1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciArrayKlass.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciInstanceKlass.hpp"
29 #include "ci/ciUtilities.inline.hpp"
30 #include "ci/ciSymbols.hpp"
31 #include "classfile/vmIntrinsics.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compileLog.hpp"
34 #include "gc/shared/barrierSet.hpp"
35 #include "gc/shared/c2/barrierSetC2.hpp"
36 #include "jfr/support/jfrIntrinsics.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/accessDecorators.hpp"
39 #include "oops/klass.inline.hpp"
40 #include "oops/layoutKind.hpp"
41 #include "oops/objArrayKlass.hpp"
42 #include "opto/addnode.hpp"
43 #include "opto/arraycopynode.hpp"
44 #include "opto/c2compiler.hpp"
45 #include "opto/castnode.hpp"
46 #include "opto/cfgnode.hpp"
47 #include "opto/convertnode.hpp"
48 #include "opto/countbitsnode.hpp"
49 #include "opto/graphKit.hpp"
50 #include "opto/idealKit.hpp"
51 #include "opto/library_call.hpp"
52 #include "opto/inlinetypenode.hpp"
53 #include "opto/mathexactnode.hpp"
54 #include "opto/mulnode.hpp"
55 #include "opto/narrowptrnode.hpp"
56 #include "opto/opaquenode.hpp"
57 #include "opto/opcodes.hpp"
58 #include "opto/parse.hpp"
59 #include "opto/runtime.hpp"
60 #include "opto/rootnode.hpp"
61 #include "opto/subnode.hpp"
62 #include "opto/type.hpp"
63 #include "opto/vectornode.hpp"
64 #include "prims/jvmtiExport.hpp"
65 #include "prims/jvmtiThreadState.hpp"
66 #include "prims/unsafe.hpp"
67 #include "runtime/jniHandles.inline.hpp"
68 #include "runtime/objectMonitor.hpp"
69 #include "runtime/sharedRuntime.hpp"
70 #include "runtime/stubRoutines.hpp"
71 #include "utilities/globalDefinitions.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/powerOfTwo.hpp"
74
75 //---------------------------make_vm_intrinsic----------------------------
76 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
77 vmIntrinsicID id = m->intrinsic_id();
78 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
79
80 if (!m->is_loaded()) {
81 // Do not attempt to inline unloaded methods.
82 return nullptr;
83 }
84
85 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
86 bool is_available = false;
87
88 {
89 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
90 // the compiler must transition to '_thread_in_vm' state because both
91 // methods access VM-internal data.
92 VM_ENTRY_MARK;
93 methodHandle mh(THREAD, m->get_Method());
94 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
95 if (is_available && is_virtual) {
96 is_available = vmIntrinsics::does_virtual_dispatch(id);
97 }
98 }
99
100 if (is_available) {
101 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
102 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
103 return new LibraryIntrinsic(m, is_virtual,
104 vmIntrinsics::predicates_needed(id),
105 vmIntrinsics::does_virtual_dispatch(id),
106 id);
107 } else {
108 return nullptr;
109 }
110 }
111
112 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
113 LibraryCallKit kit(jvms, this);
114 Compile* C = kit.C;
115 int nodes = C->unique();
116 #ifndef PRODUCT
117 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
118 char buf[1000];
119 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
120 tty->print_cr("Intrinsic %s", str);
121 }
122 #endif
123 ciMethod* callee = kit.callee();
124 const int bci = kit.bci();
125 #ifdef ASSERT
126 Node* ctrl = kit.control();
127 #endif
128 // Try to inline the intrinsic.
129 if (callee->check_intrinsic_candidate() &&
130 kit.try_to_inline(_last_predicate)) {
131 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
132 : "(intrinsic)";
133 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
134 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
135 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
136 if (C->log()) {
137 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
138 vmIntrinsics::name_at(intrinsic_id()),
139 (is_virtual() ? " virtual='1'" : ""),
140 C->unique() - nodes);
141 }
142 // Push the result from the inlined method onto the stack.
143 kit.push_result();
144 return kit.transfer_exceptions_into_jvms();
145 }
146
147 // The intrinsic bailed out
148 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
149 if (jvms->has_method()) {
150 // Not a root compile.
