_base_oop is null or
2664 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2665 // and the unsafe access is made to large offset
2666 // (i.e., larger than the maximum offset necessary for any
2667 // field access)
2668 ) {
2669 IdealKit ideal = IdealKit(this);
2670 #define __ ideal.
2671 IdealVariable normalized_result(ideal);
2672 __ declarations_done();
2673 __ set(normalized_result, p);
2674 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2675 __ set(normalized_result, ideal.ConI(1));
2676 ideal.end_if();
2677 final_sync(ideal);
2678 p = __ value(normalized_result);
2679 #undef __
2680 }
2681 }
2682 if (type == T_ADDRESS) {
2683 p = gvn().transform(new CastP2XNode(nullptr, p));
2684 p = ConvX2UL(p);
2685 }
2686 // The load node has the control of the preceding MemBarCPUOrder. All
2687 // following nodes will have the control of the MemBarCPUOrder inserted at
2688 // the end of this method. So, pushing the load onto the stack at a later
2689 // point is fine.
2690 set_result(p);
2691 } else {
2692 if (bt == T_ADDRESS) {
2693 // Repackage the long as a pointer.
2694 val = ConvL2X(val);
2695 val = gvn().transform(new CastX2PNode(val));
2696 }
2697 if (is_flat) {
2698 val->as_InlineType()->store_flat(this, base, adr, false, false, true, decorators);
2699 } else {
2700 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2701 }
2702 }
2703
2704 return true;
2705 }
2706
2707 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2708 Node* receiver = argument(0);
2709 Node* value = argument(1);
2710
2711 const Type* type = gvn().type(value);
2712 if (!type->is_inlinetypeptr()) {
2713 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2714 return false;
2715 }
2716
2717 null_check(receiver);
2718 if (stopped()) {
2719 return true;
2720 }
2721
2722 value = null_check(value);
2723 if (stopped()) {
2724 return true;
2725 }
2726
2727 ciInlineKlass* vk = type->inline_klass();
2728 Node* klass = makecon(TypeKlassPtr::make(vk));
2729 Node* obj = new_instance(klass);
2730 AllocateNode::Ideal_allocation(obj)->_larval = true;
2731
2732 assert(value->is_InlineType(), "must be an InlineTypeNode");
2733 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2734 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2735
2736 set_result(obj);
2737 return true;
2738 }
2739
2740 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2741 Node* receiver = argument(0);
2742 Node* buffer = argument(1);
2743
2744 const Type* type = gvn().type(buffer);
2745 if (!type->is_inlinetypeptr()) {
2746 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2747 return false;
2748 }
2749
2750 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2751 if (alloc == nullptr) {
2752 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2753 return false;
2754 }
2755
2756 null_check(receiver);
2757 if (stopped()) {
2758 return true;
2759 }
2760
2761 // Unset the larval bit in the object header
2762 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2763 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2764 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2765
2766 // We must ensure that the buffer is properly published
2767 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2768 assert(!type->maybe_null(), "result of an allocation should not be null");
2769 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
2770 return true;
2771 }
2772
2773 //----------------------------inline_unsafe_load_store----------------------------
2774 // This method serves a couple of different customers (depending on LoadStoreKind):
2775 //
2776 // LS_cmp_swap:
2777 //
2778 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2779 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2780 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2781 //
2782 // LS_cmp_swap_weak:
2783 //
2784 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2785 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2786 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2787 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2788 //
2789 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2790 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2791 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2792 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2793 //
2794 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2795 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2796 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2797 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2798 //
2799 // LS_cmp_exchange:
2800 //
2801 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2802 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2803 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2804 //
2805 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2806 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2807 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2808 //
2809 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2810 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2811 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2812 //
2813 // LS_get_add:
2814 //
2815 // int getAndAddInt( Object o, long offset, int delta)
2816 // long getAndAddLong(Object o, long offset, long delta)
2817 //
2818 // LS_get_set:
2819 //
2820 // int getAndSet(Object o, long offset, int newValue)
2821 // long getAndSet(Object o, long offset, long newValue)
2822 // Object getAndSet(Object o, long offset, Object newValue)
2823 //
2824 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2825 // This basic scheme here is the same as inline_unsafe_access, but
2826 // differs in enough details that combining them would make the code
2827 // overly confusing. (This is a true fact! I originally combined
2828 // them, but even I was confused by it!) As much code/comments as
2829 // possible are retained from inline_unsafe_access though to make
2830 // the correspondences clearer. - dl
2831
2832 if (callee()->is_static()) return false; // caller must have the capability!
2833
2834 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2835 decorators |= mo_decorator_for_access_kind(access_kind);
2836
2837 #ifndef PRODUCT
2838 BasicType rtype;
2839 {
2840 ResourceMark rm;
2841 // Check the signatures.
2842 ciSignature* sig = callee()->signature();
2843 rtype = sig->return_type()->basic_type();
2844 switch(kind) {
2845 case LS_get_add:
2846 case LS_get_set: {
2847 // Check the signatures.
2848 #ifdef ASSERT
2849 assert(rtype == type, "get and set must return the expected type");
2850 assert(sig->count() == 3, "get and set has 3 arguments");
2851 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2852 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2853 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2854 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2855 #endif // ASSERT
2856 break;
2857 }
2858 case LS_cmp_swap:
2859 case LS_cmp_swap_weak: {
2860 // Check the signatures.
2861 #ifdef ASSERT
2862 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2863 assert(sig->count() == 4, "CAS has 4 arguments");
2864 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2865 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2866 #endif // ASSERT
2867 break;
2868 }
2869 case LS_cmp_exchange: {
2870 // Check the signatures.
2871 #ifdef ASSERT
2872 assert(rtype == type, "CAS must return the expected type");
2873 assert(sig->count() == 4, "CAS has 4 arguments");
2874 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2875 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2876 #endif // ASSERT
2877 break;
2878 }
2879 default:
2880 ShouldNotReachHere();
2881 }
2882 }
2883 #endif //PRODUCT
2884
2885 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2886
2887 // Get arguments:
2888 Node* receiver = nullptr;
2889 Node* base = nullptr;
2890 Node* offset = nullptr;
2891 Node* oldval = nullptr;
2892 Node* newval = nullptr;
2893 switch(kind) {
2894 case LS_cmp_swap:
2895 case LS_cmp_swap_weak:
2896 case LS_cmp_exchange: {
2897 const bool two_slot_type = type2size[type] == 2;
2898 receiver = argument(0); // type: oop
2899 base = argument(1); // type: oop
2900 offset = argument(2); // type: long
2901 oldval = argument(4); // type: oop, int, or long
2902 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2903 break;
2904 }
2905 case LS_get_add:
2906 case LS_get_set: {
2907 receiver = argument(0); // type: oop
2908 base = argument(1); // type: oop
2909 offset = argument(2); // type: long
2910 oldval = nullptr;
2911 newval = argument(4); // type: oop, int, or long
2912 break;
2913 }
2914 default:
2915 ShouldNotReachHere();
2916 }
2917
2918 // Build field offset expression.
2919 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2920 // to be plain byte offsets, which are also the same as those accepted
2921 // by oopDesc::field_addr.
2922 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2923 // 32-bit machines ignore the high half of long offsets
2924 offset = ConvL2X(offset);
2925 // Save state and restore on bailout
2926 uint old_sp = sp();
2927 SafePointNode* old_map = clone_map();
2928 Node* adr = make_unsafe_address(base, offset,type, false);
2929 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2930
2931 Compile::AliasType* alias_type = C->alias_type(adr_type);
2932 BasicType bt = alias_type->basic_type();
2933 if (bt != T_ILLEGAL &&
2934 (is_reference_type(bt) != (type == T_OBJECT))) {
2935 // Don't intrinsify mismatched object accesses.
2936 set_map(old_map);
2937 set_sp(old_sp);
2938 return false;
2939 }
2940
2941 destruct_map_clone(old_map);
2942
2943 // For CAS, unlike inline_unsafe_access, there seems no point in
2944 // trying to refine types. Just use the coarse types here.
2945 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2946 const Type *value_type = Type::get_const_basic_type(type);
2947
2948 switch (kind) {
2949 case LS_get_set:
2950 case LS_cmp_exchange: {
2951 if (type == T_OBJECT) {
2952 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2953 if (tjp != nullptr) {
2954 value_type = tjp;
2955 }
2956 }
2957 break;
2958 }
2959 case LS_cmp_swap:
2960 case LS_cmp_swap_weak:
2961 case LS_get_add:
2962 break;
2963 default:
2964 ShouldNotReachHere();
2965 }
2966
2967 // Null check receiver.
2968 receiver = null_check(receiver);
2969 if (stopped()) {
2970 return true;
2971 }
2972
2973 int alias_idx = C->get_alias_index(adr_type);
2974
2975 if (is_reference_type(type)) {
2976 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2977
2978 if (oldval != nullptr && oldval->is_InlineType()) {
2979 // Re-execute the unsafe access if allocation triggers deoptimization.
2980 PreserveReexecuteState preexecs(this);
2981 jvms()->set_should_reexecute(true);
2982 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2983 }
2984 if (newval != nullptr && newval->is_InlineType()) {
2985 // Re-execute the unsafe access if allocation triggers deoptimization.
2986 PreserveReexecuteState preexecs(this);
2987 jvms()->set_should_reexecute(true);
2988 newval = newval->as_InlineType()->buffer(this)->get_oop();
2989 }
2990
2991 // Transformation of a value which could be null pointer (CastPP #null)
2992 // could be delayed during Parse (for example, in adjust_map_after_if()).
2993 // Execute transformation here to avoid barrier generation in such case.
2994 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2995 newval = _gvn.makecon(TypePtr::NULL_PTR);
2996
2997 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2998 // Refine the value to a null constant, when it is known to be null
2999 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3000 }
3001 }
3002
3003 Node* result = nullptr;
3004 switch (kind) {
3005 case LS_cmp_exchange: {
3006 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3007 oldval, newval, value_type, type, decorators);
3008 break;
3009 }
3010 case LS_cmp_swap_weak:
3011 decorators |= C2_WEAK_CMPXCHG;
3012 case LS_cmp_swap: {
3013 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3014 oldval, newval, value_type, type, decorators);
3015 break;
3016 }
3017 case LS_get_set: {
3018 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3019 newval, value_type, type, decorators);
3020 break;
3021 }
3022 case LS_get_add: {
3023 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3024 newval, value_type, type, decorators);
3025 break;
3026 }
3027 default:
3028 ShouldNotReachHere();
3029 }
3030
3031 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3032 set_result(result);
3033 return true;
3034 }
3035
3036 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3037 // Regardless of form, don't allow previous ld/st to move down,
3038 // then issue acquire, release, or volatile mem_bar.
3039 insert_mem_bar(Op_MemBarCPUOrder);
3040 switch(id) {
3041 case vmIntrinsics::_loadFence:
3042 insert_mem_bar(Op_LoadFence);
3043 return true;
3044 case vmIntrinsics::_storeFence:
3045 insert_mem_bar(Op_StoreFence);
3046 return true;
3047 case vmIntrinsics::_storeStoreFence:
3048 insert_mem_bar(Op_StoreStoreFence);
3049 return true;
3050 case vmIntrinsics::_fullFence:
3051 insert_mem_bar(Op_MemBarVolatile);
3052 return true;
3053 default:
3054 fatal_unexpected_iid(id);
3055 return false;
3056 }
3057 }
3058
3059 bool LibraryCallKit::inline_onspinwait() {
3060 insert_mem_bar(Op_OnSpinWait);
3061 return true;
3062 }
3063
3064 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3065 if (!kls->is_Con()) {
3066 return true;
3067 }
3068 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3069 if (klsptr == nullptr) {
3070 return true;
3071 }
3072 ciInstanceKlass* ik = klsptr->instance_klass();
3073 // don't need a guard for a klass that is already initialized
3074 return !ik->is_initialized();
3075 }
3076
3077 //----------------------------inline_unsafe_writeback0-------------------------
3078 // public native void Unsafe.writeback0(long address)
3079 bool LibraryCallKit::inline_unsafe_writeback0() {
3080 if (!Matcher::has_match_rule(Op_CacheWB)) {
3081 return false;
3082 }
3083 #ifndef PRODUCT
3084 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3085 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3086 ciSignature* sig = callee()->signature();
3087 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3088 #endif
3089 null_check_receiver(); // null-check, then ignore
3090 Node *addr = argument(1);
3091 addr = new CastX2PNode(addr);
3092 addr = _gvn.transform(addr);
3093 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3094 flush = _gvn.transform(flush);
3095 set_memory(flush, TypeRawPtr::BOTTOM);
3096 return true;
3097 }
3098
3099 //----------------------------inline_unsafe_writeback0-------------------------
3100 // public native void Unsafe.writeback0(long address)
3101 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3102 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3103 return false;
3104 }
3105 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3106 return false;
3107 }
3108 #ifndef PRODUCT
3109 assert(Matcher::has_match_rule(Op_CacheWB),
3110 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3111 : "found match rule for CacheWBPostSync but not CacheWB"));
3112
3113 #endif
3114 null_check_receiver(); // null-check, then ignore
3115 Node *sync;
3116 if (is_pre) {
3117 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3118 } else {
3119 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3120 }
3121 sync = _gvn.transform(sync);
3122 set_memory(sync, TypeRawPtr::BOTTOM);
3123 return true;
3124 }
3125
3126 //----------------------------inline_unsafe_allocate---------------------------
3127 // public native Object Unsafe.allocateInstance(Class<?> cls);
3128 bool LibraryCallKit::inline_unsafe_allocate() {
3129
3130 #if INCLUDE_JVMTI
3131 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3132 return false;
3133 }
3134 #endif //INCLUDE_JVMTI
3135
3136 if (callee()->is_static()) return false; // caller must have the capability!
3137
3138 null_check_receiver(); // null-check, then ignore
3139 Node* cls = null_check(argument(1));
3140 if (stopped()) return true;
3141
3142 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3143 kls = null_check(kls);
3144 if (stopped()) return true; // argument was like int.class
3145
3146 #if INCLUDE_JVMTI
3147 // Don't try to access new allocated obj in the intrinsic.
3148 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3149 // Deoptimize and allocate in interpreter instead.
3150 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3151 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3152 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3153 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3154 {
3155 BuildCutout unless(this, tst, PROB_MAX);
3156 uncommon_trap(Deoptimization::Reason_intrinsic,
3157 Deoptimization::Action_make_not_entrant);
3158 }
3159 if (stopped()) {
3160 return true;
3161 }
3162 #endif //INCLUDE_JVMTI
3163
3164 Node* test = nullptr;
3165 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3166 // Note: The argument might still be an illegal value like
3167 // Serializable.class or Object[].class. The runtime will handle it.
3168 // But we must make an explicit check for initialization.
3169 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3170 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3171 // can generate code to load it as unsigned byte.
3172 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3173 Node* bits = intcon(InstanceKlass::fully_initialized);
3174 test = _gvn.transform(new SubINode(inst, bits));
3175 // The 'test' is non-zero if we need to take a slow path.
3176 }
3177 Node* obj = nullptr;
3178 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3179 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3180 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3181 } else {
3182 obj = new_instance(kls, test);
3183 }
3184 set_result(obj);
3185 return true;
3186 }
3187
3188 //------------------------inline_native_time_funcs--------------
3189 // inline code for System.currentTimeMillis() and System.nanoTime()
3190 // these have the same type and signature
3191 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3192 const TypeFunc* tf = OptoRuntime::void_long_Type();
3193 const TypePtr* no_memory_effects = nullptr;
3194 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3195 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3196 #ifdef ASSERT
3197 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3198 assert(value_top == top(), "second value must be top");
3199 #endif
3200 set_result(value);
3201 return true;
3202 }
3203
3204
3205 #if INCLUDE_JVMTI
3206
3207 // When notifications are disabled then just update the VTMS transition bit and return.
3208 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3209 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3210 if (!DoJVMTIVirtualThreadTransitions) {
3211 return true;
3212 }
3213 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3214 IdealKit ideal(this);
3215
3216 Node* ONE = ideal.ConI(1);
3217 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3218 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3219 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3220
3221 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3222 sync_kit(ideal);
3223 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3224 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3225 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3226 ideal.sync_kit(this);
3227 } ideal.else_(); {
3228 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3229 Node* thread = ideal.thread();
3230 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3231 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3232
3233 sync_kit(ideal);
3234 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3235 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3236
3237 ideal.sync_kit(this);
3238 } ideal.end_if();
3239 final_sync(ideal);
3240
3241 return true;
3242 }
3243
3244 // Always update the is_disable_suspend bit.
3245 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3246 if (!DoJVMTIVirtualThreadTransitions) {
3247 return true;
3248 }
3249 IdealKit ideal(this);
3250
3251 {
3252 // unconditionally update the is_disable_suspend bit in current JavaThread
3253 Node* thread = ideal.thread();
3254 Node* arg = _gvn.transform(argument(0)); // argument for notification
3255 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3256 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3257
3258 sync_kit(ideal);
3259 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3260 ideal.sync_kit(this);
3261 }
3262 final_sync(ideal);
3263
3264 return true;
3265 }
3266
3267 #endif // INCLUDE_JVMTI
3268
3269 #ifdef JFR_HAVE_INTRINSICS
3270
3271 /**
3272 * if oop->klass != null
3273 * // normal class
3274 * epoch = _epoch_state ? 2 : 1
3275 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3276 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3277 * }
3278 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3279 * else
3280 * // primitive class
3281 * if oop->array_klass != null
3282 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3283 * else
3284 * id = LAST_TYPE_ID + 1 // void class path
3285 * if (!signaled)
3286 * signaled = true
3287 */
3288 bool LibraryCallKit::inline_native_classID() {
3289 Node* cls = argument(0);
3290
3291 IdealKit ideal(this);
3292 #define __ ideal.
3293 IdealVariable result(ideal); __ declarations_done();
3294 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3295 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3296 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3297
3298
3299 __ if_then(kls, BoolTest::ne, null()); {
3300 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3301 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3302
3303 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3304 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3305 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3306 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3307 mask = _gvn.transform(new OrLNode(mask, epoch));
3308 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3309
3310 float unlikely = PROB_UNLIKELY(0.999);
3311 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3312 sync_kit(ideal);
3313 make_runtime_call(RC_LEAF,
3314 OptoRuntime::class_id_load_barrier_Type(),
3315 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3316 "class id load barrier",
3317 TypePtr::BOTTOM,
3318 kls);
3319 ideal.sync_kit(this);
3320 } __ end_if();
3321
3322 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3323 } __ else_(); {
3324 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3325 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3326 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3327 __ if_then(array_kls, BoolTest::ne, null()); {
3328 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3329 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3330 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3331 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3332 } __ else_(); {
3333 // void class case
3334 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3335 } __ end_if();
3336
3337 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3338 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3339 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3340 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3341 } __ end_if();
3342 } __ end_if();
3343
3344 final_sync(ideal);
3345 set_result(ideal.value(result));
3346 #undef __
3347 return true;
3348 }
3349
3350 //------------------------inline_native_jvm_commit------------------
3351 bool LibraryCallKit::inline_native_jvm_commit() {
3352 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3353
3354 // Save input memory and i_o state.
3355 Node* input_memory_state = reset_memory();
3356 set_all_memory(input_memory_state);
3357 Node* input_io_state = i_o();
3358
3359 // TLS.
3360 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3361 // Jfr java buffer.
3362 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3363 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3364 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3365
3366 // Load the current value of the notified field in the JfrThreadLocal.
3367 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3368 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3369
3370 // Test for notification.
3371 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3372 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3373 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3374
3375 // True branch, is notified.
3376 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3377 set_control(is_notified);
3378
3379 // Reset notified state.
3380 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3381 Node* notified_reset_memory = reset_memory();
3382
3383 // 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.
3384 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3385 // Convert the machine-word to a long.
3386 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3387
3388 // False branch, not notified.
3389 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3390 set_control(not_notified);
3391 set_all_memory(input_memory_state);
3392
3393 // Arg is the next position as a long.
3394 Node* arg = argument(0);
3395 // Convert long to machine-word.
3396 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3397
3398 // Store the next_position to the underlying jfr java buffer.
3399 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3400
3401 Node* commit_memory = reset_memory();
3402 set_all_memory(commit_memory);
3403
3404 // 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.
3405 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3406 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3407 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3408
3409 // And flags with lease constant.
3410 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3411
3412 // Branch on lease to conditionalize returning the leased java buffer.
3413 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3414 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3415 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3416
3417 // False branch, not a lease.
3418 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3419
3420 // True branch, is lease.
3421 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3422 set_control(is_lease);
3423
3424 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3425 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3426 OptoRuntime::void_void_Type(),
3427 SharedRuntime::jfr_return_lease(),
3428 "return_lease", TypePtr::BOTTOM);
3429 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3430
3431 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3432 record_for_igvn(lease_compare_rgn);
3433 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3434 record_for_igvn(lease_compare_mem);
3435 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3436 record_for_igvn(lease_compare_io);
3437 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3438 record_for_igvn(lease_result_value);
3439
3440 // Update control and phi nodes.
3441 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3442 lease_compare_rgn->init_req(_false_path, not_lease);
3443
3444 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3445 lease_compare_mem->init_req(_false_path, commit_memory);
3446
3447 lease_compare_io->init_req(_true_path, i_o());
3448 lease_compare_io->init_req(_false_path, input_io_state);
3449
3450 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3451 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3452
3453 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3454 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3455 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3456 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3457
3458 // Update control and phi nodes.
3459 result_rgn->init_req(_true_path, is_notified);
3460 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3461
3462 result_mem->init_req(_true_path, notified_reset_memory);
3463 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3464
3465 result_io->init_req(_true_path, input_io_state);
3466 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3467
3468 result_value->init_req(_true_path, current_pos);
3469 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3470
3471 // Set output state.
3472 set_control(_gvn.transform(result_rgn));
3473 set_all_memory(_gvn.transform(result_mem));
3474 set_i_o(_gvn.transform(result_io));
3475 set_result(result_rgn, result_value);
3476 return true;
3477 }
3478
3479 /*
3480 * The intrinsic is a model of this pseudo-code:
3481 *
3482 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3483 * jobject h_event_writer = tl->java_event_writer();
3484 * if (h_event_writer == nullptr) {
3485 * return nullptr;
3486 * }
3487 * oop threadObj = Thread::threadObj();
3488 * oop vthread = java_lang_Thread::vthread(threadObj);
3489 * traceid tid;
3490 * bool pinVirtualThread;
3491 * bool excluded;
3492 * if (vthread != threadObj) { // i.e. current thread is virtual
3493 * tid = java_lang_Thread::tid(vthread);
3494 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3495 * pinVirtualThread = VMContinuations;
3496 * excluded = vthread_epoch_raw & excluded_mask;
3497 * if (!excluded) {
3498 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3499 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3500 * if (vthread_epoch != current_epoch) {
3501 * write_checkpoint();
3502 * }
3503 * }
3504 * } else {
3505 * tid = java_lang_Thread::tid(threadObj);
3506 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3507 * pinVirtualThread = false;
3508 * excluded = thread_epoch_raw & excluded_mask;
3509 * }
3510 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3511 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3512 * if (tid_in_event_writer != tid) {
3513 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3514 * setField(event_writer, "excluded", excluded);
3515 * setField(event_writer, "threadID", tid);
3516 * }
3517 * return event_writer
3518 */
3519 bool LibraryCallKit::inline_native_getEventWriter() {
3520 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3521
3522 // Save input memory and i_o state.
3523 Node* input_memory_state = reset_memory();
3524 set_all_memory(input_memory_state);
3525 Node* input_io_state = i_o();
3526
3527 // The most significant bit of the u2 is used to denote thread exclusion
3528 Node* excluded_shift = _gvn.intcon(15);
3529 Node* excluded_mask = _gvn.intcon(1 << 15);
3530 // The epoch generation is the range [1-32767]
3531 Node* epoch_mask = _gvn.intcon(32767);
3532
3533 // TLS
3534 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3535
3536 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3537 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3538
3539 // Load the eventwriter jobject handle.
3540 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3541
3542 // Null check the jobject handle.
3543 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3544 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3545 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3546
3547 // False path, jobj is null.
3548 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3549
3550 // True path, jobj is not null.
3551 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3552
3553 set_control(jobj_is_not_null);
3554
3555 // Load the threadObj for the CarrierThread.
3556 Node* threadObj = generate_current_thread(tls_ptr);
3557
3558 // Load the vthread.
3559 Node* vthread = generate_virtual_thread(tls_ptr);
3560
3561 // If vthread != threadObj, this is a virtual thread.
3562 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3563 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3564 IfNode* iff_vthread_not_equal_threadObj =
3565 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3566
3567 // False branch, fallback to threadObj.
3568 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3569 set_control(vthread_equal_threadObj);
3570
3571 // Load the tid field from the vthread object.
3572 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3573
3574 // Load the raw epoch value from the threadObj.
3575 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3576 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3577 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3578 TypeInt::CHAR, T_CHAR,
3579 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3580
3581 // Mask off the excluded information from the epoch.
3582 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3583
3584 // True branch, this is a virtual thread.
3585 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3586 set_control(vthread_not_equal_threadObj);
3587
3588 // Load the tid field from the vthread object.
3589 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3590
3591 // Continuation support determines if a virtual thread should be pinned.
3592 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3593 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3594
3595 // Load the raw epoch value from the vthread.