151 const char* msg;
152 if (callee->intrinsic_candidate()) {
153 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
154 } else {
155 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
156 : "failed to inline (intrinsic), method not annotated";
157 }
158 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
159 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
160 } else {
161 // Root compile
162 ResourceMark rm;
163 stringStream msg_stream;
164 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
165 vmIntrinsics::name_at(intrinsic_id()),
166 is_virtual() ? " (virtual)" : "", bci);
167 const char *msg = msg_stream.freeze();
168 log_debug(jit, inlining)("%s", msg);
169 if (C->print_intrinsics() || C->print_inlining()) {
170 tty->print("%s", msg);
171 }
172 }
173 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
174
175 return nullptr;
176 }
177
178 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
179 LibraryCallKit kit(jvms, this);
180 Compile* C = kit.C;
181 int nodes = C->unique();
182 _last_predicate = predicate;
183 #ifndef PRODUCT
184 assert(is_predicated() && predicate < predicates_count(), "sanity");
185 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
186 char buf[1000];
187 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
188 tty->print_cr("Predicate for intrinsic %s", str);
189 }
190 #endif
191 ciMethod* callee = kit.callee();
192 const int bci = kit.bci();
193
194 Node* slow_ctl = kit.try_to_predicate(predicate);
195 if (!kit.failing()) {
196 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
197 : "(intrinsic, predicate)";
198 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
199 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
200
201 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
202 if (C->log()) {
203 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
204 vmIntrinsics::name_at(intrinsic_id()),
205 (is_virtual() ? " virtual='1'" : ""),
206 C->unique() - nodes);
207 }
208 return slow_ctl; // Could be null if the check folds.
209 }
210
211 // The intrinsic bailed out
212 if (jvms->has_method()) {
213 // Not a root compile.
214 const char* msg = "failed to generate predicate for intrinsic";
215 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
216 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
217 } else {
218 // Root compile
219 ResourceMark rm;
220 stringStream msg_stream;
221 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
222 vmIntrinsics::name_at(intrinsic_id()),
223 is_virtual() ? " (virtual)" : "", bci);
224 const char *msg = msg_stream.freeze();
225 log_debug(jit, inlining)("%s", msg);
226 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dtanh:
259 case vmIntrinsics::_dabs:
260 case vmIntrinsics::_fabs:
261 case vmIntrinsics::_iabs:
262 case vmIntrinsics::_labs:
263 case vmIntrinsics::_datan2:
264 case vmIntrinsics::_dsqrt:
265 case vmIntrinsics::_dsqrt_strict:
266 case vmIntrinsics::_dexp:
267 case vmIntrinsics::_dlog:
268 case vmIntrinsics::_dlog10:
269 case vmIntrinsics::_dpow:
270 case vmIntrinsics::_dcopySign:
271 case vmIntrinsics::_fcopySign:
272 case vmIntrinsics::_dsignum:
273 case vmIntrinsics::_roundF:
274 case vmIntrinsics::_roundD:
275 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
276
277 case vmIntrinsics::_notify:
278 case vmIntrinsics::_notifyAll:
279 return inline_notify(intrinsic_id());
280
281 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
282 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
283 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
284 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
285 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
286 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
287 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
288 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
289 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
290 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
291 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
292 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
293 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
294 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
295
296 case vmIntrinsics::_arraycopy: return inline_arraycopy();
297
298 case vmIntrinsics::_arraySort: return inline_array_sort();
299 case vmIntrinsics::_arrayPartition: return inline_array_partition();
300
301 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
302 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
303 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
304 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
305
306 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
307 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
308 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
309 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
310 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
311 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
312 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
313 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
314
315 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
316
317 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
318
319 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
320 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
321 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
322 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
323
324 case vmIntrinsics::_compressStringC:
325 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
326 case vmIntrinsics::_inflateStringC:
327 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
328
329 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
330 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
331 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
332 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
333 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
334 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
335 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
336 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
337 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
338 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
339 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
340 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
341
342 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
343 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
344 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
345 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
346 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
347 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
348 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
349 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
350 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
351 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
352
353 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
354 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
355 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
356 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
357 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
358 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
359 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
360 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
361 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
362
363 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
364 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
365 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
366 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
367 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
368 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
369 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
370 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
371 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
372
373 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
379 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
380 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
381 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
382
383 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
384 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
385 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
386 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
387 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
388 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
389 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
390 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
391 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
392
393 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
394 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
395 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
396 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
397 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
398 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
399 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
400 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
401 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
402
403 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
404 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
405 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
406 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
407 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
408 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
409 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
410 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
411 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
412
413 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
414 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
415 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
416 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
417 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
418 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
419 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
420 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
421 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
422
423 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
424 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
425
426 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
427 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
428 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
429 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
431
432 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
444 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
445 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
446 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
447 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
448 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
449 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
450 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
451 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
452
453 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
462 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
463 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
464 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
465 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
466 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
467 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
468
469 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
470 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
471 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
472 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
473
474 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
475 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
476 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
477 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
479
480 case vmIntrinsics::_loadFence:
481 case vmIntrinsics::_storeFence:
482 case vmIntrinsics::_storeStoreFence:
483 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
484
485 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
486
487 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
488 case vmIntrinsics::_currentThread: return inline_native_currentThread();