3596 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3597 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3598 TypeInt::CHAR, T_CHAR,
3599 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3600
3601 // Mask off the excluded information from the epoch.
3602 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3603
3604 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3605 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3606 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3607 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3608
3609 // False branch, vthread is excluded, no need to write epoch info.
3610 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3611
3612 // True branch, vthread is included, update epoch info.
3613 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3614 set_control(included);
3615
3616 // Get epoch value.
3617 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3618
3619 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3620 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3621 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3622
3623 // Compare the epoch in the vthread to the current epoch generation.
3624 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3625 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3626 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3627
3628 // False path, epoch is equal, checkpoint information is valid.
3629 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3630
3631 // True path, epoch is not equal, write a checkpoint for the vthread.
3632 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3633
3634 set_control(epoch_is_not_equal);
3635
3636 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3637 // The call also updates the native thread local thread id and the vthread with the current epoch.
3638 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3639 OptoRuntime::jfr_write_checkpoint_Type(),
3640 SharedRuntime::jfr_write_checkpoint(),
3641 "write_checkpoint", TypePtr::BOTTOM);
3642 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3643
3644 // vthread epoch != current epoch
3645 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3646 record_for_igvn(epoch_compare_rgn);
3647 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3648 record_for_igvn(epoch_compare_mem);
3649 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3650 record_for_igvn(epoch_compare_io);
3651
3652 // Update control and phi nodes.
3653 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3654 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3655 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3656 epoch_compare_mem->init_req(_false_path, input_memory_state);
3657 epoch_compare_io->init_req(_true_path, i_o());
3658 epoch_compare_io->init_req(_false_path, input_io_state);
3659
3660 // excluded != true
3661 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3662 record_for_igvn(exclude_compare_rgn);
3663 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3664 record_for_igvn(exclude_compare_mem);
3665 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3666 record_for_igvn(exclude_compare_io);
3667
3668 // Update control and phi nodes.
3669 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3670 exclude_compare_rgn->init_req(_false_path, excluded);
3671 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3672 exclude_compare_mem->init_req(_false_path, input_memory_state);
3673 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3674 exclude_compare_io->init_req(_false_path, input_io_state);
3675
3676 // vthread != threadObj
3677 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3678 record_for_igvn(vthread_compare_rgn);
3679 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3680 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3681 record_for_igvn(vthread_compare_io);
3682 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3683 record_for_igvn(tid);
3684 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3685 record_for_igvn(exclusion);
3686 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3687 record_for_igvn(pinVirtualThread);
3688
3689 // Update control and phi nodes.
3690 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3691 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3692 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3693 vthread_compare_mem->init_req(_false_path, input_memory_state);
3694 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3695 vthread_compare_io->init_req(_false_path, input_io_state);
3696 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3697 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3698 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3699 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3700 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3701 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3702
3703 // Update branch state.
3704 set_control(_gvn.transform(vthread_compare_rgn));
3705 set_all_memory(_gvn.transform(vthread_compare_mem));
3706 set_i_o(_gvn.transform(vthread_compare_io));
3707
3708 // Load the event writer oop by dereferencing the jobject handle.
3709 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3710 assert(klass_EventWriter->is_loaded(), "invariant");
3711 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3712 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3713 const TypeOopPtr* const xtype = aklass->as_instance_type();
3714 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3715 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3716
3717 // Load the current thread id from the event writer object.
3718 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3719 // Get the field offset to, conditionally, store an updated tid value later.
3720 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3721 // Get the field offset to, conditionally, store an updated exclusion value later.
3722 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3723 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3724 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3725
3726 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3727 record_for_igvn(event_writer_tid_compare_rgn);
3728 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3729 record_for_igvn(event_writer_tid_compare_mem);
3730 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3731 record_for_igvn(event_writer_tid_compare_io);
3732
3733 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3734 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3735 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3736 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3737
3738 // False path, tids are the same.
3739 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3740
3741 // True path, tid is not equal, need to update the tid in the event writer.
3742 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3743 record_for_igvn(tid_is_not_equal);
3744
3745 // Store the pin state to the event writer.
3746 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3747
3748 // Store the exclusion state to the event writer.
3749 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3750 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3751
3752 // Store the tid to the event writer.
3753 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3754
3755 // Update control and phi nodes.
3756 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3757 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3758 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3759 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3760 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3761 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3762
3763 // Result of top level CFG, Memory, IO and Value.
3764 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3765 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3766 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3767 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3768
3769 // Result control.
3770 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3771 result_rgn->init_req(_false_path, jobj_is_null);
3772
3773 // Result memory.
3774 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3775 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3776
3777 // Result IO.
3778 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3779 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3780
3781 // Result value.
3782 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3783 result_value->init_req(_false_path, null()); // return null
3784
3785 // Set output state.
3786 set_control(_gvn.transform(result_rgn));
3787 set_all_memory(_gvn.transform(result_mem));
3788 set_i_o(_gvn.transform(result_io));
3789 set_result(result_rgn, result_value);
3790 return true;
3791 }
3792
3793 /*
3794 * The intrinsic is a model of this pseudo-code:
3795 *
3796 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3797 * if (carrierThread != thread) { // is virtual thread
3798 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3799 * bool excluded = vthread_epoch_raw & excluded_mask;
3800 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3801 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3802 * if (!excluded) {
3803 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3804 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3805 * }
3806 * Atomic::release_store(&tl->_vthread, true);
3807 * return;
3808 * }
3809 * Atomic::release_store(&tl->_vthread, false);
3810 */
3811 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3812 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3813
3814 Node* input_memory_state = reset_memory();
3815 set_all_memory(input_memory_state);
3816
3817 // The most significant bit of the u2 is used to denote thread exclusion
3818 Node* excluded_mask = _gvn.intcon(1 << 15);
3819 // The epoch generation is the range [1-32767]
3820 Node* epoch_mask = _gvn.intcon(32767);
3821
3822 Node* const carrierThread = generate_current_thread(jt);
3823 // If thread != carrierThread, this is a virtual thread.
3824 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3825 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3826 IfNode* iff_thread_not_equal_carrierThread =
3827 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3828
3829 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3830
3831 // False branch, is carrierThread.
3832 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3833 // Store release
3834 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3835
3836 set_all_memory(input_memory_state);
3837
3838 // True branch, is virtual thread.
3839 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3840 set_control(thread_not_equal_carrierThread);
3841
3842 // Load the raw epoch value from the vthread.
3843 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3844 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3845 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3846
3847 // Mask off the excluded information from the epoch.
3848 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3849
3850 // Load the tid field from the thread.
3851 Node* tid = load_field_from_object(thread, "tid", "J");
3852
3853 // Store the vthread tid to the jfr thread local.
3854 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3855 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3856
3857 // Branch is_excluded to conditionalize updating the epoch .
3858 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3859 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3860 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3861
3862 // True branch, vthread is excluded, no need to write epoch info.
3863 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3864 set_control(excluded);
3865 Node* vthread_is_excluded = _gvn.intcon(1);
3866
3867 // False branch, vthread is included, update epoch info.
3868 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3869 set_control(included);
3870 Node* vthread_is_included = _gvn.intcon(0);
3871
3872 // Get epoch value.
3873 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3874
3875 // Store the vthread epoch to the jfr thread local.
3876 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3877 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3878
3879 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3880 record_for_igvn(excluded_rgn);
3881 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3882 record_for_igvn(excluded_mem);
3883 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3884 record_for_igvn(exclusion);
3885
3886 // Merge the excluded control and memory.
3887 excluded_rgn->init_req(_true_path, excluded);
3888 excluded_rgn->init_req(_false_path, included);
3889 excluded_mem->init_req(_true_path, tid_memory);
3890 excluded_mem->init_req(_false_path, included_memory);
3891 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3892 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3893
3894 // Set intermediate state.
3895 set_control(_gvn.transform(excluded_rgn));
3896 set_all_memory(excluded_mem);
3897
3898 // Store the vthread exclusion state to the jfr thread local.
3899 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3900 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3901
3902 // Store release
3903 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3904
3905 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3906 record_for_igvn(thread_compare_rgn);
3907 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3908 record_for_igvn(thread_compare_mem);
3909 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3910 record_for_igvn(vthread);
3911
3912 // Merge the thread_compare control and memory.
3913 thread_compare_rgn->init_req(_true_path, control());
3914 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3915 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3916 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3917
3918 // Set output state.
3919 set_control(_gvn.transform(thread_compare_rgn));
3920 set_all_memory(_gvn.transform(thread_compare_mem));
3921 }
3922
3923 #endif // JFR_HAVE_INTRINSICS
3924
3925 //------------------------inline_native_currentCarrierThread------------------
3926 bool LibraryCallKit::inline_native_currentCarrierThread() {
3927 Node* junk = nullptr;
3928 set_result(generate_current_thread(junk));
3929 return true;
3930 }
3931
3932 //------------------------inline_native_currentThread------------------
3933 bool LibraryCallKit::inline_native_currentThread() {
3934 Node* junk = nullptr;
3935 set_result(generate_virtual_thread(junk));
3936 return true;
3937 }
3938
3939 //------------------------inline_native_setVthread------------------
3940 bool LibraryCallKit::inline_native_setCurrentThread() {
3941 assert(C->method()->changes_current_thread(),
3942 "method changes current Thread but is not annotated ChangesCurrentThread");
3943 Node* arr = argument(1);
3944 Node* thread = _gvn.transform(new ThreadLocalNode());
3945 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3946 Node* thread_obj_handle
3947 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3948 thread_obj_handle = _gvn.transform(thread_obj_handle);
3949 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3950 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3951
3952 // Change the _monitor_owner_id of the JavaThread
3953 Node* tid = load_field_from_object(arr, "tid", "J");
3954 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3955 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3956
3957 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3958 return true;
3959 }
3960
3961 const Type* LibraryCallKit::scopedValueCache_type() {
3962 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3963 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3964 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3965
3966 // Because we create the scopedValue cache lazily we have to make the
3967 // type of the result BotPTR.
3968 bool xk = etype->klass_is_exact();
3969 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3970 return objects_type;
3971 }
3972
3973 Node* LibraryCallKit::scopedValueCache_helper() {
3974 Node* thread = _gvn.transform(new ThreadLocalNode());
3975 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3976 // We cannot use immutable_memory() because we might flip onto a
3977 // different carrier thread, at which point we'll need to use that
3978 // carrier thread's cache.
3979 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3980 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3981 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3982 }
3983
3984 //------------------------inline_native_scopedValueCache------------------
3985 bool LibraryCallKit::inline_native_scopedValueCache() {
3986 Node* cache_obj_handle = scopedValueCache_helper();
3987 const Type* objects_type = scopedValueCache_type();
3988 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3989
3990 return true;
3991 }
3992
3993 //------------------------inline_native_setScopedValueCache------------------
3994 bool LibraryCallKit::inline_native_setScopedValueCache() {
3995 Node* arr = argument(0);
3996 Node* cache_obj_handle = scopedValueCache_helper();
3997 const Type* objects_type = scopedValueCache_type();
3998
3999 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4000 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4001
4002 return true;
4003 }
4004
4005 //------------------------inline_native_Continuation_pin and unpin-----------
4006
4007 // Shared implementation routine for both pin and unpin.
4008 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4009 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4010
4011 // Save input memory.
4012 Node* input_memory_state = reset_memory();
4013 set_all_memory(input_memory_state);
4014
4015 // TLS
4016 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4017 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4018 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4019
4020 // Null check the last continuation object.
4021 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4022 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4023 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4024
4025 // False path, last continuation is null.
4026 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4027
4028 // True path, last continuation is not null.
4029 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4030
4031 set_control(continuation_is_not_null);
4032
4033 // Load the pin count from the last continuation.
4034 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4035 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4036
4037 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4038 Node* pin_count_rhs;
4039 if (unpin) {
4040 pin_count_rhs = _gvn.intcon(0);
4041 } else {
4042 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4043 }
4044 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4045 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4046 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4047
4048 // True branch, pin count over/underflow.
4049 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4050 {
4051 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4052 // which will throw IllegalStateException for pin count over/underflow.
4053 // No memory changed so far - we can use memory create by reset_memory()
4054 // at the beginning of this intrinsic. No need to call reset_memory() again.
4055 PreserveJVMState pjvms(this);
4056 set_control(pin_count_over_underflow);
4057 uncommon_trap(Deoptimization::Reason_intrinsic,
4058 Deoptimization::Action_none);
4059 assert(stopped(), "invariant");
4060 }
4061
4062 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4063 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4064 set_control(valid_pin_count);
4065
4066 Node* next_pin_count;
4067 if (unpin) {
4068 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4069 } else {
4070 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4071 }
4072
4073 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4074
4075 // Result of top level CFG and Memory.
4076 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4077 record_for_igvn(result_rgn);
4078 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4079 record_for_igvn(result_mem);
4080
4081 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4082 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4083 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4084 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4085
4086 // Set output state.
4087 set_control(_gvn.transform(result_rgn));
4088 set_all_memory(_gvn.transform(result_mem));
4089
4090 return true;
4091 }
4092
4093 //-----------------------load_klass_from_mirror_common-------------------------
4094 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4095 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4096 // and branch to the given path on the region.
4097 // If never_see_null, take an uncommon trap on null, so we can optimistically
4098 // compile for the non-null case.
4099 // If the region is null, force never_see_null = true.
4100 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4101 bool never_see_null,
4102 RegionNode* region,
4103 int null_path,
4104 int offset) {
4105 if (region == nullptr) never_see_null = true;
4106 Node* p = basic_plus_adr(mirror, offset);
4107 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4108 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4109 Node* null_ctl = top();
4110 kls = null_check_oop(kls, &null_ctl, never_see_null);
4111 if (region != nullptr) {
4112 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4113 region->init_req(null_path, null_ctl);
4114 } else {
4115 assert(null_ctl == top(), "no loose ends");
4116 }
4117 return kls;
4118 }
4119
4120 //--------------------(inline_native_Class_query helpers)---------------------
4121 // Use this for JVM_ACC_INTERFACE.
4122 // Fall through if (mods & mask) == bits, take the guard otherwise.
4123 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4124 ByteSize offset, const Type* type, BasicType bt) {
4125 // Branch around if the given klass has the given modifier bit set.
4126 // Like generate_guard, adds a new path onto the region.
4127 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4128 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4129 Node* mask = intcon(modifier_mask);
4130 Node* bits = intcon(modifier_bits);
4131 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4132 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4133 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4134 return generate_fair_guard(bol, region);
4135 }
4136
4137 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4138 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4139 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4140 }
4141
4142 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4143 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4144 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4145 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4146 }
4147
4148 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4149 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4150 }
4151
4152 //-------------------------inline_native_Class_query-------------------
4153 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4154 const Type* return_type = TypeInt::BOOL;
4155 Node* prim_return_value = top(); // what happens if it's a primitive class?
4156 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4157 bool expect_prim = false; // most of these guys expect to work on refs
4158
4159 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4160
4161 Node* mirror = argument(0);
4162 Node* obj = top();
4163
4164 switch (id) {
4165 case vmIntrinsics::_isInstance:
4166 // nothing is an instance of a primitive type
4167 prim_return_value = intcon(0);
4168 obj = argument(1);
4169 break;
4170 case vmIntrinsics::_isHidden:
4171 prim_return_value = intcon(0);
4172 break;
4173 case vmIntrinsics::_getSuperclass:
4174 prim_return_value = null();
4175 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4176 break;
4177 case vmIntrinsics::_getClassAccessFlags:
4178 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4179 return_type = TypeInt::CHAR;
4180 break;
4181 default:
4182 fatal_unexpected_iid(id);
4183 break;
4184 }
4185
4186 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4187 if (mirror_con == nullptr) return false; // cannot happen?
4188
4189 #ifndef PRODUCT
4190 if (C->print_intrinsics() || C->print_inlining()) {
4191 ciType* k = mirror_con->java_mirror_type();
4192 if (k) {
4193 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4194 k->print_name();
4195 tty->cr();
4196 }
4197 }
4198 #endif
4199
4200 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4201 RegionNode* region = new RegionNode(PATH_LIMIT);
4202 record_for_igvn(region);
4203 PhiNode* phi = new PhiNode(region, return_type);
4204
4205 // The mirror will never be null of Reflection.getClassAccessFlags, however
4206 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4207 // if it is. See bug 4774291.
4208
4209 // For Reflection.getClassAccessFlags(), the null check occurs in
4210 // the wrong place; see inline_unsafe_access(), above, for a similar
4211 // situation.
4212 mirror = null_check(mirror);
4213 // If mirror or obj is dead, only null-path is taken.
4214 if (stopped()) return true;
4215
4216 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4217
4218 // Now load the mirror's klass metaobject, and null-check it.
4219 // Side-effects region with the control path if the klass is null.
4220 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4221 // If kls is null, we have a primitive mirror.
4222 phi->init_req(_prim_path, prim_return_value);
4223 if (stopped()) { set_result(region, phi); return true; }
4224 bool safe_for_replace = (region->in(_prim_path) == top());
4225
4226 Node* p; // handy temp
4227 Node* null_ctl;
4228
4229 // Now that we have the non-null klass, we can perform the real query.
4230 // For constant classes, the query will constant-fold in LoadNode::Value.
4231 Node* query_value = top();
4232 switch (id) {
4233 case vmIntrinsics::_isInstance:
4234 // nothing is an instance of a primitive type
4235 query_value = gen_instanceof(obj, kls, safe_for_replace);
4236 break;
4237
4238 case vmIntrinsics::_isHidden:
4239 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4240 if (generate_hidden_class_guard(kls, region) != nullptr)
4241 // A guard was added. If the guard is taken, it was an hidden class.
4242 phi->add_req(intcon(1));
4243 // If we fall through, it's a plain class.
4244 query_value = intcon(0);
4245 break;
4246
4247
4248 case vmIntrinsics::_getSuperclass:
4249 // The rules here are somewhat unfortunate, but we can still do better
4250 // with random logic than with a JNI call.
4251 // Interfaces store null or Object as _super, but must report null.
4252 // Arrays store an intermediate super as _super, but must report Object.
4253 // Other types can report the actual _super.
4254 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4255 if (generate_interface_guard(kls, region) != nullptr)
4256 // A guard was added. If the guard is taken, it was an interface.
4257 phi->add_req(null());
4258 if (generate_array_guard(kls, region) != nullptr)
4259 // A guard was added. If the guard is taken, it was an array.
4260 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4261 // If we fall through, it's a plain class. Get its _super.
4262 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4263 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4264 null_ctl = top();
4265 kls = null_check_oop(kls, &null_ctl);
4266 if (null_ctl != top()) {
4267 // If the guard is taken, Object.superClass is null (both klass and mirror).
4268 region->add_req(null_ctl);
4269 phi ->add_req(null());
4270 }
4271 if (!stopped()) {
4272 query_value = load_mirror_from_klass(kls);
4273 }
4274 break;
4275
4276 case vmIntrinsics::_getClassAccessFlags:
4277 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4278 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4279 break;
4280
4281 default:
4282 fatal_unexpected_iid(id);
4283 break;
4284 }
4285
4286 // Fall-through is the normal case of a query to a real class.
4287 phi->init_req(1, query_value);
4288 region->init_req(1, control());
4289
4290 C->set_has_split_ifs(true); // Has chance for split-if optimization
4291 set_result(region, phi);
4292 return true;
4293 }
4294
4295
4296 //-------------------------inline_Class_cast-------------------
4297 bool LibraryCallKit::inline_Class_cast() {
4298 Node* mirror = argument(0); // Class
4299 Node* obj = argument(1);
4300 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4301 if (mirror_con == nullptr) {
4302 return false; // dead path (mirror->is_top()).
4303 }
4304 if (obj == nullptr || obj->is_top()) {
4305 return false; // dead path
4306 }
4307 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4308
4309 // First, see if Class.cast() can be folded statically.
4310 // java_mirror_type() returns non-null for compile-time Class constants.
4311 bool is_null_free_array = false;
4312 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4313 if (tm != nullptr && tm->is_klass() &&
4314 tp != nullptr) {
4315 if (!tp->is_loaded()) {
4316 // Don't use intrinsic when class is not loaded.
4317 return false;
4318 } else {
4319 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4320 if (is_null_free_array) {
4321 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4322 }
4323 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4324 if (static_res == Compile::SSC_always_true) {
4325 // isInstance() is true - fold the code.
4326 set_result(obj);
4327 return true;
4328 } else if (static_res == Compile::SSC_always_false) {
4329 // Don't use intrinsic, have to throw ClassCastException.
4330 // If the reference is null, the non-intrinsic bytecode will
4331 // be optimized appropriately.
4332 return false;
4333 }
4334 }
4335 }
4336
4337 // Bailout intrinsic and do normal inlining if exception path is frequent.
4338 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4339 return false;
4340 }
4341
4342 // Generate dynamic checks.
4343 // Class.cast() is java implementation of _checkcast bytecode.
4344 // Do checkcast (Parse::do_checkcast()) optimizations here.
4345
4346 mirror = null_check(mirror);
4347 // If mirror is dead, only null-path is taken.
4348 if (stopped()) {
4349 return true;
4350 }
4351
4352 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4353 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4354 RegionNode* region = new RegionNode(PATH_LIMIT);
4355 record_for_igvn(region);
4356
4357 // Now load the mirror's klass metaobject, and null-check it.
4358 // If kls is null, we have a primitive mirror and
4359 // nothing is an instance of a primitive type.
4360 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4361
4362 Node* res = top();
4363 Node* io = i_o();
4364 Node* mem = merged_memory();
4365 if (!stopped()) {
4366
4367 Node* bad_type_ctrl = top();
4368 // Do checkcast optimizations.
4369 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4370 region->init_req(_bad_type_path, bad_type_ctrl);
4371 }
4372 if (region->in(_prim_path) != top() ||
4373 region->in(_bad_type_path) != top() ||
4374 region->in(_npe_path) != top()) {
4375 // Let Interpreter throw ClassCastException.
4376 PreserveJVMState pjvms(this);
4377 set_control(_gvn.transform(region));
4378 // Set IO and memory because gen_checkcast may override them when buffering inline types
4379 set_i_o(io);
4380 set_all_memory(mem);
4381 uncommon_trap(Deoptimization::Reason_intrinsic,
4382 Deoptimization::Action_maybe_recompile);
4383 }
4384 if (!stopped()) {
4385 set_result(res);
4386 }
4387 return true;
4388 }
4389
4390
4391 //--------------------------inline_native_subtype_check------------------------
4392 // This intrinsic takes the JNI calls out of the heart of
4393 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4394 bool LibraryCallKit::inline_native_subtype_check() {
4395 // Pull both arguments off the stack.
4396 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4397 args[0] = argument(0);
4398 args[1] = argument(1);
4399 Node* klasses[2]; // corresponding Klasses: superk, subk
4400 klasses[0] = klasses[1] = top();
4401
4402 enum {
4403 // A full decision tree on {superc is prim, subc is prim}:
4404 _prim_0_path = 1, // {P,N} => false
4405 // {P,P} & superc!=subc => false
4406 _prim_same_path, // {P,P} & superc==subc => true
4407 _prim_1_path, // {N,P} => false
4408 _ref_subtype_path, // {N,N} & subtype check wins => true
4409 _both_ref_path, // {N,N} & subtype check loses => false
4410 PATH_LIMIT
4411 };
4412
4413 RegionNode* region = new RegionNode(PATH_LIMIT);
4414 RegionNode* prim_region = new RegionNode(2);
4415 Node* phi = new PhiNode(region, TypeInt::BOOL);
4416 record_for_igvn(region);
4417 record_for_igvn(prim_region);
4418
4419 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4420 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4421 int class_klass_offset = java_lang_Class::klass_offset();
4422
4423 // First null-check both mirrors and load each mirror's klass metaobject.
4424 int which_arg;
4425 for (which_arg = 0; which_arg <= 1; which_arg++) {
4426 Node* arg = args[which_arg];
4427 arg = null_check(arg);
4428 if (stopped()) break;
4429 args[which_arg] = arg;
4430
4431 Node* p = basic_plus_adr(arg, class_klass_offset);
4432 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4433 klasses[which_arg] = _gvn.transform(kls);
4434 }
4435
4436 // Having loaded both klasses, test each for null.
4437 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4438 for (which_arg = 0; which_arg <= 1; which_arg++) {
4439 Node* kls = klasses[which_arg];
4440 Node* null_ctl = top();
4441 kls = null_check_oop(kls, &null_ctl, never_see_null);
4442 if (which_arg == 0) {
4443 prim_region->init_req(1, null_ctl);
4444 } else {
4445 region->init_req(_prim_1_path, null_ctl);
4446 }
4447 if (stopped()) break;
4448 klasses[which_arg] = kls;
4449 }
4450
4451 if (!stopped()) {
4452 // now we have two reference types, in klasses[0..1]
4453 Node* subk = klasses[1]; // the argument to isAssignableFrom
4454 Node* superk = klasses[0]; // the receiver
4455 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4456 region->set_req(_ref_subtype_path, control());
4457 }
4458
4459 // If both operands are primitive (both klasses null), then
4460 // we must return true when they are identical primitives.
4461 // It is convenient to test this after the first null klass check.
4462 // This path is also used if superc is a value mirror.
4463 set_control(_gvn.transform(prim_region));
4464 if (!stopped()) {
4465 // Since superc is primitive, make a guard for the superc==subc case.