489 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
490
491 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
492 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
493
494 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
495 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
496
497 #if INCLUDE_JVMTI
498 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
499 "notifyJvmtiStart", true, false);
500 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
501 "notifyJvmtiEnd", false, true);
502 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
503 "notifyJvmtiMount", false, false);
504 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
505 "notifyJvmtiUnmount", false, false);
506 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
507 #endif
508
509 #ifdef JFR_HAVE_INTRINSICS
510 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
511 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
512 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
513 #endif
514 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
515 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
516 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
517 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
518 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
519 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
520 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
521 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
522 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
523 case vmIntrinsics::_getLength: return inline_native_getLength();
524 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
525 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
526 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
527 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
528 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
529 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
530 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
531
532 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
533 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
534 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
535 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
536 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
537
538 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
539
540 case vmIntrinsics::_isInstance:
541 case vmIntrinsics::_isHidden:
542 case vmIntrinsics::_getSuperclass:
543 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
544
545 case vmIntrinsics::_floatToRawIntBits:
546 case vmIntrinsics::_floatToIntBits:
547 case vmIntrinsics::_intBitsToFloat:
548 case vmIntrinsics::_doubleToRawLongBits:
549 case vmIntrinsics::_doubleToLongBits:
550 case vmIntrinsics::_longBitsToDouble:
551 case vmIntrinsics::_floatToFloat16:
552 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
553 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
554 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
555 case vmIntrinsics::_floatIsFinite:
556 case vmIntrinsics::_floatIsInfinite:
557 case vmIntrinsics::_doubleIsFinite:
558 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
559
560 case vmIntrinsics::_numberOfLeadingZeros_i:
561 case vmIntrinsics::_numberOfLeadingZeros_l:
562 case vmIntrinsics::_numberOfTrailingZeros_i:
563 case vmIntrinsics::_numberOfTrailingZeros_l:
564 case vmIntrinsics::_bitCount_i:
565 case vmIntrinsics::_bitCount_l:
566 case vmIntrinsics::_reverse_i:
567 case vmIntrinsics::_reverse_l:
568 case vmIntrinsics::_reverseBytes_i:
569 case vmIntrinsics::_reverseBytes_l:
570 case vmIntrinsics::_reverseBytes_s:
571 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
572
573 case vmIntrinsics::_compress_i:
574 case vmIntrinsics::_compress_l:
575 case vmIntrinsics::_expand_i:
576 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
577
578 case vmIntrinsics::_compareUnsigned_i:
579 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
580
581 case vmIntrinsics::_divideUnsigned_i:
582 case vmIntrinsics::_divideUnsigned_l:
583 case vmIntrinsics::_remainderUnsigned_i:
584 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
585
586 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
587
588 case vmIntrinsics::_Reference_get: return inline_reference_get();
589 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
590 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
591 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
592 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
593
594 case vmIntrinsics::_Class_cast: return inline_Class_cast();
595
596 case vmIntrinsics::_aescrypt_encryptBlock:
597 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
598
599 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
600 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
601 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
602
603 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
604 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
605 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
606
607 case vmIntrinsics::_counterMode_AESCrypt:
608 return inline_counterMode_AESCrypt(intrinsic_id());
609
610 case vmIntrinsics::_galoisCounterMode_AESCrypt:
611 return inline_galoisCounterMode_AESCrypt();
612
613 case vmIntrinsics::_md5_implCompress:
614 case vmIntrinsics::_sha_implCompress:
615 case vmIntrinsics::_sha2_implCompress:
616 case vmIntrinsics::_sha5_implCompress:
617 case vmIntrinsics::_sha3_implCompress:
618 return inline_digestBase_implCompress(intrinsic_id());
619 case vmIntrinsics::_double_keccak:
620 return inline_double_keccak();
621
622 case vmIntrinsics::_digestBase_implCompressMB:
623 return inline_digestBase_implCompressMB(predicate);
624
625 case vmIntrinsics::_multiplyToLen:
626 return inline_multiplyToLen();
627
628 case vmIntrinsics::_squareToLen:
629 return inline_squareToLen();
630
631 case vmIntrinsics::_mulAdd:
632 return inline_mulAdd();
633
634 case vmIntrinsics::_montgomeryMultiply:
635 return inline_montgomeryMultiply();
636 case vmIntrinsics::_montgomerySquare:
637 return inline_montgomerySquare();
638
639 case vmIntrinsics::_bigIntegerRightShiftWorker:
640 return inline_bigIntegerShift(true);
641 case vmIntrinsics::_bigIntegerLeftShiftWorker:
642 return inline_bigIntegerShift(false);
643
644 case vmIntrinsics::_vectorizedMismatch:
645 return inline_vectorizedMismatch();
646
647 case vmIntrinsics::_ghash_processBlocks:
648 return inline_ghash_processBlocks();
649 case vmIntrinsics::_chacha20Block:
650 return inline_chacha20Block();
651 case vmIntrinsics::_kyberNtt:
652 return inline_kyberNtt();
653 case vmIntrinsics::_kyberInverseNtt:
654 return inline_kyberInverseNtt();
655 case vmIntrinsics::_kyberNttMult:
656 return inline_kyberNttMult();
657 case vmIntrinsics::_kyberAddPoly_2:
658 return inline_kyberAddPoly_2();
659 case vmIntrinsics::_kyberAddPoly_3:
660 return inline_kyberAddPoly_3();
661 case vmIntrinsics::_kyber12To16:
662 return inline_kyber12To16();
663 case vmIntrinsics::_kyberBarrettReduce:
664 return inline_kyberBarrettReduce();
665 case vmIntrinsics::_dilithiumAlmostNtt:
666 return inline_dilithiumAlmostNtt();
667 case vmIntrinsics::_dilithiumAlmostInverseNtt:
668 return inline_dilithiumAlmostInverseNtt();
669 case vmIntrinsics::_dilithiumNttMult:
670 return inline_dilithiumNttMult();
671 case vmIntrinsics::_dilithiumMontMulByConstant:
672 return inline_dilithiumMontMulByConstant();
673 case vmIntrinsics::_dilithiumDecomposePoly:
674 return inline_dilithiumDecomposePoly();
675 case vmIntrinsics::_base64_encodeBlock:
676 return inline_base64_encodeBlock();
677 case vmIntrinsics::_base64_decodeBlock:
678 return inline_base64_decodeBlock();
679 case vmIntrinsics::_poly1305_processBlocks:
680 return inline_poly1305_processBlocks();
681 case vmIntrinsics::_intpoly_montgomeryMult_P256:
682 return inline_intpoly_montgomeryMult_P256();
683 case vmIntrinsics::_intpoly_assign:
684 return inline_intpoly_assign();
685 case vmIntrinsics::_encodeISOArray:
686 case vmIntrinsics::_encodeByteISOArray:
687 return inline_encodeISOArray(false);
688 case vmIntrinsics::_encodeAsciiArray:
689 return inline_encodeISOArray(true);
690
691 case vmIntrinsics::_updateCRC32:
692 return inline_updateCRC32();
693 case vmIntrinsics::_updateBytesCRC32:
694 return inline_updateBytesCRC32();
695 case vmIntrinsics::_updateByteBufferCRC32:
696 return inline_updateByteBufferCRC32();
697
698 case vmIntrinsics::_updateBytesCRC32C:
699 return inline_updateBytesCRC32C();
700 case vmIntrinsics::_updateDirectByteBufferCRC32C:
701 return inline_updateDirectByteBufferCRC32C();
702
703 case vmIntrinsics::_updateBytesAdler32:
704 return inline_updateBytesAdler32();
705 case vmIntrinsics::_updateByteBufferAdler32:
706 return inline_updateByteBufferAdler32();
707
708 case vmIntrinsics::_profileBoolean:
709 return inline_profileBoolean();
710 case vmIntrinsics::_isCompileConstant:
711 return inline_isCompileConstant();
712
713 case vmIntrinsics::_countPositives:
714 return inline_countPositives();
715
716 case vmIntrinsics::_fmaD:
717 case vmIntrinsics::_fmaF:
718 return inline_fma(intrinsic_id());
719
720 case vmIntrinsics::_isDigit:
721 case vmIntrinsics::_isLowerCase:
722 case vmIntrinsics::_isUpperCase:
723 case vmIntrinsics::_isWhitespace:
724 return inline_character_compare(intrinsic_id());
725
726 case vmIntrinsics::_min:
727 case vmIntrinsics::_max:
728 case vmIntrinsics::_min_strict:
729 case vmIntrinsics::_max_strict:
730 case vmIntrinsics::_minL:
731 case vmIntrinsics::_maxL:
732 case vmIntrinsics::_minF:
733 case vmIntrinsics::_maxF:
734 case vmIntrinsics::_minD:
735 case vmIntrinsics::_maxD:
736 case vmIntrinsics::_minF_strict:
737 case vmIntrinsics::_maxF_strict:
738 case vmIntrinsics::_minD_strict:
739 case vmIntrinsics::_maxD_strict:
740 return inline_min_max(intrinsic_id());
741
742 case vmIntrinsics::_VectorUnaryOp:
743 return inline_vector_nary_operation(1);
744 case vmIntrinsics::_VectorBinaryOp:
745 return inline_vector_nary_operation(2);
746 case vmIntrinsics::_VectorUnaryLibOp:
747 return inline_vector_call(1);
748 case vmIntrinsics::_VectorBinaryLibOp:
749 return inline_vector_call(2);
750 case vmIntrinsics::_VectorTernaryOp:
751 return inline_vector_nary_operation(3);
752 case vmIntrinsics::_VectorFromBitsCoerced:
753 return inline_vector_frombits_coerced();
754 case vmIntrinsics::_VectorMaskOp:
755 return inline_vector_mask_operation();
756 case vmIntrinsics::_VectorLoadOp:
757 return inline_vector_mem_operation(/*is_store=*/false);
758 case vmIntrinsics::_VectorLoadMaskedOp:
759 return inline_vector_mem_masked_operation(/*is_store*/false);
760 case vmIntrinsics::_VectorStoreOp:
761 return inline_vector_mem_operation(/*is_store=*/true);
762 case vmIntrinsics::_VectorStoreMaskedOp:
763 return inline_vector_mem_masked_operation(/*is_store=*/true);
764 case vmIntrinsics::_VectorGatherOp:
765 return inline_vector_gather_scatter(/*is_scatter*/ false);
766 case vmIntrinsics::_VectorScatterOp:
767 return inline_vector_gather_scatter(/*is_scatter*/ true);
768 case vmIntrinsics::_VectorReductionCoerced:
769 return inline_vector_reduction();
770 case vmIntrinsics::_VectorTest:
771 return inline_vector_test();
772 case vmIntrinsics::_VectorBlend:
773 return inline_vector_blend();
774 case vmIntrinsics::_VectorRearrange:
775 return inline_vector_rearrange();
776 case vmIntrinsics::_VectorSelectFrom:
777 return inline_vector_select_from();
778 case vmIntrinsics::_VectorCompare:
779 return inline_vector_compare();
780 case vmIntrinsics::_VectorBroadcastInt:
781 return inline_vector_broadcast_int();
782 case vmIntrinsics::_VectorConvert:
783 return inline_vector_convert();
784 case vmIntrinsics::_VectorInsert:
785 return inline_vector_insert();
786 case vmIntrinsics::_VectorExtract:
787 return inline_vector_extract();
788 case vmIntrinsics::_VectorCompressExpand:
789 return inline_vector_compress_expand();
790 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
791 return inline_vector_select_from_two_vectors();
792 case vmIntrinsics::_IndexVector:
793 return inline_index_vector();
794 case vmIntrinsics::_IndexPartiallyInUpperRange:
795 return inline_index_partially_in_upper_range();
796
797 case vmIntrinsics::_getObjectSize:
798 return inline_getObjectSize();
799
800 case vmIntrinsics::_blackhole:
801 return inline_blackhole();
802
803 default:
804 // If you get here, it may be that someone has added a new intrinsic
805 // to the list in vmIntrinsics.hpp without implementing it here.