4466 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4467 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4468 generate_fair_guard(bol_eq, region);
4469 if (region->req() == PATH_LIMIT+1) {
4470 // A guard was added. If the added guard is taken, superc==subc.
4471 region->swap_edges(PATH_LIMIT, _prim_same_path);
4472 region->del_req(PATH_LIMIT);
4473 }
4474 region->set_req(_prim_0_path, control()); // Not equal after all.
4475 }
4476
4477 // these are the only paths that produce 'true':
4478 phi->set_req(_prim_same_path, intcon(1));
4479 phi->set_req(_ref_subtype_path, intcon(1));
4480
4481 // pull together the cases:
4482 assert(region->req() == PATH_LIMIT, "sane region");
4483 for (uint i = 1; i < region->req(); i++) {
4484 Node* ctl = region->in(i);
4485 if (ctl == nullptr || ctl == top()) {
4486 region->set_req(i, top());
4487 phi ->set_req(i, top());
4488 } else if (phi->in(i) == nullptr) {
4489 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4490 }
4491 }
4492
4493 set_control(_gvn.transform(region));
4494 set_result(_gvn.transform(phi));
4495 return true;
4496 }
4497
4498 //---------------------generate_array_guard_common------------------------
4499 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4500
4501 if (stopped()) {
4502 return nullptr;
4503 }
4504
4505 // Like generate_guard, adds a new path onto the region.
4506 jint layout_con = 0;
4507 Node* layout_val = get_layout_helper(kls, layout_con);
4508 if (layout_val == nullptr) {
4509 bool query = 0;
4510 switch(kind) {
4511 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4512 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4513 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4514 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4515 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4516 default:
4517 ShouldNotReachHere();
4518 }
4519 if (!query) {
4520 return nullptr; // never a branch
4521 } else { // always a branch
4522 Node* always_branch = control();
4523 if (region != nullptr)
4524 region->add_req(always_branch);
4525 set_control(top());
4526 return always_branch;
4527 }
4528 }
4529 unsigned int value = 0;
4530 BoolTest::mask btest = BoolTest::illegal;
4531 switch(kind) {
4532 case ObjectArray:
4533 case NonObjectArray: {
4534 value = Klass::_lh_array_tag_obj_value;
4535 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4536 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4537 break;
4538 }
4539 case TypeArray: {
4540 value = Klass::_lh_array_tag_type_value;
4541 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4542 btest = BoolTest::eq;
4543 break;
4544 }
4545 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4546 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4547 default:
4548 ShouldNotReachHere();
4549 }
4550 // Now test the correct condition.
4551 jint nval = (jint)value;
4552 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4553 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4554 Node* ctrl = generate_fair_guard(bol, region);
4555 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4556 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4557 // Keep track of the fact that 'obj' is an array to prevent
4558 // array specific accesses from floating above the guard.
4559 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4560 }
4561 return ctrl;
4562 }
4563
4564 // public static native Object[] newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4565 // public static native Object[] newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4566 // public static native Object[] newNullableAtomicArray(Class<?> componentType, int length);
4567 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4568 assert(null_free || atomic, "nullable implies atomic");
4569 Node* componentType = argument(0);
4570 Node* length = argument(1);
4571 Node* init_val = null_free ? argument(2) : nullptr;
4572
4573 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4574 if (tp != nullptr) {
4575 ciInstanceKlass* ik = tp->instance_klass();
4576 if (ik == C->env()->Class_klass()) {
4577 ciType* t = tp->java_mirror_type();
4578 if (t != nullptr && t->is_inlinetype()) {
4579 ciInlineKlass* vk = t->as_inline_klass();
4580 bool flat = vk->maybe_flat_in_array();
4581 if (flat && atomic) {
4582 // Only flat if we have a corresponding atomic layout
4583 flat = null_free ? vk->has_atomic_layout() : vk->has_nullable_atomic_layout();
4584 }
4585 // TODO 8350865 refactor
4586 if (flat && !atomic) {
4587 flat = vk->has_non_atomic_layout();
4588 }
4589
4590 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4591 if (UseZGC && null_free && !flat) {
4592 return false;
4593 }
4594
4595 ciArrayKlass* array_klass = ciArrayKlass::make(t, flat, null_free, atomic);
4596 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4597 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4598 if (null_free) {
4599 if (init_val->is_InlineType()) {
4600 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4601 // Zeroing is enough because the init value is the all-zero value
4602 init_val = nullptr;
4603 } else {
4604 init_val = init_val->as_InlineType()->buffer(this);
4605 }
4606 }
4607 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4608 }
4609 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4610 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4611 assert(arytype->is_null_free() == null_free, "inconsistency");
4612 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4613 assert(arytype->is_flat() == flat, "inconsistency");
4614 assert(arytype->is_aryptr()->is_not_flat() == !flat, "inconsistency");
4615 set_result(obj);
4616 return true;
4617 }
4618 }
4619 }
4620 }
4621 return false;
4622 }
4623
4624 //-----------------------inline_native_newArray--------------------------
4625 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4626 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4627 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4628 Node* mirror;
4629 Node* count_val;
4630 if (uninitialized) {
4631 null_check_receiver();
4632 mirror = argument(1);
4633 count_val = argument(2);
4634 } else {
4635 mirror = argument(0);
4636 count_val = argument(1);
4637 }
4638
4639 mirror = null_check(mirror);
4640 // If mirror or obj is dead, only null-path is taken.
4641 if (stopped()) return true;
4642
4643 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4644 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4645 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4646 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4647 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4648
4649 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4650 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4651 result_reg, _slow_path);
4652 Node* normal_ctl = control();
4653 Node* no_array_ctl = result_reg->in(_slow_path);
4654
4655 // Generate code for the slow case. We make a call to newArray().
4656 set_control(no_array_ctl);
4657 if (!stopped()) {
4658 // Either the input type is void.class, or else the
4659 // array klass has not yet been cached. Either the
4660 // ensuing call will throw an exception, or else it
4661 // will cache the array klass for next time.
4662 PreserveJVMState pjvms(this);
4663 CallJavaNode* slow_call = nullptr;
4664 if (uninitialized) {
4665 // Generate optimized virtual call (holder class 'Unsafe' is final)
4666 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4667 } else {
4668 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4669 }
4670 Node* slow_result = set_results_for_java_call(slow_call);
4671 // this->control() comes from set_results_for_java_call
4672 result_reg->set_req(_slow_path, control());
4673 result_val->set_req(_slow_path, slow_result);
4674 result_io ->set_req(_slow_path, i_o());
4675 result_mem->set_req(_slow_path, reset_memory());
4676 }
4677
4678 set_control(normal_ctl);
4679 if (!stopped()) {
4680 // Normal case: The array type has been cached in the java.lang.Class.
4681 // The following call works fine even if the array type is polymorphic.
4682 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4683 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4684 result_reg->init_req(_normal_path, control());
4685 result_val->init_req(_normal_path, obj);
4686 result_io ->init_req(_normal_path, i_o());
4687 result_mem->init_req(_normal_path, reset_memory());
4688
4689 if (uninitialized) {
4690 // Mark the allocation so that zeroing is skipped
4691 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4692 alloc->maybe_set_complete(&_gvn);
4693 }
4694 }
4695
4696 // Return the combined state.
4697 set_i_o( _gvn.transform(result_io) );
4698 set_all_memory( _gvn.transform(result_mem));
4699
4700 C->set_has_split_ifs(true); // Has chance for split-if optimization
4701 set_result(result_reg, result_val);
4702 return true;
4703 }
4704
4705 //----------------------inline_native_getLength--------------------------
4706 // public static native int java.lang.reflect.Array.getLength(Object array);
4707 bool LibraryCallKit::inline_native_getLength() {
4708 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4709
4710 Node* array = null_check(argument(0));
4711 // If array is dead, only null-path is taken.
4712 if (stopped()) return true;
4713
4714 // Deoptimize if it is a non-array.
4715 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4716
4717 if (non_array != nullptr) {
4718 PreserveJVMState pjvms(this);
4719 set_control(non_array);
4720 uncommon_trap(Deoptimization::Reason_intrinsic,
4721 Deoptimization::Action_maybe_recompile);
4722 }
4723
4724 // If control is dead, only non-array-path is taken.
4725 if (stopped()) return true;
4726
4727 // The works fine even if the array type is polymorphic.
4728 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4729 Node* result = load_array_length(array);
4730
4731 C->set_has_split_ifs(true); // Has chance for split-if optimization
4732 set_result(result);
4733 return true;
4734 }
4735
4736 //------------------------inline_array_copyOf----------------------------
4737 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4738 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4739 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4740 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4741
4742 // Get the arguments.
4743 Node* original = argument(0);
4744 Node* start = is_copyOfRange? argument(1): intcon(0);
4745 Node* end = is_copyOfRange? argument(2): argument(1);
4746 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4747
4748 Node* newcopy = nullptr;
4749
4750 // Set the original stack and the reexecute bit for the interpreter to reexecute
4751 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4752 { PreserveReexecuteState preexecs(this);
4753 jvms()->set_should_reexecute(true);
4754
4755 array_type_mirror = null_check(array_type_mirror);
4756 original = null_check(original);
4757
4758 // Check if a null path was taken unconditionally.
4759 if (stopped()) return true;
4760
4761 Node* orig_length = load_array_length(original);
4762
4763 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4764 klass_node = null_check(klass_node);
4765
4766 RegionNode* bailout = new RegionNode(1);
4767 record_for_igvn(bailout);
4768
4769 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4770 // Bail out if that is so.
4771 // Inline type array may have object field that would require a
4772 // write barrier. Conservatively, go to slow path.
4773 // TODO 8251971: Optimize for the case when flat src/dst are later found
4774 // to not contain oops (i.e., move this check to the macro expansion phase).
4775 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4776 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4777 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4778 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4779 // Can src array be flat and contain oops?
4780 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4781 // Can dest array be flat and contain oops?
4782 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4783 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4784 if (not_objArray != nullptr) {
4785 // Improve the klass node's type from the new optimistic assumption:
4786 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4787 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4788 Node* cast = new CastPPNode(control(), klass_node, akls);
4789 klass_node = _gvn.transform(cast);
4790 }
4791
4792 // Bail out if either start or end is negative.
4793 generate_negative_guard(start, bailout, &start);
4794 generate_negative_guard(end, bailout, &end);
4795
4796 Node* length = end;
4797 if (_gvn.type(start) != TypeInt::ZERO) {
4798 length = _gvn.transform(new SubINode(end, start));
4799 }
4800
4801 // Bail out if length is negative (i.e., if start > end).
4802 // Without this the new_array would throw
4803 // NegativeArraySizeException but IllegalArgumentException is what
4804 // should be thrown
4805 generate_negative_guard(length, bailout, &length);
4806
4807 // Handle inline type arrays
4808 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4809 if (!stopped()) {
4810 // TODO JDK-8329224
4811 if (!orig_t->is_null_free()) {
4812 // Not statically known to be null free, add a check
4813 generate_fair_guard(null_free_array_test(original), bailout);
4814 }
4815 orig_t = _gvn.type(original)->isa_aryptr();
4816 if (orig_t != nullptr && orig_t->is_flat()) {
4817 // Src is flat, check that dest is flat as well
4818 if (exclude_flat) {
4819 // Dest can't be flat, bail out
4820 bailout->add_req(control());
4821 set_control(top());
4822 } else {
4823 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4824 }
4825 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
4826 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4827 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4828 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4829 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4830 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4831 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4832 if (orig_t != nullptr) {
4833 orig_t = orig_t->cast_to_not_flat();
4834 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4835 }
4836 }
4837 if (!can_validate) {
4838 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4839 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4840 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4841 generate_fair_guard(flat_array_test(klass_node), bailout);
4842 generate_fair_guard(null_free_array_test(original), bailout);
4843 }
4844 }
4845
4846 // Bail out if start is larger than the original length
4847 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4848 generate_negative_guard(orig_tail, bailout, &orig_tail);
4849
4850 if (bailout->req() > 1) {
4851 PreserveJVMState pjvms(this);
4852 set_control(_gvn.transform(bailout));
4853 uncommon_trap(Deoptimization::Reason_intrinsic,
4854 Deoptimization::Action_maybe_recompile);
4855 }
4856
4857 if (!stopped()) {
4858 // How many elements will we copy from the original?
4859 // The answer is MinI(orig_tail, length).
4860 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4861
4862 // Generate a direct call to the right arraycopy function(s).
4863 // We know the copy is disjoint but we might not know if the
4864 // oop stores need checking.
4865 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4866 // This will fail a store-check if x contains any non-nulls.
4867
4868 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4869 // loads/stores but it is legal only if we're sure the
4870 // Arrays.copyOf would succeed. So we need all input arguments
4871 // to the copyOf to be validated, including that the copy to the
4872 // new array won't trigger an ArrayStoreException. That subtype
4873 // check can be optimized if we know something on the type of
4874 // the input array from type speculation.
4875 if (_gvn.type(klass_node)->singleton()) {
4876 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4877 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4878
4879 int test = C->static_subtype_check(superk, subk);
4880 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4881 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4882 if (t_original->speculative_type() != nullptr) {
4883 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4884 }
4885 }
4886 }
4887
4888 bool validated = false;
4889 // Reason_class_check rather than Reason_intrinsic because we
4890 // want to intrinsify even if this traps.
4891 if (can_validate) {
4892 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4893
4894 if (not_subtype_ctrl != top()) {
4895 PreserveJVMState pjvms(this);
4896 set_control(not_subtype_ctrl);
4897 uncommon_trap(Deoptimization::Reason_class_check,
4898 Deoptimization::Action_make_not_entrant);
4899 assert(stopped(), "Should be stopped");
4900 }
4901 validated = true;
4902 }
4903
4904 if (!stopped()) {
4905 newcopy = new_array(klass_node, length, 0); // no arguments to push
4906
4907 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4908 load_object_klass(original), klass_node);
4909 if (!is_copyOfRange) {
4910 ac->set_copyof(validated);
4911 } else {
4912 ac->set_copyofrange(validated);
4913 }
4914 Node* n = _gvn.transform(ac);
4915 if (n == ac) {
4916 ac->connect_outputs(this);
4917 } else {
4918 assert(validated, "shouldn't transform if all arguments not validated");
4919 set_all_memory(n);
4920 }
4921 }
4922 }
4923 } // original reexecute is set back here
4924
4925 C->set_has_split_ifs(true); // Has chance for split-if optimization
4926 if (!stopped()) {
4927 set_result(newcopy);
4928 }
4929 return true;
4930 }
4931
4932
4933 //----------------------generate_virtual_guard---------------------------
4934 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4935 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4936 RegionNode* slow_region) {
4937 ciMethod* method = callee();
4938 int vtable_index = method->vtable_index();
4939 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4940 "bad index %d", vtable_index);
4941 // Get the Method* out of the appropriate vtable entry.
4942 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4943 vtable_index*vtableEntry::size_in_bytes() +
4944 in_bytes(vtableEntry::method_offset());
4945 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4946 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4947
4948 // Compare the target method with the expected method (e.g., Object.hashCode).
4949 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4950
4951 Node* native_call = makecon(native_call_addr);
4952 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4953 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4954
4955 return generate_slow_guard(test_native, slow_region);
4956 }
4957
4958 //-----------------------generate_method_call----------------------------
4959 // Use generate_method_call to make a slow-call to the real
4960 // method if the fast path fails. An alternative would be to
4961 // use a stub like OptoRuntime::slow_arraycopy_Java.
4962 // This only works for expanding the current library call,
4963 // not another intrinsic. (E.g., don't use this for making an
4964 // arraycopy call inside of the copyOf intrinsic.)
4965 CallJavaNode*
4966 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4967 // When compiling the intrinsic method itself, do not use this technique.
4968 guarantee(callee() != C->method(), "cannot make slow-call to self");
4969
4970 ciMethod* method = callee();
4971 // ensure the JVMS we have will be correct for this call
4972 guarantee(method_id == method->intrinsic_id(), "must match");
4973
4974 const TypeFunc* tf = TypeFunc::make(method);
4975 if (res_not_null) {
4976 assert(tf->return_type() == T_OBJECT, "");
4977 const TypeTuple* range = tf->range_cc();
4978 const Type** fields = TypeTuple::fields(range->cnt());
4979 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4980 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4981 tf = TypeFunc::make(tf->domain_cc(), new_range);
4982 }
4983 CallJavaNode* slow_call;
4984 if (is_static) {
4985 assert(!is_virtual, "");
4986 slow_call = new CallStaticJavaNode(C, tf,
4987 SharedRuntime::get_resolve_static_call_stub(), method);
4988 } else if (is_virtual) {
4989 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4990 int vtable_index = Method::invalid_vtable_index;
4991 if (UseInlineCaches) {
4992 // Suppress the vtable call
4993 } else {
4994 // hashCode and clone are not a miranda methods,
4995 // so the vtable index is fixed.
4996 // No need to use the linkResolver to get it.
4997 vtable_index = method->vtable_index();
4998 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4999 "bad index %d", vtable_index);
5000 }
5001 slow_call = new CallDynamicJavaNode(tf,
5002 SharedRuntime::get_resolve_virtual_call_stub(),
5003 method, vtable_index);
5004 } else { // neither virtual nor static: opt_virtual
5005 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5006 slow_call = new CallStaticJavaNode(C, tf,
5007 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5008 slow_call->set_optimized_virtual(true);
5009 }
5010 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5011 // To be able to issue a direct call (optimized virtual or virtual)
5012 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5013 // about the method being invoked should be attached to the call site to
5014 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5015 slow_call->set_override_symbolic_info(true);
5016 }
5017 set_arguments_for_java_call(slow_call);
5018 set_edges_for_java_call(slow_call);
5019 return slow_call;
5020 }
5021
5022
5023 /**
5024 * Build special case code for calls to hashCode on an object. This call may
5025 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5026 * slightly different code.
5027 */
5028 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5029 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5030 assert(!(is_virtual && is_static), "either virtual, special, or static");
5031
5032 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5033
5034 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5035 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5036 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5037 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5038 Node* obj = argument(0);
5039
5040 // Don't intrinsify hashcode on inline types for now.
5041 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
5042 if (gvn().type(obj)->is_inlinetypeptr()) {
5043 return false;
5044 }
5045
5046 if (!is_static) {
5047 // Check for hashing null object
5048 obj = null_check_receiver();
5049 if (stopped()) return true; // unconditionally null
5050 result_reg->init_req(_null_path, top());
5051 result_val->init_req(_null_path, top());
5052 } else {
5053 // Do a null check, and return zero if null.
5054 // System.identityHashCode(null) == 0
5055 Node* null_ctl = top();
5056 obj = null_check_oop(obj, &null_ctl);
5057 result_reg->init_req(_null_path, null_ctl);
5058 result_val->init_req(_null_path, _gvn.intcon(0));
5059 }
5060
5061 // Unconditionally null? Then return right away.
5062 if (stopped()) {
5063 set_control( result_reg->in(_null_path));
5064 if (!stopped())
5065 set_result(result_val->in(_null_path));
5066 return true;
5067 }
5068
5069 // We only go to the fast case code if we pass a number of guards. The
5070 // paths which do not pass are accumulated in the slow_region.
5071 RegionNode* slow_region = new RegionNode(1);
5072 record_for_igvn(slow_region);
5073
5074 // If this is a virtual call, we generate a funny guard. We pull out
5075 // the vtable entry corresponding to hashCode() from the target object.
5076 // If the target method which we are calling happens to be the native
5077 // Object hashCode() method, we pass the guard. We do not need this
5078 // guard for non-virtual calls -- the caller is known to be the native
5079 // Object hashCode().
5080 if (is_virtual) {
5081 // After null check, get the object's klass.
5082 Node* obj_klass = load_object_klass(obj);
5083 generate_virtual_guard(obj_klass, slow_region);
5084 }
5085
5086 // Get the header out of the object, use LoadMarkNode when available
5087 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5088 // The control of the load must be null. Otherwise, the load can move before
5089 // the null check after castPP removal.
5090 Node* no_ctrl = nullptr;
5091 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5092
5093 if (!UseObjectMonitorTable) {
5094 // Test the header to see if it is safe to read w.r.t. locking.
5095 // This also serves as guard against inline types
5096 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5097 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5098 if (LockingMode == LM_LIGHTWEIGHT) {
5099 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5100 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5101 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5102
5103 generate_slow_guard(test_monitor, slow_region);
5104 } else {
5105 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5106 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5107 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5108
5109 generate_slow_guard(test_not_unlocked, slow_region);
5110 }
5111 }
5112
5113 // Get the hash value and check to see that it has been properly assigned.
5114 // We depend on hash_mask being at most 32 bits and avoid the use of
5115 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5116 // vm: see markWord.hpp.
5117 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5118 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5119 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5120 // This hack lets the hash bits live anywhere in the mark object now, as long
5121 // as the shift drops the relevant bits into the low 32 bits. Note that
5122 // Java spec says that HashCode is an int so there's no point in capturing
5123 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5124 hshifted_header = ConvX2I(hshifted_header);
5125 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5126
5127 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5128 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5129 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5130
5131 generate_slow_guard(test_assigned, slow_region);
5132
5133 Node* init_mem = reset_memory();
5134 // fill in the rest of the null path:
5135 result_io ->init_req(_null_path, i_o());
5136 result_mem->init_req(_null_path, init_mem);
5137
5138 result_val->init_req(_fast_path, hash_val);
5139 result_reg->init_req(_fast_path, control());
5140 result_io ->init_req(_fast_path, i_o());
5141 result_mem->init_req(_fast_path, init_mem);
5142
5143 // Generate code for the slow case. We make a call to hashCode().
5144 set_control(_gvn.transform(slow_region));
5145 if (!stopped()) {
5146 // No need for PreserveJVMState, because we're using up the present state.
5147 set_all_memory(init_mem);
5148 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5149 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5150 Node* slow_result = set_results_for_java_call(slow_call);
5151 // this->control() comes from set_results_for_java_call
5152 result_reg->init_req(_slow_path, control());
5153 result_val->init_req(_slow_path, slow_result);
5154 result_io ->set_req(_slow_path, i_o());
5155 result_mem ->set_req(_slow_path, reset_memory());
5156 }
5157
5158 // Return the combined state.
5159 set_i_o( _gvn.transform(result_io) );
5160 set_all_memory( _gvn.transform(result_mem));
5161
5162 set_result(result_reg, result_val);
5163 return true;
5164 }
5165
5166 //---------------------------inline_native_getClass----------------------------
5167 // public final native Class<?> java.lang.Object.getClass();
5168 //
5169 // Build special case code for calls to getClass on an object.
5170 bool LibraryCallKit::inline_native_getClass() {
5171 Node* obj = argument(0);
5172 if (obj->is_InlineType()) {
5173 const Type* t = _gvn.type(obj);
5174 if (t->maybe_null()) {
5175 null_check(obj);
5176 }
5177 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5178 return true;
5179 }
5180 obj = null_check_receiver();
5181 if (stopped()) return true;
5182 set_result(load_mirror_from_klass(load_object_klass(obj)));
5183 return true;
5184 }
5185
5186 //-----------------inline_native_Reflection_getCallerClass---------------------
5187 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5188 //
5189 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5190 //
5191 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5192 // in that it must skip particular security frames and checks for
5193 // caller sensitive methods.
5194 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5195 #ifndef PRODUCT
5196 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5197 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5198 }
5199 #endif
5200
5201 if (!jvms()->has_method()) {
5202 #ifndef PRODUCT
5203 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5204 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5205 }
5206 #endif
5207 return false;
5208 }
5209
5210 // Walk back up the JVM state to find the caller at the required
5211 // depth.
5212 JVMState* caller_jvms = jvms();
5213
5214 // Cf. JVM_GetCallerClass
5215 // NOTE: Start the loop at depth 1 because the current JVM state does
5216 // not include the Reflection.getCallerClass() frame.
5217 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5218 ciMethod* m = caller_jvms->method();
5219 switch (n) {
5220 case 0:
5221 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5222 break;
5223 case 1:
5224 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5225 if (!m->caller_sensitive()) {
5226 #ifndef PRODUCT
5227 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5228 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5229 }
5230 #endif
5231 return false; // bail-out; let JVM_GetCallerClass do the work
5232 }
5233 break;
5234 default:
5235 if (!m->is_ignored_by_security_stack_walk()) {
5236 // We have reached the desired frame; return the holder class.
5237 // Acquire method holder as java.lang.Class and push as constant.
5238 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5239 ciInstance* caller_mirror = caller_klass->java_mirror();
5240 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5241
5242 #ifndef PRODUCT
5243 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5244 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());
5245 tty->print_cr(" JVM state at this point:");
5246 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5247 ciMethod* m = jvms()->of_depth(i)->method();
5248 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5249 }
5250 }
5251 #endif
5252 return true;
5253 }
5254 break;
5255 }
5256 }
5257
5258 #ifndef PRODUCT
5259 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5260 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5261 tty->print_cr(" JVM state at this point:");
5262 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5263 ciMethod* m = jvms()->of_depth(i)->method();
5264 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5265 }
5266 }
5267 #endif
5268
5269 return false; // bail-out; let JVM_GetCallerClass do the work
5270 }
5271
5272 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5273 Node* arg = argument(0);
5274 Node* result = nullptr;
5275
5276 switch (id) {
5277 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5278 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5279 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5280 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5281 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5282 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5283
5284 case vmIntrinsics::_doubleToLongBits: {
5285 // two paths (plus control) merge in a wood
5286 RegionNode *r = new RegionNode(3);
5287 Node *phi = new PhiNode(r, TypeLong::LONG);
5288
5289 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5290 // Build the boolean node
5291 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5292
5293 // Branch either way.
5294 // NaN case is less traveled, which makes all the difference.
5295 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5296 Node *opt_isnan = _gvn.transform(ifisnan);
5297 assert( opt_isnan->is_If(), "Expect an IfNode");
5298 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5299 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5300
5301 set_control(iftrue);
5302
5303 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5304 Node *slow_result = longcon(nan_bits); // return NaN
5305 phi->init_req(1, _gvn.transform( slow_result ));
5306 r->init_req(1, iftrue);
5307
5308 // Else fall through
5309 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5310 set_control(iffalse);
5311
5312 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5313 r->init_req(2, iffalse);
5314
5315 // Post merge
5316 set_control(_gvn.transform(r));
5317 record_for_igvn(r);
5318
5319 C->set_has_split_ifs(true); // Has chance for split-if optimization
5320 result = phi;
5321 assert(result->bottom_type()->isa_long(), "must be");
5322 break;
5323 }
5324
5325 case vmIntrinsics::_floatToIntBits: {
5326 // two paths (plus control) merge in a wood
5327 RegionNode *r = new RegionNode(3);
5328 Node *phi = new PhiNode(r, TypeInt::INT);
5329
5330 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5331 // Build the boolean node
5332 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5333
5334 // Branch either way.
5335 // NaN case is less traveled, which makes all the difference.
5336 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5337 Node *opt_isnan = _gvn.transform(ifisnan);
5338 assert( opt_isnan->is_If(), "Expect an IfNode");
5339 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5340 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5341
5342 set_control(iftrue);
5343
5344 static const jint nan_bits = 0x7fc00000;
5345 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5346 phi->init_req(1, _gvn.transform( slow_result ));
5347 r->init_req(1, iftrue);
5348
5349 // Else fall through
5350 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5351 set_control(iffalse);
5352
5353 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5354 r->init_req(2, iffalse);
5355
5356 // Post merge
5357 set_control(_gvn.transform(r));
5358 record_for_igvn(r);
5359
5360 C->set_has_split_ifs(true); // Has chance for split-if optimization
5361 result = phi;
5362 assert(result->bottom_type()->isa_int(), "must be");
5363 break;
5364 }
5365
5366 default:
5367 fatal_unexpected_iid(id);
5368 break;
5369 }
5370 set_result(_gvn.transform(result));
5371 return true;
5372 }
5373
5374 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5375 Node* arg = argument(0);
5376 Node* result = nullptr;
5377
5378 switch (id) {
5379 case vmIntrinsics::_floatIsInfinite:
5380 result = new IsInfiniteFNode(arg);
5381 break;
5382 case vmIntrinsics::_floatIsFinite:
5383 result = new IsFiniteFNode(arg);
5384 break;
5385 case vmIntrinsics::_doubleIsInfinite:
5386 result = new IsInfiniteDNode(arg);
5387 break;
5388 case vmIntrinsics::_doubleIsFinite:
5389 result = new IsFiniteDNode(arg);
5390 break;
5391 default:
5392 fatal_unexpected_iid(id);
5393 break;
5394 }
5395 set_result(_gvn.transform(result));
5396 return true;
5397 }
5398
5399 //----------------------inline_unsafe_copyMemory-------------------------
5400 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5401
5402 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5403 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5404 const Type* base_t = gvn.type(base);
5405
5406 bool in_native = (base_t == TypePtr::NULL_PTR);
5407 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5408 bool is_mixed = !in_heap && !in_native;
5409
5410 if (is_mixed) {
5411 return true; // mixed accesses can touch both on-heap and off-heap memory
5412 }
5413 if (in_heap) {
5414 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5415 if (!is_prim_array) {
5416 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5417 // there's not enough type information available to determine proper memory slice for it.
5418 return true;
5419 }
5420 }
5421 return false;
5422 }
5423
5424 bool LibraryCallKit::inline_unsafe_copyMemory() {
5425 if (callee()->is_static()) return false; // caller must have the capability!
5426 null_check_receiver(); // null-check receiver
5427 if (stopped()) return true;
5428
5429 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5430
5431 Node* src_base = argument(1); // type: oop
5432 Node* src_off = ConvL2X(argument(2)); // type: long
5433 Node* dst_base = argument(4); // type: oop
5434 Node* dst_off = ConvL2X(argument(5)); // type: long
5435 Node* size = ConvL2X(argument(7)); // type: long
5436
5437 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5438 "fieldOffset must be byte-scaled");
5439
5440 Node* src_addr = make_unsafe_address(src_base, src_off);
5441 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5442
5443 Node* thread = _gvn.transform(new ThreadLocalNode());
5444 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5445 BasicType doing_unsafe_access_bt = T_BYTE;
5446 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5447
5448 // update volatile field
5449 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5450
5451 int flags = RC_LEAF | RC_NO_FP;
5452
5453 const TypePtr* dst_type = TypePtr::BOTTOM;
5454
5455 // Adjust memory effects of the runtime call based on input values.
5456 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5457 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5458 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5459
5460 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5461 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5462 flags |= RC_NARROW_MEM; // narrow in memory
5463 }
5464 }
5465
5466 // Call it. Note that the length argument is not scaled.
5467 make_runtime_call(flags,
5468 OptoRuntime::fast_arraycopy_Type(),
5469 StubRoutines::unsafe_arraycopy(),
5470 "unsafe_arraycopy",
5471 dst_type,
5472 src_addr, dst_addr, size XTOP);
5473
5474 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5475
5476 return true;
5477 }
5478
5479 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5480 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5481 bool LibraryCallKit::inline_unsafe_setMemory() {
5482 if (callee()->is_static()) return false; // caller must have the capability!
5483 null_check_receiver(); // null-check receiver
5484 if (stopped()) return true;
5485
5486 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5487
5488 Node* dst_base = argument(1); // type: oop
5489 Node* dst_off = ConvL2X(argument(2)); // type: long
5490 Node* size = ConvL2X(argument(4)); // type: long
5491 Node* byte = argument(6); // type: byte
5492
5493 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5494 "fieldOffset must be byte-scaled");
5495
5496 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5497
5498 Node* thread = _gvn.transform(new ThreadLocalNode());
5499 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5500 BasicType doing_unsafe_access_bt = T_BYTE;
5501 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5502
5503 // update volatile field
5504 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5505
5506 int flags = RC_LEAF | RC_NO_FP;
5507
5508 const TypePtr* dst_type = TypePtr::BOTTOM;
5509
5510 // Adjust memory effects of the runtime call based on input values.
5511 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5512 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5513
5514 flags |= RC_NARROW_MEM; // narrow in memory
5515 }
5516
5517 // Call it. Note that the length argument is not scaled.
5518 make_runtime_call(flags,
5519 OptoRuntime::unsafe_setmemory_Type(),
5520 StubRoutines::unsafe_setmemory(),
5521 "unsafe_setmemory",
5522 dst_type,
5523 dst_addr, size XTOP, byte);
5524
5525 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5526
5527 return true;
5528 }
5529
5530 #undef XTOP
5531
5532 //----------------------inline_unsafe_isFlatArray------------------------
5533 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5534 // This intrinsic exploits assumptions made by the native implementation
5535 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5536 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5537 Node* cls = argument(1);
5538 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5539 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p,
5540 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5541 Node* result = flat_array_test(kls);
5542 set_result(result);
5543 return true;
5544 }
5545
5546 //------------------------clone_coping-----------------------------------
5547 // Helper function for inline_native_clone.
5548 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5549 assert(obj_size != nullptr, "");
5550 Node* raw_obj = alloc_obj->in(1);
5551 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5552
5553 AllocateNode* alloc = nullptr;
5554 if (ReduceBulkZeroing &&
5555 // If we are implementing an array clone without knowing its source type
5556 // (can happen when compiling the array-guarded branch of a reflective
5557 // Object.clone() invocation), initialize the array within the allocation.
5558 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5559 // to a runtime clone call that assumes fully initialized source arrays.
5560 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5561 // We will be completely responsible for initializing this object -
5562 // mark Initialize node as complete.
5563 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5564 // The object was just allocated - there should be no any stores!
5565 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5566 // Mark as complete_with_arraycopy so that on AllocateNode
5567 // expansion, we know this AllocateNode is initialized by an array
5568 // copy and a StoreStore barrier exists after the array copy.
5569 alloc->initialization()->set_complete_with_arraycopy();
5570 }
5571
5572 Node* size = _gvn.transform(obj_size);
5573 access_clone(obj, alloc_obj, size, is_array);
5574
5575 // Do not let reads from the cloned object float above the arraycopy.
5576 if (alloc != nullptr) {
5577 // Do not let stores that initialize this object be reordered with
5578 // a subsequent store that would make this object accessible by
5579 // other threads.
5580 // Record what AllocateNode this StoreStore protects so that
5581 // escape analysis can go from the MemBarStoreStoreNode to the
5582 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5583 // based on the escape status of the AllocateNode.
5584 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5585 } else {
5586 insert_mem_bar(Op_MemBarCPUOrder);
5587 }
5588 }
5589
5590 //------------------------inline_native_clone----------------------------
5591 // protected native Object java.lang.Object.clone();
5592 //
5593 // Here are the simple edge cases:
5594 // null receiver => normal trap
5595 // virtual and clone was overridden => slow path to out-of-line clone
5596 // not cloneable or finalizer => slow path to out-of-line Object.clone
5597 //
5598 // The general case has two steps, allocation and copying.
5599 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5600 //
5601 // Copying also has two cases, oop arrays and everything else.
5602 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5603 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5604 //
5605 // These steps fold up nicely if and when the cloned object's klass
5606 // can be sharply typed as an object array, a type array, or an instance.
5607 //
5608 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5609 PhiNode* result_val;
5610
5611 // Set the reexecute bit for the interpreter to reexecute
5612 // the bytecode that invokes Object.clone if deoptimization happens.
5613 { PreserveReexecuteState preexecs(this);
5614 jvms()->set_should_reexecute(true);
5615
5616 Node* obj = argument(0);
5617 obj = null_check_receiver();
5618 if (stopped()) return true;
5619
5620 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5621 if (obj_type->is_inlinetypeptr()) {
5622 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5623 // no identity.
5624 set_result(obj);
5625 return true;
5626 }
5627
5628 // If we are going to clone an instance, we need its exact type to
5629 // know the number and types of fields to convert the clone to
5630 // loads/stores. Maybe a speculative type can help us.
5631 if (!obj_type->klass_is_exact() &&
5632 obj_type->speculative_type() != nullptr &&
5633 obj_type->speculative_type()->is_instance_klass() &&
5634 !obj_type->speculative_type()->is_inlinetype()) {
5635 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5636 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5637 !spec_ik->has_injected_fields()) {
5638 if (!obj_type->isa_instptr() ||
5639 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5640 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5641 }
5642 }
5643 }
5644
5645 // Conservatively insert a memory barrier on all memory slices.
5646 // Do not let writes into the original float below the clone.
5647 insert_mem_bar(Op_MemBarCPUOrder);
5648
5649 // paths into result_reg:
5650 enum {
5651 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5652 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5653 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5654 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5655 PATH_LIMIT
5656 };
5657 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5658 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5659 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5660 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5661 record_for_igvn(result_reg);
5662
5663 // TODO 8350865 For arrays, this might be folded and then not account for atomic arrays
5664 Node* obj_klass = load_object_klass(obj);
5665 // We only go to the fast case code if we pass a number of guards.
5666 // The paths which do not pass are accumulated in the slow_region.
5667 RegionNode* slow_region = new RegionNode(1);
5668 record_for_igvn(slow_region);
5669
5670 Node* array_obj = obj;
5671 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5672 if (array_ctl != nullptr) {
5673 // It's an array.
5674 PreserveJVMState pjvms(this);
5675 set_control(array_ctl);
5676
5677 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5678 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5679 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5680 obj_type->can_be_inline_array() &&
5681 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5682 // Flat inline type array may have object field that would require a
5683 // write barrier. Conservatively, go to slow path.
5684 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5685 }
5686
5687 if (!stopped()) {
5688 Node* obj_length = load_array_length(array_obj);
5689 Node* array_size = nullptr; // Size of the array without object alignment padding.
5690 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5691
5692 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5693 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5694 // If it is an oop array, it requires very special treatment,
5695 // because gc barriers are required when accessing the array.
5696 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5697 if (is_obja != nullptr) {
5698 PreserveJVMState pjvms2(this);
5699 set_control(is_obja);
5700 // Generate a direct call to the right arraycopy function(s).
5701 // Clones are always tightly coupled.
5702 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5703 ac->set_clone_oop_array();
5704 Node* n = _gvn.transform(ac);
5705 assert(n == ac, "cannot disappear");
5706 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5707
5708 result_reg->init_req(_objArray_path, control());
5709 result_val->init_req(_objArray_path, alloc_obj);
5710 result_i_o ->set_req(_objArray_path, i_o());
5711 result_mem ->set_req(_objArray_path, reset_memory());
5712 }
5713 }
5714 // Otherwise, there are no barriers to worry about.
5715 // (We can dispense with card marks if we know the allocation
5716 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5717 // causes the non-eden paths to take compensating steps to
5718 // simulate a fresh allocation, so that no further
5719 // card marks are required in compiled code to initialize
5720 // the object.)
5721
5722 if (!stopped()) {
5723 copy_to_clone(obj, alloc_obj, array_size, true);
5724
5725 // Present the results of the copy.
5726 result_reg->init_req(_array_path, control());
5727 result_val->init_req(_array_path, alloc_obj);
5728 result_i_o ->set_req(_array_path, i_o());
5729 result_mem ->set_req(_array_path, reset_memory());
5730 }
5731 }
5732 }
5733
5734 if (!stopped()) {
5735 // It's an instance (we did array above). Make the slow-path tests.
5736 // If this is a virtual call, we generate a funny guard. We grab
5737 // the vtable entry corresponding to clone() from the target object.
5738 // If the target method which we are calling happens to be the
5739 // Object clone() method, we pass the guard. We do not need this
5740 // guard for non-virtual calls; the caller is known to be the native
5741 // Object clone().
5742 if (is_virtual) {
5743 generate_virtual_guard(obj_klass, slow_region);
5744 }
5745
5746 // The object must be easily cloneable and must not have a finalizer.
5747 // Both of these conditions may be checked in a single test.
5748 // We could optimize the test further, but we don't care.
5749 generate_misc_flags_guard(obj_klass,
5750 // Test both conditions:
5751 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5752 // Must be cloneable but not finalizer:
5753 KlassFlags::_misc_is_cloneable_fast,
5754 slow_region);
5755 }
5756
5757 if (!stopped()) {
5758 // It's an instance, and it passed the slow-path tests.
5759 PreserveJVMState pjvms(this);
5760 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5761 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5762 // is reexecuted if deoptimization occurs and there could be problems when merging
5763 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5764 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5765
5766 copy_to_clone(obj, alloc_obj, obj_size, false);
5767
5768 // Present the results of the slow call.
5769 result_reg->init_req(_instance_path, control());
5770 result_val->init_req(_instance_path, alloc_obj);
5771 result_i_o ->set_req(_instance_path, i_o());
5772 result_mem ->set_req(_instance_path, reset_memory());
5773 }
5774
5775 // Generate code for the slow case. We make a call to clone().
5776 set_control(_gvn.transform(slow_region));
5777 if (!stopped()) {
5778 PreserveJVMState pjvms(this);
5779 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5780 // We need to deoptimize on exception (see comment above)
5781 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5782 // this->control() comes from set_results_for_java_call
5783 result_reg->init_req(_slow_path, control());
5784 result_val->init_req(_slow_path, slow_result);
5785 result_i_o ->set_req(_slow_path, i_o());
5786 result_mem ->set_req(_slow_path, reset_memory());
5787 }
5788
5789 // Return the combined state.
5790 set_control( _gvn.transform(result_reg));
5791 set_i_o( _gvn.transform(result_i_o));
5792 set_all_memory( _gvn.transform(result_mem));
5793 } // original reexecute is set back here
5794
5795 set_result(_gvn.transform(result_val));
5796 return true;
5797 }
5798
5799 // If we have a tightly coupled allocation, the arraycopy may take care
5800 // of the array initialization. If one of the guards we insert between
5801 // the allocation and the arraycopy causes a deoptimization, an
5802 // uninitialized array will escape the compiled method. To prevent that
5803 // we set the JVM state for uncommon traps between the allocation and
5804 // the arraycopy to the state before the allocation so, in case of
5805 // deoptimization, we'll reexecute the allocation and the
5806 // initialization.
5807 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5808 if (alloc != nullptr) {
5809 ciMethod* trap_method = alloc->jvms()->method();
5810 int trap_bci = alloc->jvms()->bci();
5811
5812 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5813 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5814 // Make sure there's no store between the allocation and the
5815 // arraycopy otherwise visible side effects could be rexecuted
5816 // in case of deoptimization and cause incorrect execution.
5817 bool no_interfering_store = true;
5818 Node* mem = alloc->in(TypeFunc::Memory);
5819 if (mem->is_MergeMem()) {
5820 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5821 Node* n = mms.memory();
5822 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5823 assert(n->is_Store(), "what else?");
5824 no_interfering_store = false;
5825 break;
5826 }
5827 }
5828 } else {
5829 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5830 Node* n = mms.memory();
5831 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5832 assert(n->is_Store(), "what else?");
5833 no_interfering_store = false;
5834 break;
5835 }
5836 }
5837 }
5838
5839 if (no_interfering_store) {
5840 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5841
5842 JVMState* saved_jvms = jvms();
5843 saved_reexecute_sp = _reexecute_sp;
5844
5845 set_jvms(sfpt->jvms());
5846 _reexecute_sp = jvms()->sp();
5847
5848 return saved_jvms;
5849 }
5850 }
5851 }
5852 return nullptr;
5853 }
5854
5855 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5856 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5857 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5858 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5859 uint size = alloc->req();
5860 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5861 old_jvms->set_map(sfpt);
5862 for (uint i = 0; i < size; i++) {
5863 sfpt->init_req(i, alloc->in(i));
5864 }
5865 int adjustment = 1;
5866 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5867 if (ary_klass_ptr->is_null_free()) {
5868 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
5869 // also requires the componentType and initVal on stack for re-execution.
5870 // Re-create and push the componentType.
5871 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5872 ciInstance* instance = klass->component_mirror_instance();
5873 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5874 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5875 adjustment++;
5876 }
5877 // re-push array length for deoptimization
5878 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5879 if (ary_klass_ptr->is_null_free()) {
5880 // Re-create and push the initVal.
5881 Node* init_val = alloc->in(AllocateNode::InitValue);
5882 if (init_val == nullptr) {
5883 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
5884 } else if (UseCompressedOops) {
5885 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
5886 }
5887 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
5888 adjustment++;
5889 }
5890 old_jvms->set_sp(old_jvms->sp() + adjustment);
5891 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5892 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5893 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5894 old_jvms->set_should_reexecute(true);
5895
5896 sfpt->set_i_o(map()->i_o());
5897 sfpt->set_memory(map()->memory());
5898 sfpt->set_control(map()->control());
5899 return sfpt;
5900 }
5901
5902 // In case of a deoptimization, we restart execution at the
5903 // allocation, allocating a new array. We would leave an uninitialized
5904 // array in the heap that GCs wouldn't expect. Move the allocation
5905 // after the traps so we don't allocate the array if we
5906 // deoptimize. This is possible because tightly_coupled_allocation()
5907 // guarantees there's no observer of the allocated array at this point
5908 // and the control flow is simple enough.
5909 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5910 int saved_reexecute_sp, uint new_idx) {
5911 if (saved_jvms_before_guards != nullptr && !stopped()) {
5912 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5913
5914 assert(alloc != nullptr, "only with a tightly coupled allocation");
5915 // restore JVM state to the state at the arraycopy
5916 saved_jvms_before_guards->map()->set_control(map()->control());
5917 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5918 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5919 // If we've improved the types of some nodes (null check) while
5920 // emitting the guards, propagate them to the current state
5921 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5922 set_jvms(saved_jvms_before_guards);
5923 _reexecute_sp = saved_reexecute_sp;
5924
5925 // Remove the allocation from above the guards
5926 CallProjections* callprojs = alloc->extract_projections(true);
5927 InitializeNode* init = alloc->initialization();
5928 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5929 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5930 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5931
5932 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5933 // the allocation (i.e. is only valid if the allocation succeeds):
5934 // 1) replace CastIINode with AllocateArrayNode's length here
5935 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5936 //
5937 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5938 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5939 Node* init_control = init->proj_out(TypeFunc::Control);
5940 Node* alloc_length = alloc->Ideal_length();
5941 #ifdef ASSERT
5942 Node* prev_cast = nullptr;
5943 #endif
5944 for (uint i = 0; i < init_control->outcnt(); i++) {
5945 Node* init_out = init_control->raw_out(i);
5946 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5947 #ifdef ASSERT
5948 if (prev_cast == nullptr) {
5949 prev_cast = init_out;
5950 } else {
5951 if (prev_cast->cmp(*init_out) == false) {
5952 prev_cast->dump();
5953 init_out->dump();
5954 assert(false, "not equal CastIINode");
5955 }
5956 }
5957 #endif
5958 C->gvn_replace_by(init_out, alloc_length);
5959 }
5960 }
5961 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5962
5963 // move the allocation here (after the guards)
5964 _gvn.hash_delete(alloc);
5965 alloc->set_req(TypeFunc::Control, control());
5966 alloc->set_req(TypeFunc::I_O, i_o());
5967 Node *mem = reset_memory();
5968 set_all_memory(mem);
5969 alloc->set_req(TypeFunc::Memory, mem);
5970 set_control(init->proj_out_or_null(TypeFunc::Control));
5971 set_i_o(callprojs->fallthrough_ioproj);
5972
5973 // Update memory as done in GraphKit::set_output_for_allocation()
5974 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5975 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5976 if (ary_type->isa_aryptr() && length_type != nullptr) {
5977 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5978 }
5979 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5980 int elemidx = C->get_alias_index(telemref);
5981 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5982 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5983
5984 Node* allocx = _gvn.transform(alloc);
5985 assert(allocx == alloc, "where has the allocation gone?");
5986 assert(dest->is_CheckCastPP(), "not an allocation result?");
5987
5988 _gvn.hash_delete(dest);
5989 dest->set_req(0, control());
5990 Node* destx = _gvn.transform(dest);
5991 assert(destx == dest, "where has the allocation result gone?");
5992
5993 array_ideal_length(alloc, ary_type, true);
5994 }
5995 }
5996
5997 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5998 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5999 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6000 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6001 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6002 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6003 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6004 JVMState* saved_jvms_before_guards) {
6005 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6006 // There is at least one unrelated uncommon trap which needs to be replaced.
6007 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6008
6009 JVMState* saved_jvms = jvms();
6010 const int saved_reexecute_sp = _reexecute_sp;
6011 set_jvms(sfpt->jvms());
6012 _reexecute_sp = jvms()->sp();
6013
6014 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6015
6016 // Restore state
6017 set_jvms(saved_jvms);
6018 _reexecute_sp = saved_reexecute_sp;
6019 }
6020 }
6021
6022 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6023 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6024 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6025 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6026 while (if_proj->is_IfProj()) {
6027 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6028 if (uncommon_trap != nullptr) {
6029 create_new_uncommon_trap(uncommon_trap);
6030 }
6031 assert(if_proj->in(0)->is_If(), "must be If");
6032 if_proj = if_proj->in(0)->in(0);
6033 }
6034 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6035 "must have reached control projection of init node");
6036 }
6037
6038 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6039 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6040 assert(trap_request != 0, "no valid UCT trap request");
6041 PreserveJVMState pjvms(this);
6042 set_control(uncommon_trap_call->in(0));
6043 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6044 Deoptimization::trap_request_action(trap_request));
6045 assert(stopped(), "Should be stopped");
6046 _gvn.hash_delete(uncommon_trap_call);
6047 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6048 }
6049
6050 // Common checks for array sorting intrinsics arguments.
6051 // Returns `true` if checks passed.
6052 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6053 // check address of the class
6054 if (elementType == nullptr || elementType->is_top()) {
6055 return false; // dead path
6056 }
6057 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6058 if (elem_klass == nullptr) {
6059 return false; // dead path
6060 }
6061 // java_mirror_type() returns non-null for compile-time Class constants only
6062 ciType* elem_type = elem_klass->java_mirror_type();
6063 if (elem_type == nullptr) {
6064 return false;
6065 }
6066 bt = elem_type->basic_type();
6067 // Disable the intrinsic if the CPU does not support SIMD sort
6068 if (!Matcher::supports_simd_sort(bt)) {
6069 return false;
6070 }
6071 // check address of the array
6072 if (obj == nullptr || obj->is_top()) {
6073 return false; // dead path
6074 }
6075 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6076 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6077 return false; // failed input validation
6078 }
6079 return true;
6080 }
6081
6082 //------------------------------inline_array_partition-----------------------
6083 bool LibraryCallKit::inline_array_partition() {
6084 address stubAddr = StubRoutines::select_array_partition_function();
6085 if (stubAddr == nullptr) {
6086 return false; // Intrinsic's stub is not implemented on this platform
6087 }
6088 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6089
6090 // no receiver because it is a static method
6091 Node* elementType = argument(0);
6092 Node* obj = argument(1);
6093 Node* offset = argument(2); // long
6094 Node* fromIndex = argument(4);
6095 Node* toIndex = argument(5);
6096 Node* indexPivot1 = argument(6);
6097 Node* indexPivot2 = argument(7);
6098 // PartitionOperation: argument(8) is ignored
6099
6100 Node* pivotIndices = nullptr;
6101 BasicType bt = T_ILLEGAL;
6102
6103 if (!check_array_sort_arguments(elementType, obj, bt)) {
6104 return false;
6105 }
6106 null_check(obj);
6107 // If obj is dead, only null-path is taken.