806 #ifndef PRODUCT
807 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
808 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
809 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
810 }
811 #endif
812 return false;
813 }
814 }
815
816 Node* LibraryCallKit::try_to_predicate(int predicate) {
817 if (!jvms()->has_method()) {
818 // Root JVMState has a null method.
819 assert(map()->memory()->Opcode() == Op_Parm, "");
820 // Insert the memory aliasing node
821 set_all_memory(reset_memory());
822 }
823 assert(merged_memory(), "");
824
825 switch (intrinsic_id()) {
826 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
827 return inline_cipherBlockChaining_AESCrypt_predicate(false);
828 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
829 return inline_cipherBlockChaining_AESCrypt_predicate(true);
830 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
831 return inline_electronicCodeBook_AESCrypt_predicate(false);
832 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
833 return inline_electronicCodeBook_AESCrypt_predicate(true);
834 case vmIntrinsics::_counterMode_AESCrypt:
835 return inline_counterMode_AESCrypt_predicate();
836 case vmIntrinsics::_digestBase_implCompressMB:
837 return inline_digestBase_implCompressMB_predicate(predicate);
838 case vmIntrinsics::_galoisCounterMode_AESCrypt:
839 return inline_galoisCounterMode_AESCrypt_predicate();
840
841 default:
842 // If you get here, it may be that someone has added a new intrinsic
843 // to the list in vmIntrinsics.hpp without implementing it here.
844 #ifndef PRODUCT
845 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
846 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
847 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
848 }
849 #endif
850 Node* slow_ctl = control();
851 set_control(top()); // No fast path intrinsic
852 return slow_ctl;
853 }
854 }
855
856 //------------------------------set_result-------------------------------
857 // Helper function for finishing intrinsics.
858 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
859 record_for_igvn(region);
860 set_control(_gvn.transform(region));
861 set_result( _gvn.transform(value));
862 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
863 }
864
865 //------------------------------generate_guard---------------------------
866 // Helper function for generating guarded fast-slow graph structures.
867 // The given 'test', if true, guards a slow path. If the test fails
868 // then a fast path can be taken. (We generally hope it fails.)
869 // In all cases, GraphKit::control() is updated to the fast path.
870 // The returned value represents the control for the slow path.
871 // The return value is never 'top'; it is either a valid control
872 // or null if it is obvious that the slow path can never be taken.
873 // Also, if region and the slow control are not null, the slow edge
874 // is appended to the region.
875 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
876 if (stopped()) {
877 // Already short circuited.
878 return nullptr;
879 }
880
881 // Build an if node and its projections.
882 // If test is true we take the slow path, which we assume is uncommon.
883 if (_gvn.type(test) == TypeInt::ZERO) {
884 // The slow branch is never taken. No need to build this guard.
885 return nullptr;
886 }
887
888 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
889
890 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
891 if (if_slow == top()) {
892 // The slow branch is never taken. No need to build this guard.
893 return nullptr;
894 }
895
896 if (region != nullptr)
897 region->add_req(if_slow);
898
899 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
900 set_control(if_fast);
901
902 return if_slow;
903 }
904
905 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
906 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
907 }
908 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
909 return generate_guard(test, region, PROB_FAIR);
910 }
911
912 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
913 Node* *pos_index) {
914 if (stopped())
915 return nullptr; // already stopped
916 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
917 return nullptr; // index is already adequately typed
918 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
919 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
920 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
921 if (is_neg != nullptr && pos_index != nullptr) {
922 // Emulate effect of Parse::adjust_map_after_if.
923 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
924 (*pos_index) = _gvn.transform(ccast);
925 }
926 return is_neg;
927 }
928
929 // Make sure that 'position' is a valid limit index, in [0..length].
930 // There are two equivalent plans for checking this:
931 // A. (offset + copyLength) unsigned<= arrayLength
932 // B. offset <= (arrayLength - copyLength)
933 // We require that all of the values above, except for the sum and
934 // difference, are already known to be non-negative.
935 // Plan A is robust in the face of overflow, if offset and copyLength
936 // are both hugely positive.