6108 if (stopped()) {
6109 return true;
6110 }
6111 // Set the original stack and the reexecute bit for the interpreter to reexecute
6112 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6113 { PreserveReexecuteState preexecs(this);
6114 jvms()->set_should_reexecute(true);
6115
6116 Node* obj_adr = make_unsafe_address(obj, offset);
6117
6118 // create the pivotIndices array of type int and size = 2
6119 Node* size = intcon(2);
6120 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6121 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6122 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6123 guarantee(alloc != nullptr, "created above");
6124 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6125
6126 // pass the basic type enum to the stub
6127 Node* elemType = intcon(bt);
6128
6129 // Call the stub
6130 const char *stubName = "array_partition_stub";
6131 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6132 stubAddr, stubName, TypePtr::BOTTOM,
6133 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6134 indexPivot1, indexPivot2);
6135
6136 } // original reexecute is set back here
6137
6138 if (!stopped()) {
6139 set_result(pivotIndices);
6140 }
6141
6142 return true;
6143 }
6144
6145
6146 //------------------------------inline_array_sort-----------------------
6147 bool LibraryCallKit::inline_array_sort() {
6148 address stubAddr = StubRoutines::select_arraysort_function();
6149 if (stubAddr == nullptr) {
6150 return false; // Intrinsic's stub is not implemented on this platform
6151 }
6152 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6153
6154 // no receiver because it is a static method
6155 Node* elementType = argument(0);
6156 Node* obj = argument(1);
6157 Node* offset = argument(2); // long
6158 Node* fromIndex = argument(4);
6159 Node* toIndex = argument(5);
6160 // SortOperation: argument(6) is ignored
6161
6162 BasicType bt = T_ILLEGAL;
6163
6164 if (!check_array_sort_arguments(elementType, obj, bt)) {
6165 return false;
6166 }
6167 null_check(obj);
6168 // If obj is dead, only null-path is taken.
6169 if (stopped()) {
6170 return true;
6171 }
6172 Node* obj_adr = make_unsafe_address(obj, offset);
6173
6174 // pass the basic type enum to the stub
6175 Node* elemType = intcon(bt);
6176
6177 // Call the stub.
6178 const char *stubName = "arraysort_stub";
6179 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6180 stubAddr, stubName, TypePtr::BOTTOM,
6181 obj_adr, elemType, fromIndex, toIndex);
6182
6183 return true;
6184 }
6185
6186
6187 //------------------------------inline_arraycopy-----------------------
6188 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6189 // Object dest, int destPos,
6190 // int length);
6191 bool LibraryCallKit::inline_arraycopy() {
6192 // Get the arguments.
6193 Node* src = argument(0); // type: oop
6194 Node* src_offset = argument(1); // type: int
6195 Node* dest = argument(2); // type: oop
6196 Node* dest_offset = argument(3); // type: int
6197 Node* length = argument(4); // type: int
6198
6199 uint new_idx = C->unique();
6200
6201 // Check for allocation before we add nodes that would confuse
6202 // tightly_coupled_allocation()
6203 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6204
6205 int saved_reexecute_sp = -1;
6206 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6207 // See arraycopy_restore_alloc_state() comment
6208 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6209 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6210 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6211 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6212
6213 // The following tests must be performed
6214 // (1) src and dest are arrays.
6215 // (2) src and dest arrays must have elements of the same BasicType
6216 // (3) src and dest must not be null.
6217 // (4) src_offset must not be negative.
6218 // (5) dest_offset must not be negative.
6219 // (6) length must not be negative.
6220 // (7) src_offset + length must not exceed length of src.
6221 // (8) dest_offset + length must not exceed length of dest.
6222 // (9) each element of an oop array must be assignable
6223
6224 // (3) src and dest must not be null.
6225 // always do this here because we need the JVM state for uncommon traps
6226 Node* null_ctl = top();
6227 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6228 assert(null_ctl->is_top(), "no null control here");
6229 dest = null_check(dest, T_ARRAY);
6230
6231 if (!can_emit_guards) {
6232 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6233 // guards but the arraycopy node could still take advantage of a
6234 // tightly allocated allocation. tightly_coupled_allocation() is
6235 // called again to make sure it takes the null check above into
6236 // account: the null check is mandatory and if it caused an
6237 // uncommon trap to be emitted then the allocation can't be
6238 // considered tightly coupled in this context.
6239 alloc = tightly_coupled_allocation(dest);
6240 }
6241
6242 bool validated = false;
6243
6244 const Type* src_type = _gvn.type(src);
6245 const Type* dest_type = _gvn.type(dest);
6246 const TypeAryPtr* top_src = src_type->isa_aryptr();
6247 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6248
6249 // Do we have the type of src?
6250 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6251 // Do we have the type of dest?
6252 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6253 // Is the type for src from speculation?
6254 bool src_spec = false;
6255 // Is the type for dest from speculation?
6256 bool dest_spec = false;
6257
6258 if ((!has_src || !has_dest) && can_emit_guards) {
6259 // We don't have sufficient type information, let's see if
6260 // speculative types can help. We need to have types for both src
6261 // and dest so that it pays off.
6262
6263 // Do we already have or could we have type information for src
6264 bool could_have_src = has_src;
6265 // Do we already have or could we have type information for dest
6266 bool could_have_dest = has_dest;
6267
6268 ciKlass* src_k = nullptr;
6269 if (!has_src) {
6270 src_k = src_type->speculative_type_not_null();
6271 if (src_k != nullptr && src_k->is_array_klass()) {
6272 could_have_src = true;
6273 }
6274 }
6275
6276 ciKlass* dest_k = nullptr;
6277 if (!has_dest) {
6278 dest_k = dest_type->speculative_type_not_null();
6279 if (dest_k != nullptr && dest_k->is_array_klass()) {
6280 could_have_dest = true;
6281 }
6282 }
6283
6284 if (could_have_src && could_have_dest) {
6285 // This is going to pay off so emit the required guards
6286 if (!has_src) {
6287 src = maybe_cast_profiled_obj(src, src_k, true);
6288 src_type = _gvn.type(src);
6289 top_src = src_type->isa_aryptr();
6290 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6291 src_spec = true;
6292 }
6293 if (!has_dest) {
6294 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6295 dest_type = _gvn.type(dest);
6296 top_dest = dest_type->isa_aryptr();
6297 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6298 dest_spec = true;
6299 }
6300 }
6301 }
6302
6303 if (has_src && has_dest && can_emit_guards) {
6304 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6305 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6306 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6307 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6308
6309 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6310 // If both arrays are object arrays then having the exact types
6311 // for both will remove the need for a subtype check at runtime
6312 // before the call and may make it possible to pick a faster copy
6313 // routine (without a subtype check on every element)
6314 // Do we have the exact type of src?
6315 bool could_have_src = src_spec;
6316 // Do we have the exact type of dest?
6317 bool could_have_dest = dest_spec;
6318 ciKlass* src_k = nullptr;
6319 ciKlass* dest_k = nullptr;
6320 if (!src_spec) {
6321 src_k = src_type->speculative_type_not_null();
6322 if (src_k != nullptr && src_k->is_array_klass()) {
6323 could_have_src = true;
6324 }
6325 }
6326 if (!dest_spec) {
6327 dest_k = dest_type->speculative_type_not_null();
6328 if (dest_k != nullptr && dest_k->is_array_klass()) {
6329 could_have_dest = true;
6330 }
6331 }
6332 if (could_have_src && could_have_dest) {
6333 // If we can have both exact types, emit the missing guards
6334 if (could_have_src && !src_spec) {
6335 src = maybe_cast_profiled_obj(src, src_k, true);
6336 src_type = _gvn.type(src);
6337 top_src = src_type->isa_aryptr();
6338 }
6339 if (could_have_dest && !dest_spec) {
6340 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6341 dest_type = _gvn.type(dest);
6342 top_dest = dest_type->isa_aryptr();
6343 }
6344 }
6345 }
6346 }
6347
6348 ciMethod* trap_method = method();
6349 int trap_bci = bci();
6350 if (saved_jvms_before_guards != nullptr) {
6351 trap_method = alloc->jvms()->method();
6352 trap_bci = alloc->jvms()->bci();
6353 }
6354
6355 bool negative_length_guard_generated = false;
6356
6357 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6358 can_emit_guards && !src->is_top() && !dest->is_top()) {
6359 // validate arguments: enables transformation the ArrayCopyNode
6360 validated = true;
6361
6362 RegionNode* slow_region = new RegionNode(1);
6363 record_for_igvn(slow_region);
6364
6365 // (1) src and dest are arrays.
6366 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6367 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6368
6369 // (2) src and dest arrays must have elements of the same BasicType
6370 // done at macro expansion or at Ideal transformation time
6371
6372 // (4) src_offset must not be negative.
6373 generate_negative_guard(src_offset, slow_region);
6374
6375 // (5) dest_offset must not be negative.
6376 generate_negative_guard(dest_offset, slow_region);
6377
6378 // (7) src_offset + length must not exceed length of src.
6379 generate_limit_guard(src_offset, length,
6380 load_array_length(src),
6381 slow_region);
6382
6383 // (8) dest_offset + length must not exceed length of dest.
6384 generate_limit_guard(dest_offset, length,
6385 load_array_length(dest),
6386 slow_region);
6387
6388 // (6) length must not be negative.
6389 // This is also checked in generate_arraycopy() during macro expansion, but
6390 // we also have to check it here for the case where the ArrayCopyNode will
6391 // be eliminated by Escape Analysis.
6392 if (EliminateAllocations) {
6393 generate_negative_guard(length, slow_region);
6394 negative_length_guard_generated = true;
6395 }
6396
6397 // (9) each element of an oop array must be assignable
6398 Node* dest_klass = load_object_klass(dest);
6399 if (src != dest) {
6400 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6401 slow_region->add_req(not_subtype_ctrl);
6402 }
6403
6404 // TODO 8350865 Fix below logic. Also handle atomicity.
6405 generate_fair_guard(flat_array_test(src), slow_region);
6406 generate_fair_guard(flat_array_test(dest), slow_region);
6407
6408 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6409 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6410 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6411 src_type = _gvn.type(src);
6412 top_src = src_type->isa_aryptr();
6413
6414 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6415 if (!stopped() && UseArrayFlattening) {
6416 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6417 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6418 if (top_src != nullptr && top_src->is_flat()) {
6419 // Src is flat, check that dest is flat as well
6420 if (top_dest != nullptr && !top_dest->is_flat()) {
6421 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6422 // Since dest is flat and src <: dest, dest must have the same type as src.
6423 top_dest = top_src->cast_to_exactness(false);
6424 assert(top_dest->is_flat(), "dest must be flat");
6425 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6426 }
6427 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6428 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6429 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6430 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6431 generate_fair_guard(flat_array_test(src), slow_region);
6432 if (top_src != nullptr) {
6433 top_src = top_src->cast_to_not_flat();
6434 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6435 }
6436 }
6437 }
6438
6439 {
6440 PreserveJVMState pjvms(this);
6441 set_control(_gvn.transform(slow_region));
6442 uncommon_trap(Deoptimization::Reason_intrinsic,
6443 Deoptimization::Action_make_not_entrant);
6444 assert(stopped(), "Should be stopped");
6445 }
6446 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6447 }
6448
6449 if (stopped()) {
6450 return true;
6451 }
6452
6453 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6454 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6455 // so the compiler has a chance to eliminate them: during macro expansion,
6456 // we have to set their control (CastPP nodes are eliminated).
6457 load_object_klass(src), load_object_klass(dest),
6458 load_array_length(src), load_array_length(dest));
6459
6460 ac->set_arraycopy(validated);
6461
6462 Node* n = _gvn.transform(ac);
6463 if (n == ac) {
6464 ac->connect_outputs(this);
6465 } else {
6466 assert(validated, "shouldn't transform if all arguments not validated");
6467 set_all_memory(n);
6468 }
6469 clear_upper_avx();
6470
6471
6472 return true;
6473 }
6474
6475
6476 // Helper function which determines if an arraycopy immediately follows
6477 // an allocation, with no intervening tests or other escapes for the object.
6478 AllocateArrayNode*
6479 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6480 if (stopped()) return nullptr; // no fast path
6481 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6482
6483 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6484 if (alloc == nullptr) return nullptr;
6485
6486 Node* rawmem = memory(Compile::AliasIdxRaw);
6487 // Is the allocation's memory state untouched?
6488 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6489 // Bail out if there have been raw-memory effects since the allocation.
6490 // (Example: There might have been a call or safepoint.)
6491 return nullptr;
6492 }
6493 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6494 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6495 return nullptr;
6496 }
6497
6498 // There must be no unexpected observers of this allocation.
6499 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6500 Node* obs = ptr->fast_out(i);
6501 if (obs != this->map()) {
6502 return nullptr;
6503 }
6504 }
6505
6506 // This arraycopy must unconditionally follow the allocation of the ptr.
6507 Node* alloc_ctl = ptr->in(0);
6508 Node* ctl = control();
6509 while (ctl != alloc_ctl) {
6510 // There may be guards which feed into the slow_region.
6511 // Any other control flow means that we might not get a chance
6512 // to finish initializing the allocated object.
6513 // Various low-level checks bottom out in uncommon traps. These
6514 // are considered safe since we've already checked above that
6515 // there is no unexpected observer of this allocation.
6516 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6517 assert(ctl->in(0)->is_If(), "must be If");
6518 ctl = ctl->in(0)->in(0);
6519 } else {
6520 return nullptr;
6521 }
6522 }
6523
6524 // If we get this far, we have an allocation which immediately
6525 // precedes the arraycopy, and we can take over zeroing the new object.
6526 // The arraycopy will finish the initialization, and provide
6527 // a new control state to which we will anchor the destination pointer.
6528
6529 return alloc;
6530 }
6531
6532 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6533 if (node->is_IfProj()) {
6534 Node* other_proj = node->as_IfProj()->other_if_proj();
6535 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6536 Node* obs = other_proj->fast_out(j);
6537 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6538 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6539 return obs->as_CallStaticJava();
6540 }
6541 }
6542 }
6543 return nullptr;
6544 }
6545
6546 //-------------inline_encodeISOArray-----------------------------------
6547 // encode char[] to byte[] in ISO_8859_1 or ASCII
6548 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6549 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6550 // no receiver since it is static method
6551 Node *src = argument(0);
6552 Node *src_offset = argument(1);
6553 Node *dst = argument(2);
6554 Node *dst_offset = argument(3);
6555 Node *length = argument(4);
6556
6557 src = must_be_not_null(src, true);
6558 dst = must_be_not_null(dst, true);
6559
6560 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6561 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6562 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6563 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6564 // failed array check
6565 return false;
6566 }
6567
6568 // Figure out the size and type of the elements we will be copying.
6569 BasicType src_elem = src_type->elem()->array_element_basic_type();
6570 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6571 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6572 return false;
6573 }
6574
6575 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6576 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6577 // 'src_start' points to src array + scaled offset
6578 // 'dst_start' points to dst array + scaled offset
6579
6580 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6581 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6582 enc = _gvn.transform(enc);
6583 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6584 set_memory(res_mem, mtype);
6585 set_result(enc);
6586 clear_upper_avx();
6587
6588 return true;
6589 }
6590
6591 //-------------inline_multiplyToLen-----------------------------------
6592 bool LibraryCallKit::inline_multiplyToLen() {
6593 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6594
6595 address stubAddr = StubRoutines::multiplyToLen();
6596 if (stubAddr == nullptr) {
6597 return false; // Intrinsic's stub is not implemented on this platform
6598 }
6599 const char* stubName = "multiplyToLen";
6600
6601 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6602
6603 // no receiver because it is a static method
6604 Node* x = argument(0);
6605 Node* xlen = argument(1);
6606 Node* y = argument(2);
6607 Node* ylen = argument(3);
6608 Node* z = argument(4);
6609
6610 x = must_be_not_null(x, true);
6611 y = must_be_not_null(y, true);
6612
6613 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6614 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6615 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6616 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6617 // failed array check
6618 return false;
6619 }
6620
6621 BasicType x_elem = x_type->elem()->array_element_basic_type();
6622 BasicType y_elem = y_type->elem()->array_element_basic_type();
6623 if (x_elem != T_INT || y_elem != T_INT) {
6624 return false;
6625 }
6626
6627 Node* x_start = array_element_address(x, intcon(0), x_elem);
6628 Node* y_start = array_element_address(y, intcon(0), y_elem);
6629 // 'x_start' points to x array + scaled xlen
6630 // 'y_start' points to y array + scaled ylen
6631
6632 Node* z_start = array_element_address(z, intcon(0), T_INT);
6633
6634 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6635 OptoRuntime::multiplyToLen_Type(),
6636 stubAddr, stubName, TypePtr::BOTTOM,
6637 x_start, xlen, y_start, ylen, z_start);
6638
6639 C->set_has_split_ifs(true); // Has chance for split-if optimization
6640 set_result(z);
6641 return true;
6642 }
6643
6644 //-------------inline_squareToLen------------------------------------
6645 bool LibraryCallKit::inline_squareToLen() {
6646 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6647
6648 address stubAddr = StubRoutines::squareToLen();
6649 if (stubAddr == nullptr) {
6650 return false; // Intrinsic's stub is not implemented on this platform
6651 }
6652 const char* stubName = "squareToLen";
6653
6654 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6655
6656 Node* x = argument(0);
6657 Node* len = argument(1);
6658 Node* z = argument(2);
6659 Node* zlen = argument(3);
6660
6661 x = must_be_not_null(x, true);
6662 z = must_be_not_null(z, true);
6663
6664 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6665 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6666 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6667 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6668 // failed array check
6669 return false;
6670 }
6671
6672 BasicType x_elem = x_type->elem()->array_element_basic_type();
6673 BasicType z_elem = z_type->elem()->array_element_basic_type();
6674 if (x_elem != T_INT || z_elem != T_INT) {
6675 return false;
6676 }
6677
6678
6679 Node* x_start = array_element_address(x, intcon(0), x_elem);
6680 Node* z_start = array_element_address(z, intcon(0), z_elem);
6681
6682 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6683 OptoRuntime::squareToLen_Type(),
6684 stubAddr, stubName, TypePtr::BOTTOM,
6685 x_start, len, z_start, zlen);
6686
6687 set_result(z);
6688 return true;
6689 }
6690
6691 //-------------inline_mulAdd------------------------------------------
6692 bool LibraryCallKit::inline_mulAdd() {
6693 assert(UseMulAddIntrinsic, "not implemented on this platform");
6694
6695 address stubAddr = StubRoutines::mulAdd();
6696 if (stubAddr == nullptr) {
6697 return false; // Intrinsic's stub is not implemented on this platform
6698 }
6699 const char* stubName = "mulAdd";
6700
6701 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6702
6703 Node* out = argument(0);
6704 Node* in = argument(1);
6705 Node* offset = argument(2);
6706 Node* len = argument(3);
6707 Node* k = argument(4);
6708
6709 in = must_be_not_null(in, true);
6710 out = must_be_not_null(out, true);
6711
6712 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6713 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6714 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6715 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6716 // failed array check
6717 return false;
6718 }
6719
6720 BasicType out_elem = out_type->elem()->array_element_basic_type();
6721 BasicType in_elem = in_type->elem()->array_element_basic_type();
6722 if (out_elem != T_INT || in_elem != T_INT) {
6723 return false;
6724 }
6725
6726 Node* outlen = load_array_length(out);
6727 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6728 Node* out_start = array_element_address(out, intcon(0), out_elem);
6729 Node* in_start = array_element_address(in, intcon(0), in_elem);
6730
6731 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6732 OptoRuntime::mulAdd_Type(),
6733 stubAddr, stubName, TypePtr::BOTTOM,
6734 out_start,in_start, new_offset, len, k);
6735 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6736 set_result(result);
6737 return true;
6738 }
6739
6740 //-------------inline_montgomeryMultiply-----------------------------------
6741 bool LibraryCallKit::inline_montgomeryMultiply() {
6742 address stubAddr = StubRoutines::montgomeryMultiply();
6743 if (stubAddr == nullptr) {
6744 return false; // Intrinsic's stub is not implemented on this platform
6745 }
6746
6747 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6748 const char* stubName = "montgomery_multiply";
6749
6750 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6751
6752 Node* a = argument(0);
6753 Node* b = argument(1);
6754 Node* n = argument(2);
6755 Node* len = argument(3);
6756 Node* inv = argument(4);
6757 Node* m = argument(6);
6758
6759 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6760 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6761 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6762 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6763 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6764 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6765 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6766 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6767 // failed array check
6768 return false;
6769 }
6770
6771 BasicType a_elem = a_type->elem()->array_element_basic_type();
6772 BasicType b_elem = b_type->elem()->array_element_basic_type();
6773 BasicType n_elem = n_type->elem()->array_element_basic_type();
6774 BasicType m_elem = m_type->elem()->array_element_basic_type();
6775 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6776 return false;
6777 }
6778
6779 // Make the call
6780 {
6781 Node* a_start = array_element_address(a, intcon(0), a_elem);
6782 Node* b_start = array_element_address(b, intcon(0), b_elem);
6783 Node* n_start = array_element_address(n, intcon(0), n_elem);
6784 Node* m_start = array_element_address(m, intcon(0), m_elem);
6785
6786 Node* call = make_runtime_call(RC_LEAF,
6787 OptoRuntime::montgomeryMultiply_Type(),
6788 stubAddr, stubName, TypePtr::BOTTOM,
6789 a_start, b_start, n_start, len, inv, top(),
6790 m_start);
6791 set_result(m);
6792 }
6793
6794 return true;
6795 }
6796
6797 bool LibraryCallKit::inline_montgomerySquare() {
6798 address stubAddr = StubRoutines::montgomerySquare();
6799 if (stubAddr == nullptr) {
6800 return false; // Intrinsic's stub is not implemented on this platform
6801 }
6802
6803 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6804 const char* stubName = "montgomery_square";
6805
6806 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6807
6808 Node* a = argument(0);
6809 Node* n = argument(1);
6810 Node* len = argument(2);
6811 Node* inv = argument(3);
6812 Node* m = argument(5);
6813
6814 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6815 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6816 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6817 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6818 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6819 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6820 // failed array check
6821 return false;
6822 }
6823
6824 BasicType a_elem = a_type->elem()->array_element_basic_type();
6825 BasicType n_elem = n_type->elem()->array_element_basic_type();
6826 BasicType m_elem = m_type->elem()->array_element_basic_type();
6827 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6828 return false;
6829 }
6830
6831 // Make the call
6832 {
6833 Node* a_start = array_element_address(a, intcon(0), a_elem);
6834 Node* n_start = array_element_address(n, intcon(0), n_elem);
6835 Node* m_start = array_element_address(m, intcon(0), m_elem);
6836
6837 Node* call = make_runtime_call(RC_LEAF,
6838 OptoRuntime::montgomerySquare_Type(),
6839 stubAddr, stubName, TypePtr::BOTTOM,
6840 a_start, n_start, len, inv, top(),
6841 m_start);
6842 set_result(m);
6843 }
6844
6845 return true;
6846 }
6847
6848 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6849 address stubAddr = nullptr;
6850 const char* stubName = nullptr;
6851
6852 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6853 if (stubAddr == nullptr) {
6854 return false; // Intrinsic's stub is not implemented on this platform
6855 }
6856
6857 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6858
6859 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6860
6861 Node* newArr = argument(0);
6862 Node* oldArr = argument(1);
6863 Node* newIdx = argument(2);
6864 Node* shiftCount = argument(3);
6865 Node* numIter = argument(4);
6866
6867 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6868 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6869 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6870 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6871 return false;
6872 }
6873
6874 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6875 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6876 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6877 return false;
6878 }
6879
6880 // Make the call
6881 {
6882 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6883 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6884
6885 Node* call = make_runtime_call(RC_LEAF,
6886 OptoRuntime::bigIntegerShift_Type(),
6887 stubAddr,
6888 stubName,
6889 TypePtr::BOTTOM,
6890 newArr_start,
6891 oldArr_start,
6892 newIdx,
6893 shiftCount,
6894 numIter);
6895 }
6896
6897 return true;
6898 }
6899
6900 //-------------inline_vectorizedMismatch------------------------------
6901 bool LibraryCallKit::inline_vectorizedMismatch() {
6902 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6903
6904 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6905 Node* obja = argument(0); // Object
6906 Node* aoffset = argument(1); // long
6907 Node* objb = argument(3); // Object
6908 Node* boffset = argument(4); // long
6909 Node* length = argument(6); // int
6910 Node* scale = argument(7); // int
6911
6912 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6913 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6914 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6915 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6916 scale == top()) {
6917 return false; // failed input validation
6918 }
6919
6920 Node* obja_adr = make_unsafe_address(obja, aoffset);
6921 Node* objb_adr = make_unsafe_address(objb, boffset);
6922
6923 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6924 //
6925 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6926 // if (length <= inline_limit) {
6927 // inline_path:
6928 // vmask = VectorMaskGen length
6929 // vload1 = LoadVectorMasked obja, vmask
6930 // vload2 = LoadVectorMasked objb, vmask
6931 // result1 = VectorCmpMasked vload1, vload2, vmask
6932 // } else {
6933 // call_stub_path:
6934 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6935 // }
6936 // exit_block:
6937 // return Phi(result1, result2);
6938 //
6939 enum { inline_path = 1, // input is small enough to process it all at once
6940 stub_path = 2, // input is too large; call into the VM
6941 PATH_LIMIT = 3
6942 };
6943
6944 Node* exit_block = new RegionNode(PATH_LIMIT);
6945 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6946 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6947
6948 Node* call_stub_path = control();
6949
6950 BasicType elem_bt = T_ILLEGAL;
6951
6952 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6953 if (scale_t->is_con()) {
6954 switch (scale_t->get_con()) {
6955 case 0: elem_bt = T_BYTE; break;
6956 case 1: elem_bt = T_SHORT; break;
6957 case 2: elem_bt = T_INT; break;
6958 case 3: elem_bt = T_LONG; break;
6959
6960 default: elem_bt = T_ILLEGAL; break; // not supported
6961 }
6962 }
6963
6964 int inline_limit = 0;
6965 bool do_partial_inline = false;
6966
6967 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6968 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6969 do_partial_inline = inline_limit >= 16;
6970 }
6971
6972 if (do_partial_inline) {
6973 assert(elem_bt != T_ILLEGAL, "sanity");
6974
6975 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6976 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6977 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6978
6979 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6980 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6981 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6982
6983 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6984
6985 if (!stopped()) {
6986 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6987
6988 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6989 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6990 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6991 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6992
6993 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6994 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6995 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6996 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6997
6998 exit_block->init_req(inline_path, control());
6999 memory_phi->init_req(inline_path, map()->memory());
7000 result_phi->init_req(inline_path, result);
7001
7002 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7003 clear_upper_avx();
7004 }
7005 }
7006 }
7007
7008 if (call_stub_path != nullptr) {
7009 set_control(call_stub_path);
7010
7011 Node* call = make_runtime_call(RC_LEAF,
7012 OptoRuntime::vectorizedMismatch_Type(),
7013 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7014 obja_adr, objb_adr, length, scale);
7015
7016 exit_block->init_req(stub_path, control());
7017 memory_phi->init_req(stub_path, map()->memory());
7018 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7019 }
7020
7021 exit_block = _gvn.transform(exit_block);
7022 memory_phi = _gvn.transform(memory_phi);
7023 result_phi = _gvn.transform(result_phi);
7024
7025 set_control(exit_block);
7026 set_all_memory(memory_phi);
7027 set_result(result_phi);
7028
7029 return true;
7030 }
7031
7032 //------------------------------inline_vectorizedHashcode----------------------------
7033 bool LibraryCallKit::inline_vectorizedHashCode() {
7034 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7035
7036 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7037 Node* array = argument(0);
7038 Node* offset = argument(1);
7039 Node* length = argument(2);
7040 Node* initialValue = argument(3);
7041 Node* basic_type = argument(4);
7042
7043 if (basic_type == top()) {
7044 return false; // failed input validation
7045 }
7046
7047 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7048 if (!basic_type_t->is_con()) {
7049 return false; // Only intrinsify if mode argument is constant
7050 }
7051
7052 array = must_be_not_null(array, true);
7053
7054 BasicType bt = (BasicType)basic_type_t->get_con();
7055
7056 // Resolve address of first element
7057 Node* array_start = array_element_address(array, offset, bt);
7058
7059 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7060 array_start, length, initialValue, basic_type)));
7061 clear_upper_avx();
7062
7063 return true;
7064 }
7065
7066 /**
7067 * Calculate CRC32 for byte.
7068 * int java.util.zip.CRC32.update(int crc, int b)
7069 */
7070 bool LibraryCallKit::inline_updateCRC32() {
7071 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7072 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7073 // no receiver since it is static method
7074 Node* crc = argument(0); // type: int
7075 Node* b = argument(1); // type: int
7076
7077 /*
7078 * int c = ~ crc;
7079 * b = timesXtoThe32[(b ^ c) & 0xFF];
7080 * b = b ^ (c >>> 8);
7081 * crc = ~b;
7082 */
7083
7084 Node* M1 = intcon(-1);
7085 crc = _gvn.transform(new XorINode(crc, M1));
7086 Node* result = _gvn.transform(new XorINode(crc, b));
7087 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7088
7089 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7090 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7091 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7092 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7093
7094 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7095 result = _gvn.transform(new XorINode(crc, result));
7096 result = _gvn.transform(new XorINode(result, M1));
7097 set_result(result);
7098 return true;
7099 }
7100
7101 /**
7102 * Calculate CRC32 for byte[] array.
7103 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7104 */
7105 bool LibraryCallKit::inline_updateBytesCRC32() {
7106 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7107 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7108 // no receiver since it is static method
7109 Node* crc = argument(0); // type: int
7110 Node* src = argument(1); // type: oop
7111 Node* offset = argument(2); // type: int
7112 Node* length = argument(3); // type: int
7113
7114 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7115 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7116 // failed array check
7117 return false;
7118 }
7119
7120 // Figure out the size and type of the elements we will be copying.
7121 BasicType src_elem = src_type->elem()->array_element_basic_type();
7122 if (src_elem != T_BYTE) {
7123 return false;
7124 }
7125
7126 // 'src_start' points to src array + scaled offset
7127 src = must_be_not_null(src, true);
7128 Node* src_start = array_element_address(src, offset, src_elem);
7129
7130 // We assume that range check is done by caller.
7131 // TODO: generate range check (offset+length < src.length) in debug VM.
7132
7133 // Call the stub.
7134 address stubAddr = StubRoutines::updateBytesCRC32();
7135 const char *stubName = "updateBytesCRC32";
7136
7137 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7138 stubAddr, stubName, TypePtr::BOTTOM,
7139 crc, src_start, length);
7140 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7141 set_result(result);
7142 return true;
7143 }
7144
7145 /**
7146 * Calculate CRC32 for ByteBuffer.
7147 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7148 */
7149 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7150 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7151 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7152 // no receiver since it is static method
7153 Node* crc = argument(0); // type: int
7154 Node* src = argument(1); // type: long
7155 Node* offset = argument(3); // type: int
7156 Node* length = argument(4); // type: int
7157
7158 src = ConvL2X(src); // adjust Java long to machine word
7159 Node* base = _gvn.transform(new CastX2PNode(src));
7160 offset = ConvI2X(offset);
7161
7162 // 'src_start' points to src array + scaled offset
7163 Node* src_start = basic_plus_adr(top(), base, offset);
7164
7165 // Call the stub.
7166 address stubAddr = StubRoutines::updateBytesCRC32();
7167 const char *stubName = "updateBytesCRC32";
7168
7169 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7170 stubAddr, stubName, TypePtr::BOTTOM,
7171 crc, src_start, length);
7172 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7173 set_result(result);
7174 return true;
7175 }
7176
7177 //------------------------------get_table_from_crc32c_class-----------------------
7178 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7179 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7180 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7181
7182 return table;
7183 }
7184
7185 //------------------------------inline_updateBytesCRC32C-----------------------
7186 //
7187 // Calculate CRC32C for byte[] array.
7188 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7189 //
7190 bool LibraryCallKit::inline_updateBytesCRC32C() {
7191 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7192 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7193 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7194 // no receiver since it is a static method
7195 Node* crc = argument(0); // type: int
7196 Node* src = argument(1); // type: oop
7197 Node* offset = argument(2); // type: int
7198 Node* end = argument(3); // type: int
7199
7200 Node* length = _gvn.transform(new SubINode(end, offset));
7201
7202 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7203 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7204 // failed array check
7205 return false;
7206 }
7207
7208 // Figure out the size and type of the elements we will be copying.
7209 BasicType src_elem = src_type->elem()->array_element_basic_type();
7210 if (src_elem != T_BYTE) {
7211 return false;
7212 }
7213
7214 // 'src_start' points to src array + scaled offset
7215 src = must_be_not_null(src, true);
7216 Node* src_start = array_element_address(src, offset, src_elem);
7217
7218 // static final int[] byteTable in class CRC32C
7219 Node* table = get_table_from_crc32c_class(callee()->holder());
7220 table = must_be_not_null(table, true);
7221 Node* table_start = array_element_address(table, intcon(0), T_INT);
7222
7223 // We assume that range check is done by caller.
7224 // TODO: generate range check (offset+length < src.length) in debug VM.
7225
7226 // Call the stub.
7227 address stubAddr = StubRoutines::updateBytesCRC32C();
7228 const char *stubName = "updateBytesCRC32C";
7229
7230 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7231 stubAddr, stubName, TypePtr::BOTTOM,
7232 crc, src_start, length, table_start);
7233 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7234 set_result(result);
7235 return true;
7236 }
7237
7238 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7239 //
7240 // Calculate CRC32C for DirectByteBuffer.
7241 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7242 //
7243 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7244 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7245 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7246 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7247 // no receiver since it is a static method
7248 Node* crc = argument(0); // type: int
7249 Node* src = argument(1); // type: long
7250 Node* offset = argument(3); // type: int
7251 Node* end = argument(4); // type: int
7252
7253 Node* length = _gvn.transform(new SubINode(end, offset));
7254
7255 src = ConvL2X(src); // adjust Java long to machine word
7256 Node* base = _gvn.transform(new CastX2PNode(src));
7257 offset = ConvI2X(offset);
7258
7259 // 'src_start' points to src array + scaled offset
7260 Node* src_start = basic_plus_adr(top(), base, offset);
7261
7262 // static final int[] byteTable in class CRC32C
7263 Node* table = get_table_from_crc32c_class(callee()->holder());
7264 table = must_be_not_null(table, true);
7265 Node* table_start = array_element_address(table, intcon(0), T_INT);
7266
7267 // Call the stub.
7268 address stubAddr = StubRoutines::updateBytesCRC32C();
7269 const char *stubName = "updateBytesCRC32C";
7270
7271 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7272 stubAddr, stubName, TypePtr::BOTTOM,
7273 crc, src_start, length, table_start);
7274 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7275 set_result(result);
7276 return true;
7277 }
7278
7279 //------------------------------inline_updateBytesAdler32----------------------
7280 //
7281 // Calculate Adler32 checksum for byte[] array.
7282 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7283 //
7284 bool LibraryCallKit::inline_updateBytesAdler32() {
7285 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7286 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7287 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7288 // no receiver since it is static method
7289 Node* crc = argument(0); // type: int
7290 Node* src = argument(1); // type: oop
7291 Node* offset = argument(2); // type: int
7292 Node* length = argument(3); // type: int
7293
7294 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7295 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7296 // failed array check
7297 return false;
7298 }
7299
7300 // Figure out the size and type of the elements we will be copying.
7301 BasicType src_elem = src_type->elem()->array_element_basic_type();
7302 if (src_elem != T_BYTE) {
7303 return false;
7304 }
7305
7306 // 'src_start' points to src array + scaled offset
7307 Node* src_start = array_element_address(src, offset, src_elem);
7308
7309 // We assume that range check is done by caller.
7310 // TODO: generate range check (offset+length < src.length) in debug VM.
7311
7312 // Call the stub.
7313 address stubAddr = StubRoutines::updateBytesAdler32();
7314 const char *stubName = "updateBytesAdler32";
7315
7316 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7317 stubAddr, stubName, TypePtr::BOTTOM,
7318 crc, src_start, length);
7319 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7320 set_result(result);
7321 return true;
7322 }
7323
7324 //------------------------------inline_updateByteBufferAdler32---------------
7325 //
7326 // Calculate Adler32 checksum for DirectByteBuffer.
7327 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7328 //
7329 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7330 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7331 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7332 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7333 // no receiver since it is static method
7334 Node* crc = argument(0); // type: int
7335 Node* src = argument(1); // type: long
7336 Node* offset = argument(3); // type: int
7337 Node* length = argument(4); // type: int
7338
7339 src = ConvL2X(src); // adjust Java long to machine word
7340 Node* base = _gvn.transform(new CastX2PNode(src));
7341 offset = ConvI2X(offset);
7342
7343 // 'src_start' points to src array + scaled offset
7344 Node* src_start = basic_plus_adr(top(), base, offset);
7345
7346 // Call the stub.
7347 address stubAddr = StubRoutines::updateBytesAdler32();
7348 const char *stubName = "updateBytesAdler32";
7349
7350 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7351 stubAddr, stubName, TypePtr::BOTTOM,
7352 crc, src_start, length);
7353
7354 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7355 set_result(result);
7356 return true;
7357 }
7358
7359 //----------------------------inline_reference_get----------------------------
7360 // public T java.lang.ref.Reference.get();
7361 bool LibraryCallKit::inline_reference_get() {
7362 const int referent_offset = java_lang_ref_Reference::referent_offset();
7363
7364 // Get the argument:
7365 Node* reference_obj = null_check_receiver();
7366 if (stopped()) return true;
7367
7368 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7369 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7370 decorators, /*is_static*/ false, nullptr);
7371 if (result == nullptr) return false;
7372
7373 // Add memory barrier to prevent commoning reads from this field
7374 // across safepoint since GC can change its value.
7375 insert_mem_bar(Op_MemBarCPUOrder);
7376
7377 set_result(result);
7378 return true;
7379 }
7380
7381 //----------------------------inline_reference_refersTo0----------------------------
7382 // bool java.lang.ref.Reference.refersTo0();
7383 // bool java.lang.ref.PhantomReference.refersTo0();
7384 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7385 // Get arguments:
7386 Node* reference_obj = null_check_receiver();
7387 Node* other_obj = argument(1);
7388 if (stopped()) return true;
7389
7390 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7391 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7392 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7393 decorators, /*is_static*/ false, nullptr);
7394 if (referent == nullptr) return false;
7395
7396 // Add memory barrier to prevent commoning reads from this field
7397 // across safepoint since GC can change its value.
7398 insert_mem_bar(Op_MemBarCPUOrder);
7399
7400 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7401 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7402 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7403
7404 RegionNode* region = new RegionNode(3);
7405 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7406
7407 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7408 region->init_req(1, if_true);
7409 phi->init_req(1, intcon(1));
7410
7411 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7412 region->init_req(2, if_false);
7413 phi->init_req(2, intcon(0));
7414
7415 set_control(_gvn.transform(region));
7416 record_for_igvn(region);
7417 set_result(_gvn.transform(phi));
7418 return true;
7419 }
7420
7421 //----------------------------inline_reference_clear0----------------------------
7422 // void java.lang.ref.Reference.clear0();
7423 // void java.lang.ref.PhantomReference.clear0();
7424 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7425 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7426
7427 // Get arguments
7428 Node* reference_obj = null_check_receiver();
7429 if (stopped()) return true;
7430
7431 // Common access parameters
7432 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7433 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7434 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7435 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7436 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7437
7438 Node* referent = access_load_at(reference_obj,
7439 referent_field_addr,
7440 referent_field_addr_type,
7441 val_type,
7442 T_OBJECT,
7443 decorators);
7444
7445 IdealKit ideal(this);
7446 #define __ ideal.
7447 __ if_then(referent, BoolTest::ne, null());
7448 sync_kit(ideal);
7449 access_store_at(reference_obj,
7450 referent_field_addr,
7451 referent_field_addr_type,
7452 null(),
7453 val_type,
7454 T_OBJECT,
7455 decorators);
7456 __ sync_kit(this);
7457 __ end_if();
7458 final_sync(ideal);
7459 #undef __
7460
7461 return true;
7462 }
7463
7464 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7465 DecoratorSet decorators, bool is_static,
7466 ciInstanceKlass* fromKls) {
7467 if (fromKls == nullptr) {
7468 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7469 assert(tinst != nullptr, "obj is null");
7470 assert(tinst->is_loaded(), "obj is not loaded");
7471 fromKls = tinst->instance_klass();
7472 } else {
7473 assert(is_static, "only for static field access");
7474 }
7475 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7476 ciSymbol::make(fieldTypeString),
7477 is_static);
7478
7479 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7480 if (field == nullptr) return (Node *) nullptr;
7481
7482 if (is_static) {
7483 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7484 fromObj = makecon(tip);
7485 }
7486
7487 // Next code copied from Parse::do_get_xxx():
7488
7489 // Compute address and memory type.
7490 int offset = field->offset_in_bytes();
7491 bool is_vol = field->is_volatile();
7492 ciType* field_klass = field->type();
7493 assert(field_klass->is_loaded(), "should be loaded");
7494 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7495 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7496 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7497 "slice of address and input slice don't match");
7498 BasicType bt = field->layout_type();
7499
7500 // Build the resultant type of the load
7501 const Type *type;
7502 if (bt == T_OBJECT) {
7503 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7504 } else {
7505 type = Type::get_const_basic_type(bt);
7506 }
7507
7508 if (is_vol) {
7509 decorators |= MO_SEQ_CST;
7510 }
7511
7512 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7513 }
7514
7515 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7516 bool is_exact /* true */, bool is_static /* false */,
7517 ciInstanceKlass * fromKls /* nullptr */) {
7518 if (fromKls == nullptr) {
7519 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7520 assert(tinst != nullptr, "obj is null");
7521 assert(tinst->is_loaded(), "obj is not loaded");
7522 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7523 fromKls = tinst->instance_klass();
7524 }
7525 else {
7526 assert(is_static, "only for static field access");
7527 }
7528 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7529 ciSymbol::make(fieldTypeString),
7530 is_static);
7531
7532 assert(field != nullptr, "undefined field");
7533 assert(!field->is_volatile(), "not defined for volatile fields");
7534
7535 if (is_static) {
7536 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7537 fromObj = makecon(tip);
7538 }
7539
7540 // Next code copied from Parse::do_get_xxx():
7541
7542 // Compute address and memory type.
7543 int offset = field->offset_in_bytes();
7544 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7545
7546 return adr;
7547 }
7548
7549 //------------------------------inline_aescrypt_Block-----------------------
7550 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7551 address stubAddr = nullptr;
7552 const char *stubName;
7553 assert(UseAES, "need AES instruction support");
7554
7555 switch(id) {
7556 case vmIntrinsics::_aescrypt_encryptBlock:
7557 stubAddr = StubRoutines::aescrypt_encryptBlock();
7558 stubName = "aescrypt_encryptBlock";
7559 break;
7560 case vmIntrinsics::_aescrypt_decryptBlock:
7561 stubAddr = StubRoutines::aescrypt_decryptBlock();
7562 stubName = "aescrypt_decryptBlock";
7563 break;
7564 default:
7565 break;
7566 }
7567 if (stubAddr == nullptr) return false;
7568
7569 Node* aescrypt_object = argument(0);
7570 Node* src = argument(1);
7571 Node* src_offset = argument(2);
7572 Node* dest = argument(3);
7573 Node* dest_offset = argument(4);
7574
7575 src = must_be_not_null(src, true);
7576 dest = must_be_not_null(dest, true);
7577
7578 // (1) src and dest are arrays.
7579 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7580 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7581 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7582 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7583
7584 // for the quick and dirty code we will skip all the checks.
7585 // we are just trying to get the call to be generated.
7586 Node* src_start = src;
7587 Node* dest_start = dest;
7588 if (src_offset != nullptr || dest_offset != nullptr) {
7589 assert(src_offset != nullptr && dest_offset != nullptr, "");
7590 src_start = array_element_address(src, src_offset, T_BYTE);
7591 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7592 }
7593
7594 // now need to get the start of its expanded key array
7595 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7596 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7597 if (k_start == nullptr) return false;
7598
7599 // Call the stub.
7600 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7601 stubAddr, stubName, TypePtr::BOTTOM,
7602 src_start, dest_start, k_start);
7603
7604 return true;
7605 }
7606
7607 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7608 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7609 address stubAddr = nullptr;
7610 const char *stubName = nullptr;
7611
7612 assert(UseAES, "need AES instruction support");
7613
7614 switch(id) {
7615 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7616 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7617 stubName = "cipherBlockChaining_encryptAESCrypt";
7618 break;
7619 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7620 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7621 stubName = "cipherBlockChaining_decryptAESCrypt";
7622 break;
7623 default:
7624 break;
7625 }
7626 if (stubAddr == nullptr) return false;
7627
7628 Node* cipherBlockChaining_object = argument(0);
7629 Node* src = argument(1);
7630 Node* src_offset = argument(2);
7631 Node* len = argument(3);
7632 Node* dest = argument(4);
7633 Node* dest_offset = argument(5);
7634
7635 src = must_be_not_null(src, false);
7636 dest = must_be_not_null(dest, false);
7637
7638 // (1) src and dest are arrays.
7639 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7640 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7641 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7642 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7643
7644 // checks are the responsibility of the caller
7645 Node* src_start = src;
7646 Node* dest_start = dest;
7647 if (src_offset != nullptr || dest_offset != nullptr) {
7648 assert(src_offset != nullptr && dest_offset != nullptr, "");
7649 src_start = array_element_address(src, src_offset, T_BYTE);
7650 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7651 }
7652
7653 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7654 // (because of the predicated logic executed earlier).
7655 // so we cast it here safely.
7656 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7657
7658 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7659 if (embeddedCipherObj == nullptr) return false;
7660
7661 // cast it to what we know it will be at runtime
7662 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7663 assert(tinst != nullptr, "CBC obj is null");
7664 assert(tinst->is_loaded(), "CBC obj is not loaded");
7665 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7666 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7667
7668 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7669 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7670 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7671 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7672 aescrypt_object = _gvn.transform(aescrypt_object);
7673
7674 // we need to get the start of the aescrypt_object's expanded key array
7675 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7676 if (k_start == nullptr) return false;
7677
7678 // similarly, get the start address of the r vector
7679 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7680 if (objRvec == nullptr) return false;
7681 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7682
7683 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7684 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7685 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7686 stubAddr, stubName, TypePtr::BOTTOM,
7687 src_start, dest_start, k_start, r_start, len);
7688
7689 // return cipher length (int)
7690 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7691 set_result(retvalue);
7692 return true;
7693 }
7694
7695 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7696 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7697 address stubAddr = nullptr;
7698 const char *stubName = nullptr;
7699
7700 assert(UseAES, "need AES instruction support");
7701
7702 switch (id) {
7703 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7704 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7705 stubName = "electronicCodeBook_encryptAESCrypt";
7706 break;
7707 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7708 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7709 stubName = "electronicCodeBook_decryptAESCrypt";
7710 break;
7711 default:
7712 break;
7713 }
7714
7715 if (stubAddr == nullptr) return false;
7716
7717 Node* electronicCodeBook_object = argument(0);
7718 Node* src = argument(1);
7719 Node* src_offset = argument(2);
7720 Node* len = argument(3);
7721 Node* dest = argument(4);
7722 Node* dest_offset = argument(5);
7723
7724 // (1) src and dest are arrays.
7725 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7726 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7727 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7728 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7729
7730 // checks are the responsibility of the caller
7731 Node* src_start = src;
7732 Node* dest_start = dest;
7733 if (src_offset != nullptr || dest_offset != nullptr) {
7734 assert(src_offset != nullptr && dest_offset != nullptr, "");
7735 src_start = array_element_address(src, src_offset, T_BYTE);
7736 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7737 }
7738
7739 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7740 // (because of the predicated logic executed earlier).
7741 // so we cast it here safely.
7742 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7743
7744 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7745 if (embeddedCipherObj == nullptr) return false;
7746
7747 // cast it to what we know it will be at runtime
7748 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7749 assert(tinst != nullptr, "ECB obj is null");
7750 assert(tinst->is_loaded(), "ECB obj is not loaded");
7751 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7752 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7753
7754 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7755 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7756 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7757 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7758 aescrypt_object = _gvn.transform(aescrypt_object);
7759
7760 // we need to get the start of the aescrypt_object's expanded key array
7761 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7762 if (k_start == nullptr) return false;
7763
7764 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7765 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7766 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7767 stubAddr, stubName, TypePtr::BOTTOM,
7768 src_start, dest_start, k_start, len);
7769
7770 // return cipher length (int)
7771 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7772 set_result(retvalue);
7773 return true;
7774 }
7775
7776 //------------------------------inline_counterMode_AESCrypt-----------------------
7777 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7778 assert(UseAES, "need AES instruction support");
7779 if (!UseAESCTRIntrinsics) return false;
7780
7781 address stubAddr = nullptr;
7782 const char *stubName = nullptr;
7783 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7784 stubAddr = StubRoutines::counterMode_AESCrypt();
7785 stubName = "counterMode_AESCrypt";
7786 }
7787 if (stubAddr == nullptr) return false;
7788
7789 Node* counterMode_object = argument(0);
7790 Node* src = argument(1);
7791 Node* src_offset = argument(2);
7792 Node* len = argument(3);
7793 Node* dest = argument(4);
7794 Node* dest_offset = argument(5);
7795
7796 // (1) src and dest are arrays.
7797 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7798 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7799 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7800 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7801
7802 // checks are the responsibility of the caller
7803 Node* src_start = src;
7804 Node* dest_start = dest;
7805 if (src_offset != nullptr || dest_offset != nullptr) {
7806 assert(src_offset != nullptr && dest_offset != nullptr, "");
7807 src_start = array_element_address(src, src_offset, T_BYTE);
7808 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7809 }
7810
7811 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7812 // (because of the predicated logic executed earlier).