937 //
938 // Plan B is less direct and intuitive, but it does not overflow at
939 // all, since the difference of two non-negatives is always
940 // representable. Whenever Java methods must perform the equivalent
941 // check they generally use Plan B instead of Plan A.
942 // For the moment we use Plan A.
943 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
944 Node* subseq_length,
945 Node* array_length,
946 RegionNode* region) {
947 if (stopped())
948 return nullptr; // already stopped
949 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
950 if (zero_offset && subseq_length->eqv_uncast(array_length))
951 return nullptr; // common case of whole-array copy
952 Node* last = subseq_length;
953 if (!zero_offset) // last += offset
954 last = _gvn.transform(new AddINode(last, offset));
955 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
956 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
957 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
958 return is_over;
959 }
960
961 // Emit range checks for the given String.value byte array
962 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
963 if (stopped()) {
964 return; // already stopped
965 }
966 RegionNode* bailout = new RegionNode(1);
967 record_for_igvn(bailout);
968 if (char_count) {
969 // Convert char count to byte count
970 count = _gvn.transform(new LShiftINode(count, intcon(1)));
971 }
972
973 // Offset and count must not be negative
974 generate_negative_guard(offset, bailout);
975 generate_negative_guard(count, bailout);
976 // Offset + count must not exceed length of array
977 generate_limit_guard(offset, count, load_array_length(array), bailout);
978
979 if (bailout->req() > 1) {
980 PreserveJVMState pjvms(this);
981 set_control(_gvn.transform(bailout));
982 uncommon_trap(Deoptimization::Reason_intrinsic,
983 Deoptimization::Action_maybe_recompile);
984 }
985 }
986
987 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
988 bool is_immutable) {
989 ciKlass* thread_klass = env()->Thread_klass();
990 const Type* thread_type
991 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
992
993 Node* thread = _gvn.transform(new ThreadLocalNode());
994 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
995 tls_output = thread;
996
997 Node* thread_obj_handle
998 = (is_immutable
999 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1000 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1001 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1002 thread_obj_handle = _gvn.transform(thread_obj_handle);
1003
1004 DecoratorSet decorators = IN_NATIVE;
1005 if (is_immutable) {
1006 decorators |= C2_IMMUTABLE_MEMORY;
1007 }
1008 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1009 }
1010
1011 //--------------------------generate_current_thread--------------------
1012 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1013 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1014 /*is_immutable*/false);
1015 }
1016
1017 //--------------------------generate_virtual_thread--------------------
1018 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1019 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1020 !C->method()->changes_current_thread());
1021 }
1022
1023 //------------------------------make_string_method_node------------------------
1024 // Helper method for String intrinsic functions. This version is called with
1025 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1026 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1027 // containing the lengths of str1 and str2.
1028 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1029 Node* result = nullptr;
1030 switch (opcode) {
1031 case Op_StrIndexOf:
1032 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1033 str1_start, cnt1, str2_start, cnt2, ae);
1034 break;
1035 case Op_StrComp:
1036 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1037 str1_start, cnt1, str2_start, cnt2, ae);
1038 break;
1039 case Op_StrEquals:
1040 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1041 // Use the constant length if there is one because optimized match rule may exist.
1042 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1043 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1044 break;
1045 default:
1046 ShouldNotReachHere();
1047 return nullptr;
1048 }
1049
1050 // All these intrinsics have checks.
1051 C->set_has_split_ifs(true); // Has chance for split-if optimization
1052 clear_upper_avx();
1053
1054 return _gvn.transform(result);
1055 }
1056
1057 //------------------------------inline_string_compareTo------------------------
1058 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1059 Node* arg1 = argument(0);
1060 Node* arg2 = argument(1);
1061
1062 arg1 = must_be_not_null(arg1, true);
1063 arg2 = must_be_not_null(arg2, true);
1064
1065 // Get start addr and length of first argument
1066 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1067 Node* arg1_cnt = load_array_length(arg1);
1068
1069 // Get start addr and length of second argument
1070 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1071 Node* arg2_cnt = load_array_length(arg2);
1072
1073 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1074 set_result(result);
1075 return true;
1076 }
1077
1078 //------------------------------inline_string_equals------------------------
1079 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1080 Node* arg1 = argument(0);
1081 Node* arg2 = argument(1);
1082
1083 // paths (plus control) merge
1084 RegionNode* region = new RegionNode(3);
1085 Node* phi = new PhiNode(region, TypeInt::BOOL);
1086
1087 if (!stopped()) {
1088
1089 arg1 = must_be_not_null(arg1, true);
1090 arg2 = must_be_not_null(arg2, true);
1091
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 // Check for arg1_cnt != arg2_cnt
1101 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1102 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1103 Node* if_ne = generate_slow_guard(bol, nullptr);
1104 if (if_ne != nullptr) {
1105 phi->init_req(2, intcon(0));
1106 region->init_req(2, if_ne);
1107 }
1108
1109 // Check for count == 0 is done by assembler code for StrEquals.