7813 // so we cast it here safely.
7814 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7815 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7816 if (embeddedCipherObj == nullptr) return false;
7817 // cast it to what we know it will be at runtime
7818 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7819 assert(tinst != nullptr, "CTR obj is null");
7820 assert(tinst->is_loaded(), "CTR obj is not loaded");
7821 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7822 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7823 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7824 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7825 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7826 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7827 aescrypt_object = _gvn.transform(aescrypt_object);
7828 // we need to get the start of the aescrypt_object's expanded key array
7829 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7830 if (k_start == nullptr) return false;
7831 // similarly, get the start address of the r vector
7832 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7833 if (obj_counter == nullptr) return false;
7834 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7835
7836 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7837 if (saved_encCounter == nullptr) return false;
7838 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7839 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7840
7841 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7842 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7843 OptoRuntime::counterMode_aescrypt_Type(),
7844 stubAddr, stubName, TypePtr::BOTTOM,
7845 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7846
7847 // return cipher length (int)
7848 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7849 set_result(retvalue);
7850 return true;
7851 }
7852
7853 //------------------------------get_key_start_from_aescrypt_object-----------------------
7854 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7855 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7856 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7857 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7858 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7859 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7860 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7861 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7862 if (objSessionK == nullptr) {
7863 return (Node *) nullptr;
7864 }
7865 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7866 #else
7867 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7868 #endif // PPC64
7869 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7870 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7871
7872 // now have the array, need to get the start address of the K array
7873 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7874 return k_start;
7875 }
7876
7877 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7878 // Return node representing slow path of predicate check.
7879 // the pseudo code we want to emulate with this predicate is:
7880 // for encryption:
7881 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7882 // for decryption:
7883 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7884 // note cipher==plain is more conservative than the original java code but that's OK
7885 //
7886 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7887 // The receiver was checked for null already.
7888 Node* objCBC = argument(0);
7889
7890 Node* src = argument(1);
7891 Node* dest = argument(4);
7892
7893 // Load embeddedCipher field of CipherBlockChaining object.
7894 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7895
7896 // get AESCrypt klass for instanceOf check
7897 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7898 // will have same classloader as CipherBlockChaining object
7899 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7900 assert(tinst != nullptr, "CBCobj is null");
7901 assert(tinst->is_loaded(), "CBCobj is not loaded");
7902
7903 // we want to do an instanceof comparison against the AESCrypt class
7904 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7905 if (!klass_AESCrypt->is_loaded()) {
7906 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7907 Node* ctrl = control();
7908 set_control(top()); // no regular fast path
7909 return ctrl;
7910 }
7911
7912 src = must_be_not_null(src, true);
7913 dest = must_be_not_null(dest, true);
7914
7915 // Resolve oops to stable for CmpP below.
7916 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7917
7918 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7919 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7920 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7921
7922 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7923
7924 // for encryption, we are done
7925 if (!decrypting)
7926 return instof_false; // even if it is null
7927
7928 // for decryption, we need to add a further check to avoid
7929 // taking the intrinsic path when cipher and plain are the same
7930 // see the original java code for why.
7931 RegionNode* region = new RegionNode(3);
7932 region->init_req(1, instof_false);
7933
7934 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7935 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7936 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7937 region->init_req(2, src_dest_conjoint);
7938
7939 record_for_igvn(region);
7940 return _gvn.transform(region);
7941 }
7942
7943 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7944 // Return node representing slow path of predicate check.
7945 // the pseudo code we want to emulate with this predicate is:
7946 // for encryption:
7947 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7948 // for decryption:
7949 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7950 // note cipher==plain is more conservative than the original java code but that's OK
7951 //
7952 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7953 // The receiver was checked for null already.
7954 Node* objECB = argument(0);
7955
7956 // Load embeddedCipher field of ElectronicCodeBook object.
7957 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7958
7959 // get AESCrypt klass for instanceOf check
7960 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7961 // will have same classloader as ElectronicCodeBook object
7962 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7963 assert(tinst != nullptr, "ECBobj is null");
7964 assert(tinst->is_loaded(), "ECBobj is not loaded");
7965
7966 // we want to do an instanceof comparison against the AESCrypt class
7967 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7968 if (!klass_AESCrypt->is_loaded()) {
7969 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7970 Node* ctrl = control();
7971 set_control(top()); // no regular fast path
7972 return ctrl;
7973 }
7974 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7975
7976 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7977 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7978 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7979
7980 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7981
7982 // for encryption, we are done
7983 if (!decrypting)
7984 return instof_false; // even if it is null
7985
7986 // for decryption, we need to add a further check to avoid
7987 // taking the intrinsic path when cipher and plain are the same
7988 // see the original java code for why.
7989 RegionNode* region = new RegionNode(3);
7990 region->init_req(1, instof_false);
7991 Node* src = argument(1);
7992 Node* dest = argument(4);
7993 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7994 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7995 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7996 region->init_req(2, src_dest_conjoint);
7997
7998 record_for_igvn(region);
7999 return _gvn.transform(region);
8000 }
8001
8002 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8003 // Return node representing slow path of predicate check.
8004 // the pseudo code we want to emulate with this predicate is:
8005 // for encryption:
8006 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8007 // for decryption:
8008 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8009 // note cipher==plain is more conservative than the original java code but that's OK
8010 //
8011
8012 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8013 // The receiver was checked for null already.
8014 Node* objCTR = argument(0);
8015
8016 // Load embeddedCipher field of CipherBlockChaining object.
8017 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8018
8019 // get AESCrypt klass for instanceOf check
8020 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8021 // will have same classloader as CipherBlockChaining object
8022 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8023 assert(tinst != nullptr, "CTRobj is null");
8024 assert(tinst->is_loaded(), "CTRobj is not loaded");
8025
8026 // we want to do an instanceof comparison against the AESCrypt class
8027 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8028 if (!klass_AESCrypt->is_loaded()) {
8029 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8030 Node* ctrl = control();
8031 set_control(top()); // no regular fast path
8032 return ctrl;
8033 }
8034
8035 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8036 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8037 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8038 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8039 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8040
8041 return instof_false; // even if it is null
8042 }
8043
8044 //------------------------------inline_ghash_processBlocks
8045 bool LibraryCallKit::inline_ghash_processBlocks() {
8046 address stubAddr;
8047 const char *stubName;
8048 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8049
8050 stubAddr = StubRoutines::ghash_processBlocks();
8051 stubName = "ghash_processBlocks";
8052
8053 Node* data = argument(0);
8054 Node* offset = argument(1);
8055 Node* len = argument(2);
8056 Node* state = argument(3);
8057 Node* subkeyH = argument(4);
8058
8059 state = must_be_not_null(state, true);
8060 subkeyH = must_be_not_null(subkeyH, true);
8061 data = must_be_not_null(data, true);
8062
8063 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8064 assert(state_start, "state is null");
8065 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8066 assert(subkeyH_start, "subkeyH is null");
8067 Node* data_start = array_element_address(data, offset, T_BYTE);
8068 assert(data_start, "data is null");
8069
8070 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8071 OptoRuntime::ghash_processBlocks_Type(),
8072 stubAddr, stubName, TypePtr::BOTTOM,
8073 state_start, subkeyH_start, data_start, len);
8074 return true;
8075 }
8076
8077 //------------------------------inline_chacha20Block
8078 bool LibraryCallKit::inline_chacha20Block() {
8079 address stubAddr;
8080 const char *stubName;
8081 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8082
8083 stubAddr = StubRoutines::chacha20Block();
8084 stubName = "chacha20Block";
8085
8086 Node* state = argument(0);
8087 Node* result = argument(1);
8088
8089 state = must_be_not_null(state, true);
8090 result = must_be_not_null(result, true);
8091
8092 Node* state_start = array_element_address(state, intcon(0), T_INT);
8093 assert(state_start, "state is null");
8094 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8095 assert(result_start, "result is null");
8096
8097 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8098 OptoRuntime::chacha20Block_Type(),
8099 stubAddr, stubName, TypePtr::BOTTOM,
8100 state_start, result_start);
8101 // return key stream length (int)
8102 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8103 set_result(retvalue);
8104 return true;
8105 }
8106
8107 //------------------------------inline_kyberNtt
8108 bool LibraryCallKit::inline_kyberNtt() {
8109 address stubAddr;
8110 const char *stubName;
8111 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8112 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8113
8114 stubAddr = StubRoutines::kyberNtt();
8115 stubName = "kyberNtt";
8116 if (!stubAddr) return false;
8117
8118 Node* coeffs = argument(0);
8119 Node* ntt_zetas = argument(1);
8120
8121 coeffs = must_be_not_null(coeffs, true);
8122 ntt_zetas = must_be_not_null(ntt_zetas, true);
8123
8124 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8125 assert(coeffs_start, "coeffs is null");
8126 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8127 assert(ntt_zetas_start, "ntt_zetas is null");
8128 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8129 OptoRuntime::kyberNtt_Type(),
8130 stubAddr, stubName, TypePtr::BOTTOM,
8131 coeffs_start, ntt_zetas_start);
8132 // return an int
8133 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8134 set_result(retvalue);
8135 return true;
8136 }
8137
8138 //------------------------------inline_kyberInverseNtt
8139 bool LibraryCallKit::inline_kyberInverseNtt() {
8140 address stubAddr;
8141 const char *stubName;
8142 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8143 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8144
8145 stubAddr = StubRoutines::kyberInverseNtt();
8146 stubName = "kyberInverseNtt";
8147 if (!stubAddr) return false;
8148
8149 Node* coeffs = argument(0);
8150 Node* zetas = argument(1);
8151
8152 coeffs = must_be_not_null(coeffs, true);
8153 zetas = must_be_not_null(zetas, true);
8154
8155 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8156 assert(coeffs_start, "coeffs is null");
8157 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8158 assert(zetas_start, "inverseNtt_zetas is null");
8159 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8160 OptoRuntime::kyberInverseNtt_Type(),
8161 stubAddr, stubName, TypePtr::BOTTOM,
8162 coeffs_start, zetas_start);
8163
8164 // return an int
8165 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8166 set_result(retvalue);
8167 return true;
8168 }
8169
8170 //------------------------------inline_kyberNttMult
8171 bool LibraryCallKit::inline_kyberNttMult() {
8172 address stubAddr;
8173 const char *stubName;
8174 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8175 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8176
8177 stubAddr = StubRoutines::kyberNttMult();
8178 stubName = "kyberNttMult";
8179 if (!stubAddr) return false;
8180
8181 Node* result = argument(0);
8182 Node* ntta = argument(1);
8183 Node* nttb = argument(2);
8184 Node* zetas = argument(3);
8185
8186 result = must_be_not_null(result, true);
8187 ntta = must_be_not_null(ntta, true);
8188 nttb = must_be_not_null(nttb, true);
8189 zetas = must_be_not_null(zetas, true);
8190 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8191 assert(result_start, "result is null");
8192 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8193 assert(ntta_start, "ntta is null");
8194 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8195 assert(nttb_start, "nttb is null");
8196 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8197 assert(zetas_start, "nttMult_zetas is null");
8198 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8199 OptoRuntime::kyberNttMult_Type(),
8200 stubAddr, stubName, TypePtr::BOTTOM,
8201 result_start, ntta_start, nttb_start,
8202 zetas_start);
8203
8204 // return an int
8205 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8206 set_result(retvalue);
8207
8208 return true;
8209 }
8210
8211 //------------------------------inline_kyberAddPoly_2
8212 bool LibraryCallKit::inline_kyberAddPoly_2() {
8213 address stubAddr;
8214 const char *stubName;
8215 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8216 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8217
8218 stubAddr = StubRoutines::kyberAddPoly_2();
8219 stubName = "kyberAddPoly_2";
8220 if (!stubAddr) return false;
8221
8222 Node* result = argument(0);
8223 Node* a = argument(1);
8224 Node* b = argument(2);
8225
8226 result = must_be_not_null(result, true);
8227 a = must_be_not_null(a, true);
8228 b = must_be_not_null(b, true);
8229
8230 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8231 assert(result_start, "result is null");
8232 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8233 assert(a_start, "a is null");
8234 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8235 assert(b_start, "b is null");
8236 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8237 OptoRuntime::kyberAddPoly_2_Type(),
8238 stubAddr, stubName, TypePtr::BOTTOM,
8239 result_start, a_start, b_start);
8240 // return an int
8241 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8242 set_result(retvalue);
8243 return true;
8244 }
8245
8246 //------------------------------inline_kyberAddPoly_3
8247 bool LibraryCallKit::inline_kyberAddPoly_3() {
8248 address stubAddr;
8249 const char *stubName;
8250 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8251 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8252
8253 stubAddr = StubRoutines::kyberAddPoly_3();
8254 stubName = "kyberAddPoly_3";
8255 if (!stubAddr) return false;
8256
8257 Node* result = argument(0);
8258 Node* a = argument(1);
8259 Node* b = argument(2);
8260 Node* c = argument(3);
8261
8262 result = must_be_not_null(result, true);
8263 a = must_be_not_null(a, true);
8264 b = must_be_not_null(b, true);
8265 c = must_be_not_null(c, true);
8266
8267 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8268 assert(result_start, "result is null");
8269 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8270 assert(a_start, "a is null");
8271 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8272 assert(b_start, "b is null");
8273 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8274 assert(c_start, "c is null");
8275 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8276 OptoRuntime::kyberAddPoly_3_Type(),
8277 stubAddr, stubName, TypePtr::BOTTOM,
8278 result_start, a_start, b_start, c_start);
8279 // return an int
8280 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8281 set_result(retvalue);
8282 return true;
8283 }
8284
8285 //------------------------------inline_kyber12To16
8286 bool LibraryCallKit::inline_kyber12To16() {
8287 address stubAddr;
8288 const char *stubName;
8289 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8290 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8291
8292 stubAddr = StubRoutines::kyber12To16();
8293 stubName = "kyber12To16";
8294 if (!stubAddr) return false;
8295
8296 Node* condensed = argument(0);
8297 Node* condensedOffs = argument(1);
8298 Node* parsed = argument(2);
8299 Node* parsedLength = argument(3);
8300
8301 condensed = must_be_not_null(condensed, true);
8302 parsed = must_be_not_null(parsed, true);
8303
8304 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8305 assert(condensed_start, "condensed is null");
8306 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8307 assert(parsed_start, "parsed is null");
8308 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8309 OptoRuntime::kyber12To16_Type(),
8310 stubAddr, stubName, TypePtr::BOTTOM,
8311 condensed_start, condensedOffs, parsed_start, parsedLength);
8312 // return an int
8313 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8314 set_result(retvalue);
8315 return true;
8316
8317 }
8318
8319 //------------------------------inline_kyberBarrettReduce
8320 bool LibraryCallKit::inline_kyberBarrettReduce() {
8321 address stubAddr;
8322 const char *stubName;
8323 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8324 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8325
8326 stubAddr = StubRoutines::kyberBarrettReduce();
8327 stubName = "kyberBarrettReduce";
8328 if (!stubAddr) return false;
8329
8330 Node* coeffs = argument(0);
8331
8332 coeffs = must_be_not_null(coeffs, true);
8333
8334 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8335 assert(coeffs_start, "coeffs is null");
8336 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8337 OptoRuntime::kyberBarrettReduce_Type(),
8338 stubAddr, stubName, TypePtr::BOTTOM,
8339 coeffs_start);
8340 // return an int
8341 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8342 set_result(retvalue);
8343 return true;
8344 }
8345
8346 //------------------------------inline_dilithiumAlmostNtt
8347 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8348 address stubAddr;
8349 const char *stubName;
8350 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8351 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8352
8353 stubAddr = StubRoutines::dilithiumAlmostNtt();
8354 stubName = "dilithiumAlmostNtt";
8355 if (!stubAddr) return false;
8356
8357 Node* coeffs = argument(0);
8358 Node* ntt_zetas = argument(1);
8359
8360 coeffs = must_be_not_null(coeffs, true);
8361 ntt_zetas = must_be_not_null(ntt_zetas, true);
8362
8363 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8364 assert(coeffs_start, "coeffs is null");
8365 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8366 assert(ntt_zetas_start, "ntt_zetas is null");
8367 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8368 OptoRuntime::dilithiumAlmostNtt_Type(),
8369 stubAddr, stubName, TypePtr::BOTTOM,
8370 coeffs_start, ntt_zetas_start);
8371 // return an int
8372 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8373 set_result(retvalue);
8374 return true;
8375 }
8376
8377 //------------------------------inline_dilithiumAlmostInverseNtt
8378 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8379 address stubAddr;
8380 const char *stubName;
8381 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8382 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8383
8384 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8385 stubName = "dilithiumAlmostInverseNtt";
8386 if (!stubAddr) return false;
8387
8388 Node* coeffs = argument(0);
8389 Node* zetas = argument(1);
8390
8391 coeffs = must_be_not_null(coeffs, true);
8392 zetas = must_be_not_null(zetas, true);
8393
8394 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8395 assert(coeffs_start, "coeffs is null");
8396 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8397 assert(zetas_start, "inverseNtt_zetas is null");
8398 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8399 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8400 stubAddr, stubName, TypePtr::BOTTOM,
8401 coeffs_start, zetas_start);
8402 // return an int
8403 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8404 set_result(retvalue);
8405 return true;
8406 }
8407
8408 //------------------------------inline_dilithiumNttMult
8409 bool LibraryCallKit::inline_dilithiumNttMult() {
8410 address stubAddr;
8411 const char *stubName;
8412 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8413 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8414
8415 stubAddr = StubRoutines::dilithiumNttMult();
8416 stubName = "dilithiumNttMult";
8417 if (!stubAddr) return false;
8418
8419 Node* result = argument(0);
8420 Node* ntta = argument(1);
8421 Node* nttb = argument(2);
8422 Node* zetas = argument(3);
8423
8424 result = must_be_not_null(result, true);
8425 ntta = must_be_not_null(ntta, true);
8426 nttb = must_be_not_null(nttb, true);
8427 zetas = must_be_not_null(zetas, true);
8428
8429 Node* result_start = array_element_address(result, intcon(0), T_INT);
8430 assert(result_start, "result is null");
8431 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8432 assert(ntta_start, "ntta is null");
8433 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8434 assert(nttb_start, "nttb is null");
8435 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8436 OptoRuntime::dilithiumNttMult_Type(),
8437 stubAddr, stubName, TypePtr::BOTTOM,
8438 result_start, ntta_start, nttb_start);
8439
8440 // return an int
8441 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8442 set_result(retvalue);
8443
8444 return true;
8445 }
8446
8447 //------------------------------inline_dilithiumMontMulByConstant
8448 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8449 address stubAddr;
8450 const char *stubName;
8451 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8452 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8453
8454 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8455 stubName = "dilithiumMontMulByConstant";
8456 if (!stubAddr) return false;
8457
8458 Node* coeffs = argument(0);
8459 Node* constant = argument(1);
8460
8461 coeffs = must_be_not_null(coeffs, true);
8462
8463 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8464 assert(coeffs_start, "coeffs is null");
8465 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8466 OptoRuntime::dilithiumMontMulByConstant_Type(),
8467 stubAddr, stubName, TypePtr::BOTTOM,
8468 coeffs_start, constant);
8469
8470 // return an int
8471 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8472 set_result(retvalue);
8473 return true;
8474 }
8475
8476
8477 //------------------------------inline_dilithiumDecomposePoly
8478 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8479 address stubAddr;
8480 const char *stubName;
8481 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8482 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8483
8484 stubAddr = StubRoutines::dilithiumDecomposePoly();
8485 stubName = "dilithiumDecomposePoly";
8486 if (!stubAddr) return false;
8487
8488 Node* input = argument(0);
8489 Node* lowPart = argument(1);
8490 Node* highPart = argument(2);
8491 Node* twoGamma2 = argument(3);
8492 Node* multiplier = argument(4);
8493
8494 input = must_be_not_null(input, true);
8495 lowPart = must_be_not_null(lowPart, true);
8496 highPart = must_be_not_null(highPart, true);
8497
8498 Node* input_start = array_element_address(input, intcon(0), T_INT);
8499 assert(input_start, "input is null");
8500 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8501 assert(lowPart_start, "lowPart is null");
8502 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8503 assert(highPart_start, "highPart is null");
8504
8505 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8506 OptoRuntime::dilithiumDecomposePoly_Type(),
8507 stubAddr, stubName, TypePtr::BOTTOM,
8508 input_start, lowPart_start, highPart_start,
8509 twoGamma2, multiplier);
8510
8511 // return an int
8512 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8513 set_result(retvalue);
8514 return true;
8515 }
8516
8517 bool LibraryCallKit::inline_base64_encodeBlock() {
8518 address stubAddr;
8519 const char *stubName;
8520 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8521 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8522 stubAddr = StubRoutines::base64_encodeBlock();
8523 stubName = "encodeBlock";
8524
8525 if (!stubAddr) return false;
8526 Node* base64obj = argument(0);
8527 Node* src = argument(1);
8528 Node* offset = argument(2);
8529 Node* len = argument(3);
8530 Node* dest = argument(4);
8531 Node* dp = argument(5);
8532 Node* isURL = argument(6);
8533
8534 src = must_be_not_null(src, true);
8535 dest = must_be_not_null(dest, true);
8536
8537 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8538 assert(src_start, "source array is null");
8539 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8540 assert(dest_start, "destination array is null");
8541
8542 Node* base64 = make_runtime_call(RC_LEAF,
8543 OptoRuntime::base64_encodeBlock_Type(),
8544 stubAddr, stubName, TypePtr::BOTTOM,
8545 src_start, offset, len, dest_start, dp, isURL);
8546 return true;
8547 }
8548
8549 bool LibraryCallKit::inline_base64_decodeBlock() {
8550 address stubAddr;
8551 const char *stubName;
8552 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8553 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8554 stubAddr = StubRoutines::base64_decodeBlock();
8555 stubName = "decodeBlock";
8556
8557 if (!stubAddr) return false;
8558 Node* base64obj = argument(0);
8559 Node* src = argument(1);
8560 Node* src_offset = argument(2);
8561 Node* len = argument(3);
8562 Node* dest = argument(4);
8563 Node* dest_offset = argument(5);
8564 Node* isURL = argument(6);
8565 Node* isMIME = argument(7);
8566
8567 src = must_be_not_null(src, true);
8568 dest = must_be_not_null(dest, true);
8569
8570 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8571 assert(src_start, "source array is null");
8572 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8573 assert(dest_start, "destination array is null");
8574
8575 Node* call = make_runtime_call(RC_LEAF,
8576 OptoRuntime::base64_decodeBlock_Type(),
8577 stubAddr, stubName, TypePtr::BOTTOM,
8578 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8579 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8580 set_result(result);
8581 return true;
8582 }
8583
8584 bool LibraryCallKit::inline_poly1305_processBlocks() {
8585 address stubAddr;
8586 const char *stubName;
8587 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8588 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8589 stubAddr = StubRoutines::poly1305_processBlocks();
8590 stubName = "poly1305_processBlocks";
8591
8592 if (!stubAddr) return false;
8593 null_check_receiver(); // null-check receiver
8594 if (stopped()) return true;
8595
8596 Node* input = argument(1);
8597 Node* input_offset = argument(2);
8598 Node* len = argument(3);
8599 Node* alimbs = argument(4);
8600 Node* rlimbs = argument(5);
8601
8602 input = must_be_not_null(input, true);
8603 alimbs = must_be_not_null(alimbs, true);
8604 rlimbs = must_be_not_null(rlimbs, true);
8605
8606 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8607 assert(input_start, "input array is null");
8608 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8609 assert(acc_start, "acc array is null");
8610 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8611 assert(r_start, "r array is null");
8612
8613 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8614 OptoRuntime::poly1305_processBlocks_Type(),
8615 stubAddr, stubName, TypePtr::BOTTOM,
8616 input_start, len, acc_start, r_start);
8617 return true;
8618 }
8619
8620 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8621 address stubAddr;
8622 const char *stubName;
8623 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8624 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8625 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8626 stubName = "intpoly_montgomeryMult_P256";
8627
8628 if (!stubAddr) return false;
8629 null_check_receiver(); // null-check receiver
8630 if (stopped()) return true;
8631
8632 Node* a = argument(1);
8633 Node* b = argument(2);
8634 Node* r = argument(3);
8635
8636 a = must_be_not_null(a, true);
8637 b = must_be_not_null(b, true);
8638 r = must_be_not_null(r, true);
8639
8640 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8641 assert(a_start, "a array is null");
8642 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8643 assert(b_start, "b array is null");
8644 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8645 assert(r_start, "r array is null");
8646
8647 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8648 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8649 stubAddr, stubName, TypePtr::BOTTOM,
8650 a_start, b_start, r_start);
8651 return true;
8652 }
8653
8654 bool LibraryCallKit::inline_intpoly_assign() {
8655 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8656 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8657 const char *stubName = "intpoly_assign";
8658 address stubAddr = StubRoutines::intpoly_assign();
8659 if (!stubAddr) return false;
8660
8661 Node* set = argument(0);
8662 Node* a = argument(1);
8663 Node* b = argument(2);
8664 Node* arr_length = load_array_length(a);
8665
8666 a = must_be_not_null(a, true);
8667 b = must_be_not_null(b, true);
8668
8669 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8670 assert(a_start, "a array is null");
8671 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8672 assert(b_start, "b array is null");
8673
8674 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8675 OptoRuntime::intpoly_assign_Type(),
8676 stubAddr, stubName, TypePtr::BOTTOM,
8677 set, a_start, b_start, arr_length);
8678 return true;
8679 }
8680
8681 //------------------------------inline_digestBase_implCompress-----------------------
8682 //
8683 // Calculate MD5 for single-block byte[] array.
8684 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8685 //
8686 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8687 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8688 //
8689 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8690 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8691 //
8692 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8693 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8694 //
8695 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8696 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8697 //
8698 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8699 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8700
8701 Node* digestBase_obj = argument(0);
8702 Node* src = argument(1); // type oop
8703 Node* ofs = argument(2); // type int
8704
8705 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8706 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8707 // failed array check
8708 return false;
8709 }
8710 // Figure out the size and type of the elements we will be copying.
8711 BasicType src_elem = src_type->elem()->array_element_basic_type();
8712 if (src_elem != T_BYTE) {
8713 return false;
8714 }
8715 // 'src_start' points to src array + offset
8716 src = must_be_not_null(src, true);
8717 Node* src_start = array_element_address(src, ofs, src_elem);
8718 Node* state = nullptr;
8719 Node* block_size = nullptr;
8720 address stubAddr;
8721 const char *stubName;
8722
8723 switch(id) {
8724 case vmIntrinsics::_md5_implCompress:
8725 assert(UseMD5Intrinsics, "need MD5 instruction support");
8726 state = get_state_from_digest_object(digestBase_obj, T_INT);
8727 stubAddr = StubRoutines::md5_implCompress();
8728 stubName = "md5_implCompress";
8729 break;
8730 case vmIntrinsics::_sha_implCompress:
8731 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8732 state = get_state_from_digest_object(digestBase_obj, T_INT);
8733 stubAddr = StubRoutines::sha1_implCompress();
8734 stubName = "sha1_implCompress";
8735 break;
8736 case vmIntrinsics::_sha2_implCompress:
8737 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8738 state = get_state_from_digest_object(digestBase_obj, T_INT);
8739 stubAddr = StubRoutines::sha256_implCompress();
8740 stubName = "sha256_implCompress";
8741 break;
8742 case vmIntrinsics::_sha5_implCompress:
8743 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8744 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8745 stubAddr = StubRoutines::sha512_implCompress();
8746 stubName = "sha512_implCompress";
8747 break;
8748 case vmIntrinsics::_sha3_implCompress:
8749 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8750 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8751 stubAddr = StubRoutines::sha3_implCompress();
8752 stubName = "sha3_implCompress";
8753 block_size = get_block_size_from_digest_object(digestBase_obj);
8754 if (block_size == nullptr) return false;
8755 break;
8756 default:
8757 fatal_unexpected_iid(id);
8758 return false;
8759 }
8760 if (state == nullptr) return false;
8761
8762 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8763 if (stubAddr == nullptr) return false;
8764
8765 // Call the stub.
8766 Node* call;
8767 if (block_size == nullptr) {
8768 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8769 stubAddr, stubName, TypePtr::BOTTOM,
8770 src_start, state);
8771 } else {
8772 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8773 stubAddr, stubName, TypePtr::BOTTOM,
8774 src_start, state, block_size);
8775 }
8776
8777 return true;
8778 }
8779
8780 //------------------------------inline_double_keccak
8781 bool LibraryCallKit::inline_double_keccak() {
8782 address stubAddr;
8783 const char *stubName;
8784 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8785 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8786
8787 stubAddr = StubRoutines::double_keccak();
8788 stubName = "double_keccak";
8789 if (!stubAddr) return false;
8790
8791 Node* status0 = argument(0);
8792 Node* status1 = argument(1);
8793
8794 status0 = must_be_not_null(status0, true);
8795 status1 = must_be_not_null(status1, true);
8796
8797 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8798 assert(status0_start, "status0 is null");
8799 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8800 assert(status1_start, "status1 is null");
8801 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8802 OptoRuntime::double_keccak_Type(),
8803 stubAddr, stubName, TypePtr::BOTTOM,
8804 status0_start, status1_start);
8805 // return an int
8806 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8807 set_result(retvalue);
8808 return true;
8809 }
8810
8811
8812 //------------------------------inline_digestBase_implCompressMB-----------------------
8813 //
8814 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8815 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8816 //
8817 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8818 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8819 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8820 assert((uint)predicate < 5, "sanity");
8821 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8822
8823 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8824 Node* src = argument(1); // byte[] array
8825 Node* ofs = argument(2); // type int
8826 Node* limit = argument(3); // type int
8827
8828 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8829 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8830 // failed array check
8831 return false;
8832 }
8833 // Figure out the size and type of the elements we will be copying.
8834 BasicType src_elem = src_type->elem()->array_element_basic_type();
8835 if (src_elem != T_BYTE) {
8836 return false;
8837 }
8838 // 'src_start' points to src array + offset
8839 src = must_be_not_null(src, false);
8840 Node* src_start = array_element_address(src, ofs, src_elem);
8841
8842 const char* klass_digestBase_name = nullptr;
8843 const char* stub_name = nullptr;
8844 address stub_addr = nullptr;
8845 BasicType elem_type = T_INT;
8846
8847 switch (predicate) {
8848 case 0:
8849 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8850 klass_digestBase_name = "sun/security/provider/MD5";
8851 stub_name = "md5_implCompressMB";
8852 stub_addr = StubRoutines::md5_implCompressMB();
8853 }
8854 break;
8855 case 1:
8856 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8857 klass_digestBase_name = "sun/security/provider/SHA";
8858 stub_name = "sha1_implCompressMB";
8859 stub_addr = StubRoutines::sha1_implCompressMB();
8860 }
8861 break;
8862 case 2:
8863 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8864 klass_digestBase_name = "sun/security/provider/SHA2";
8865 stub_name = "sha256_implCompressMB";
8866 stub_addr = StubRoutines::sha256_implCompressMB();
8867 }
8868 break;
8869 case 3:
8870 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8871 klass_digestBase_name = "sun/security/provider/SHA5";
8872 stub_name = "sha512_implCompressMB";
8873 stub_addr = StubRoutines::sha512_implCompressMB();
8874 elem_type = T_LONG;
8875 }
8876 break;
8877 case 4:
8878 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8879 klass_digestBase_name = "sun/security/provider/SHA3";
8880 stub_name = "sha3_implCompressMB";
8881 stub_addr = StubRoutines::sha3_implCompressMB();
8882 elem_type = T_LONG;
8883 }
8884 break;
8885 default:
8886 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8887 }
8888 if (klass_digestBase_name != nullptr) {
8889 assert(stub_addr != nullptr, "Stub is generated");
8890 if (stub_addr == nullptr) return false;
8891
8892 // get DigestBase klass to lookup for SHA klass
8893 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8894 assert(tinst != nullptr, "digestBase_obj is not instance???");
8895 assert(tinst->is_loaded(), "DigestBase is not loaded");
8896
8897 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8898 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8899 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8900 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8901 }
8902 return false;
8903 }
8904
8905 //------------------------------inline_digestBase_implCompressMB-----------------------
8906 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8907 BasicType elem_type, address stubAddr, const char *stubName,
8908 Node* src_start, Node* ofs, Node* limit) {
8909 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8910 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8911 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8912 digest_obj = _gvn.transform(digest_obj);
8913
8914 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8915 if (state == nullptr) return false;
8916
8917 Node* block_size = nullptr;
8918 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8919 block_size = get_block_size_from_digest_object(digest_obj);
8920 if (block_size == nullptr) return false;
8921 }
8922
8923 // Call the stub.
8924 Node* call;
8925 if (block_size == nullptr) {
8926 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8927 OptoRuntime::digestBase_implCompressMB_Type(false),
8928 stubAddr, stubName, TypePtr::BOTTOM,
8929 src_start, state, ofs, limit);
8930 } else {
8931 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8932 OptoRuntime::digestBase_implCompressMB_Type(true),
8933 stubAddr, stubName, TypePtr::BOTTOM,
8934 src_start, state, block_size, ofs, limit);
8935 }
8936
8937 // return ofs (int)
8938 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8939 set_result(result);
8940
8941 return true;
8942 }
8943
8944 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8945 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8946 assert(UseAES, "need AES instruction support");
8947 address stubAddr = nullptr;
8948 const char *stubName = nullptr;
8949 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8950 stubName = "galoisCounterMode_AESCrypt";
8951
8952 if (stubAddr == nullptr) return false;
8953
8954 Node* in = argument(0);
8955 Node* inOfs = argument(1);
8956 Node* len = argument(2);
8957 Node* ct = argument(3);
8958 Node* ctOfs = argument(4);
8959 Node* out = argument(5);
8960 Node* outOfs = argument(6);
8961 Node* gctr_object = argument(7);
8962 Node* ghash_object = argument(8);
8963
8964 // (1) in, ct and out are arrays.
8965 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8966 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8967 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8968 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8969 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8970 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8971
8972 // checks are the responsibility of the caller
8973 Node* in_start = in;
8974 Node* ct_start = ct;
8975 Node* out_start = out;
8976 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8977 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8978 in_start = array_element_address(in, inOfs, T_BYTE);
8979 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8980 out_start = array_element_address(out, outOfs, T_BYTE);
8981 }
8982
8983 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8984 // (because of the predicated logic executed earlier).
8985 // so we cast it here safely.
8986 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8987 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8988 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8989 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8990 Node* state = load_field_from_object(ghash_object, "state", "[J");
8991
8992 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8993 return false;
8994 }
8995 // cast it to what we know it will be at runtime
8996 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8997 assert(tinst != nullptr, "GCTR obj is null");
8998 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8999 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9000 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9001 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9002 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9003 const TypeOopPtr* xtype = aklass->as_instance_type();
9004 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9005 aescrypt_object = _gvn.transform(aescrypt_object);
9006 // we need to get the start of the aescrypt_object's expanded key array
9007 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
9008 if (k_start == nullptr) return false;
9009 // similarly, get the start address of the r vector
9010 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9011 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9012 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9013
9014
9015 // Call the stub, passing params
9016 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9017 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9018 stubAddr, stubName, TypePtr::BOTTOM,
9019 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9020
9021 // return cipher length (int)
9022 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9023 set_result(retvalue);
9024
9025 return true;
9026 }
9027
9028 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9029 // Return node representing slow path of predicate check.
9030 // the pseudo code we want to emulate with this predicate is:
9031 // for encryption:
9032 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9033 // for decryption:
9034 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9035 // note cipher==plain is more conservative than the original java code but that's OK
9036 //
9037
9038 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9039 // The receiver was checked for null already.
9040 Node* objGCTR = argument(7);
9041 // Load embeddedCipher field of GCTR object.
9042 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9043 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9044
9045 // get AESCrypt klass for instanceOf check
9046 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9047 // will have same classloader as CipherBlockChaining object
9048 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9049 assert(tinst != nullptr, "GCTR obj is null");
9050 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9051
9052 // we want to do an instanceof comparison against the AESCrypt class
9053 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9054 if (!klass_AESCrypt->is_loaded()) {
9055 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9056 Node* ctrl = control();
9057 set_control(top()); // no regular fast path
9058 return ctrl;
9059 }
9060
9061 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9062 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9063 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9064 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9065 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9066
9067 return instof_false; // even if it is null
9068 }
9069
9070 //------------------------------get_state_from_digest_object-----------------------
9071 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9072 const char* state_type;
9073 switch (elem_type) {
9074 case T_BYTE: state_type = "[B"; break;
9075 case T_INT: state_type = "[I"; break;
9076 case T_LONG: state_type = "[J"; break;
9077 default: ShouldNotReachHere();
9078 }
9079 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9080 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9081 if (digest_state == nullptr) return (Node *) nullptr;
9082
9083 // now have the array, need to get the start address of the state array
9084 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9085 return state;
9086 }
9087
9088 //------------------------------get_block_size_from_sha3_object----------------------------------
9089 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9090 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9091 assert (block_size != nullptr, "sanity");
9092 return block_size;
9093 }
9094
9095 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9096 // Return node representing slow path of predicate check.
9097 // the pseudo code we want to emulate with this predicate is:
9098 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9099 //
9100 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9101 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9102 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9103 assert((uint)predicate < 5, "sanity");
9104
9105 // The receiver was checked for null already.
9106 Node* digestBaseObj = argument(0);
9107
9108 // get DigestBase klass for instanceOf check
9109 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9110 assert(tinst != nullptr, "digestBaseObj is null");
9111 assert(tinst->is_loaded(), "DigestBase is not loaded");
9112
9113 const char* klass_name = nullptr;
9114 switch (predicate) {
9115 case 0:
9116 if (UseMD5Intrinsics) {
9117 // we want to do an instanceof comparison against the MD5 class
9118 klass_name = "sun/security/provider/MD5";
9119 }
9120 break;
9121 case 1:
9122 if (UseSHA1Intrinsics) {
9123 // we want to do an instanceof comparison against the SHA class
9124 klass_name = "sun/security/provider/SHA";
9125 }
9126 break;
9127 case 2:
9128 if (UseSHA256Intrinsics) {
9129 // we want to do an instanceof comparison against the SHA2 class
9130 klass_name = "sun/security/provider/SHA2";
9131 }
9132 break;
9133 case 3:
9134 if (UseSHA512Intrinsics) {
9135 // we want to do an instanceof comparison against the SHA5 class
9136 klass_name = "sun/security/provider/SHA5";
9137 }
9138 break;
9139 case 4:
9140 if (UseSHA3Intrinsics) {
9141 // we want to do an instanceof comparison against the SHA3 class
9142 klass_name = "sun/security/provider/SHA3";
9143 }
9144 break;
9145 default:
9146 fatal("unknown SHA intrinsic predicate: %d", predicate);
9147 }
9148
9149 ciKlass* klass = nullptr;
9150 if (klass_name != nullptr) {
9151 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9152 }
9153 if ((klass == nullptr) || !klass->is_loaded()) {
9154 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9155 Node* ctrl = control();
9156 set_control(top()); // no intrinsic path
9157 return ctrl;
9158 }
9159 ciInstanceKlass* instklass = klass->as_instance_klass();
9160
9161 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9162 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9163 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9164 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9165
9166 return instof_false; // even if it is null
9167 }
9168
9169 //-------------inline_fma-----------------------------------
9170 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9171 Node *a = nullptr;
9172 Node *b = nullptr;
9173 Node *c = nullptr;
9174 Node* result = nullptr;
9175 switch (id) {
9176 case vmIntrinsics::_fmaD:
9177 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9178 // no receiver since it is static method
9179 a = argument(0);
9180 b = argument(2);
9181 c = argument(4);
9182 result = _gvn.transform(new FmaDNode(a, b, c));
9183 break;
9184 case vmIntrinsics::_fmaF:
9185 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9186 a = argument(0);
9187 b = argument(1);
9188 c = argument(2);
9189 result = _gvn.transform(new FmaFNode(a, b, c));
9190 break;
9191 default:
9192 fatal_unexpected_iid(id); break;
9193 }
9194 set_result(result);
9195 return true;
9196 }
9197
9198 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9199 // argument(0) is receiver
9200 Node* codePoint = argument(1);
9201 Node* n = nullptr;
9202
9203 switch (id) {
9204 case vmIntrinsics::_isDigit :
9205 n = new DigitNode(control(), codePoint);
9206 break;
9207 case vmIntrinsics::_isLowerCase :
9208 n = new LowerCaseNode(control(), codePoint);
9209 break;
9210 case vmIntrinsics::_isUpperCase :
9211 n = new UpperCaseNode(control(), codePoint);
9212 break;
9213 case vmIntrinsics::_isWhitespace :
9214 n = new WhitespaceNode(control(), codePoint);
9215 break;
9216 default:
9217 fatal_unexpected_iid(id);
9218 }
9219
9220 set_result(_gvn.transform(n));
9221 return true;
9222 }
9223
9224 bool LibraryCallKit::inline_profileBoolean() {
9225 Node* counts = argument(1);
9226 const TypeAryPtr* ary = nullptr;
9227 ciArray* aobj = nullptr;
9228 if (counts->is_Con()
9229 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9230 && (aobj = ary->const_oop()->as_array()) != nullptr
9231 && (aobj->length() == 2)) {
9232 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9233 jint false_cnt = aobj->element_value(0).as_int();
9234 jint true_cnt = aobj->element_value(1).as_int();
9235
9236 if (C->log() != nullptr) {
9237 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9238 false_cnt, true_cnt);
9239 }
9240
9241 if (false_cnt + true_cnt == 0) {
9242 // According to profile, never executed.
9243 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9244 Deoptimization::Action_reinterpret);
9245 return true;
9246 }
9247
9248 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9249 // is a number of each value occurrences.
9250 Node* result = argument(0);
9251 if (false_cnt == 0 || true_cnt == 0) {
9252 // According to profile, one value has been never seen.
9253 int expected_val = (false_cnt == 0) ? 1 : 0;
9254
9255 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9256 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9257
9258 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9259 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9260 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9261
9262 { // Slow path: uncommon trap for never seen value and then reexecute
9263 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9264 // the value has been seen at least once.
9265 PreserveJVMState pjvms(this);
9266 PreserveReexecuteState preexecs(this);
9267 jvms()->set_should_reexecute(true);
9268
9269 set_control(slow_path);
9270 set_i_o(i_o());
9271
9272 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9273 Deoptimization::Action_reinterpret);
9274 }
9275 // The guard for never seen value enables sharpening of the result and
9276 // returning a constant. It allows to eliminate branches on the same value
9277 // later on.
9278 set_control(fast_path);
9279 result = intcon(expected_val);
9280 }
9281 // Stop profiling.
9282 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9283 // By replacing method body with profile data (represented as ProfileBooleanNode
9284 // on IR level) we effectively disable profiling.
9285 // It enables full speed execution once optimized code is generated.
9286 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9287 C->record_for_igvn(profile);
9288 set_result(profile);
9289 return true;
9290 } else {
9291 // Continue profiling.
9292 // Profile data isn't available at the moment. So, execute method's bytecode version.
9293 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9294 // is compiled and counters aren't available since corresponding MethodHandle
9295 // isn't a compile-time constant.
9296 return false;
9297 }
9298 }
9299
9300 bool LibraryCallKit::inline_isCompileConstant() {
9301 Node* n = argument(0);
9302 set_result(n->is_Con() ? intcon(1) : intcon(0));
9303 return true;
9304 }
9305
9306 //------------------------------- inline_getObjectSize --------------------------------------
9307 //
9308 // Calculate the runtime size of the object/array.
9309 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9310 //
9311 bool LibraryCallKit::inline_getObjectSize() {
9312 Node* obj = argument(3);
9313 Node* klass_node = load_object_klass(obj);
9314
9315 jint layout_con = Klass::_lh_neutral_value;
9316 Node* layout_val = get_layout_helper(klass_node, layout_con);
9317 int layout_is_con = (layout_val == nullptr);
9318
9319 if (layout_is_con) {
9320 // Layout helper is constant, can figure out things at compile time.
9321
9322 if (Klass::layout_helper_is_instance(layout_con)) {
9323 // Instance case: layout_con contains the size itself.
9324 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9325 set_result(size);
9326 } else {
9327 // Array case: size is round(header + element_size*arraylength).
9328 // Since arraylength is different for every array instance, we have to
9329 // compute the whole thing at runtime.
9330
9331 Node* arr_length = load_array_length(obj);
9332
9333 int round_mask = MinObjAlignmentInBytes - 1;
9334 int hsize = Klass::layout_helper_header_size(layout_con);
9335 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9336
9337 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9338 round_mask = 0; // strength-reduce it if it goes away completely
9339 }
9340 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9341 Node* header_size = intcon(hsize + round_mask);
9342
9343 Node* lengthx = ConvI2X(arr_length);
9344 Node* headerx = ConvI2X(header_size);
9345
9346 Node* abody = lengthx;
9347 if (eshift != 0) {
9348 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9349 }
9350 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9351 if (round_mask != 0) {
9352 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9353 }
9354 size = ConvX2L(size);
9355 set_result(size);
9356 }
9357 } else {
9358 // Layout helper is not constant, need to test for array-ness at runtime.
9359
9360 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9361 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9362 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9363 record_for_igvn(result_reg);
9364
9365 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9366 if (array_ctl != nullptr) {
9367 // Array case: size is round(header + element_size*arraylength).
9368 // Since arraylength is different for every array instance, we have to
9369 // compute the whole thing at runtime.
9370
9371 PreserveJVMState pjvms(this);
9372 set_control(array_ctl);
9373 Node* arr_length = load_array_length(obj);
9374
9375 int round_mask = MinObjAlignmentInBytes - 1;
9376 Node* mask = intcon(round_mask);
9377
9378 Node* hss = intcon(Klass::_lh_header_size_shift);
9379 Node* hsm = intcon(Klass::_lh_header_size_mask);
9380 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9381 header_size = _gvn.transform(new AndINode(header_size, hsm));
9382 header_size = _gvn.transform(new AddINode(header_size, mask));
9383
9384 // There is no need to mask or shift this value.
9385 // The semantics of LShiftINode include an implicit mask to 0x1F.
9386 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9387 Node* elem_shift = layout_val;
9388
9389 Node* lengthx = ConvI2X(arr_length);
9390 Node* headerx = ConvI2X(header_size);
9391
9392 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9393 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9394 if (round_mask != 0) {
9395 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9396 }
9397 size = ConvX2L(size);
9398
9399 result_reg->init_req(_array_path, control());
9400 result_val->init_req(_array_path, size);
9401 }
9402
9403 if (!stopped()) {
9404 // Instance case: the layout helper gives us instance size almost directly,
9405 // but we need to mask out the _lh_instance_slow_path_bit.
9406 Node* size = ConvI2X(layout_val);
9407 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9408 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9409 size = _gvn.transform(new AndXNode(size, mask));
9410 size = ConvX2L(size);
9411
9412 result_reg->init_req(_instance_path, control());
9413 result_val->init_req(_instance_path, size);
9414 }
9415
9416 set_result(result_reg, result_val);
9417 }
9418
9419 return true;
9420 }
9421
9422 //------------------------------- inline_blackhole --------------------------------------
9423 //
9424 // Make sure all arguments to this node are alive.
9425 // This matches methods that were requested to be blackholed through compile commands.
9426 //
9427 bool LibraryCallKit::inline_blackhole() {
9428 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9429 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9430 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9431
9432 // Blackhole node pinches only the control, not memory. This allows
9433 // the blackhole to be pinned in the loop that computes blackholed
9434 // values, but have no other side effects, like breaking the optimizations
9435 // across the blackhole.
9436
9437 Node* bh = _gvn.transform(new BlackholeNode(control()));
9438 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9439
9440 // Bind call arguments as blackhole arguments to keep them alive
9441 uint nargs = callee()->arg_size();
9442 for (uint i = 0; i < nargs; i++) {
9443 bh->add_req(argument(i));
9444 }
9445
9446 return true;
9447 }
9448
9449 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9450 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9451 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9452 return nullptr; // box klass is not Float16
9453 }
9454
9455 // Null check; get notnull casted pointer
9456 Node* null_ctl = top();
9457 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9458 // If not_null_box is dead, only null-path is taken
9459 if (stopped()) {
9460 set_control(null_ctl);
9461 return nullptr;
9462 }
9463 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9464 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9465 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9466 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9467 }
9468
9469 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9470 PreserveReexecuteState preexecs(this);
9471 jvms()->set_should_reexecute(true);
9472
9473 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9474 Node* klass_node = makecon(klass_type);
9475 Node* box = new_instance(klass_node);
9476
9477 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9478 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9479
9480 Node* field_store = _gvn.transform(access_store_at(box,
9481 value_field,
9482 value_adr_type,
9483 value,
9484 TypeInt::SHORT,
9485 T_SHORT,
9486 IN_HEAP));
9487 set_memory(field_store, value_adr_type);
9488 return box;
9489 }
9490
9491 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9492 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9493 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9494 return false;
9495 }
9496
9497 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9498 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9499 return false;
9500 }
9501
9502 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9503 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9504 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9505 ciSymbols::short_signature(),
9506 false);
9507 assert(field != nullptr, "");
9508
9509 // Transformed nodes
9510 Node* fld1 = nullptr;
9511 Node* fld2 = nullptr;
9512 Node* fld3 = nullptr;
9513 switch(num_args) {
9514 case 3:
9515 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9516 if (fld3 == nullptr) {
9517 return false;
9518 }
9519 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9520 // fall-through
9521 case 2:
9522 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9523 if (fld2 == nullptr) {
9524 return false;
9525 }
9526 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9527 // fall-through
9528 case 1:
9529 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9530 if (fld1 == nullptr) {
9531 return false;
9532 }
9533 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9534 break;
9535 default: fatal("Unsupported number of arguments %d", num_args);
9536 }
9537
9538 Node* result = nullptr;
9539 switch (id) {
9540 // Unary operations
9541 case vmIntrinsics::_sqrt_float16:
9542 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9543 break;
9544 // Ternary operations
9545 case vmIntrinsics::_fma_float16:
9546 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9547 break;
9548 default:
9549 fatal_unexpected_iid(id);
9550 break;
9551 }
9552 result = _gvn.transform(new ReinterpretHF2SNode(result));
9553 set_result(box_fp16_value(float16_box_type, field, result));
9554 return true;
9555 }
9556