1110
1111 if (!stopped()) {
1112 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1113 phi->init_req(1, equals);
1114 region->init_req(1, control());
1115 }
1116 }
1117
1118 // post merge
1119 set_control(_gvn.transform(region));
1120 record_for_igvn(region);
1121
1122 set_result(_gvn.transform(phi));
1123 return true;
1124 }
1125
1126 //------------------------------inline_array_equals----------------------------
1127 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1128 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1129 Node* arg1 = argument(0);
1130 Node* arg2 = argument(1);
1131
1132 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1133 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1134 clear_upper_avx();
1135
1136 return true;
1137 }
1138
1139
1140 //------------------------------inline_countPositives------------------------------
1141 bool LibraryCallKit::inline_countPositives() {
1142 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1143 return false;
1144 }
1145
1146 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1147 // no receiver since it is static method
1148 Node* ba = argument(0);
1149 Node* offset = argument(1);
1150 Node* len = argument(2);
1151
1152 ba = must_be_not_null(ba, true);
1153
1154 // Range checks
1155 generate_string_range_check(ba, offset, len, false);
1156 if (stopped()) {
1157 return true;
1158 }
1159 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1160 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1161 set_result(_gvn.transform(result));
1162 clear_upper_avx();
1163 return true;
1164 }
1165
1166 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1167 Node* index = argument(0);
1168 Node* length = bt == T_INT ? argument(1) : argument(2);
1169 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1170 return false;
1171 }
1172
1173 // check that length is positive
1174 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1175 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1176
1177 {
1178 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1179 uncommon_trap(Deoptimization::Reason_intrinsic,
1180 Deoptimization::Action_make_not_entrant);
1181 }
1182
1183 if (stopped()) {
1184 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1185 return true;
1186 }
1187
1188 // length is now known positive, add a cast node to make this explicit
1189 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1190 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1191 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1192 ConstraintCastNode::RegularDependency, bt);
1193 casted_length = _gvn.transform(casted_length);
1194 replace_in_map(length, casted_length);
1195 length = casted_length;
1196
1197 // Use an unsigned comparison for the range check itself
1198 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1199 BoolTest::mask btest = BoolTest::lt;
1200 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1201 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1202 _gvn.set_type(rc, rc->Value(&_gvn));
1203 if (!rc_bool->is_Con()) {
1204 record_for_igvn(rc);
1205 }
1206 set_control(_gvn.transform(new IfTrueNode(rc)));
1207 {
1208 PreserveJVMState pjvms(this);
1209 set_control(_gvn.transform(new IfFalseNode(rc)));
1210 uncommon_trap(Deoptimization::Reason_range_check,
1211 Deoptimization::Action_make_not_entrant);
1212 }
1213
1214 if (stopped()) {
1215 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1216 return true;
1217 }
1218
1219 // index is now known to be >= 0 and < length, cast it
1220 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1221 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1222 ConstraintCastNode::RegularDependency, bt);
1223 result = _gvn.transform(result);
1224 set_result(result);
1225 replace_in_map(index, result);
1226 return true;
1227 }
1228
1229 //------------------------------inline_string_indexOf------------------------
1230 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1231 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1232 return false;
1233 }
1234 Node* src = argument(0);
1235 Node* tgt = argument(1);
1236
1237 // Make the merge point
1238 RegionNode* result_rgn = new RegionNode(4);
1239 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1240
1241 src = must_be_not_null(src, true);
1242 tgt = must_be_not_null(tgt, true);
1243
1244 // Get start addr and length of source string
1245 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1246 Node* src_count = load_array_length(src);
1247
1248 // Get start addr and length of substring
1249 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1250 Node* tgt_count = load_array_length(tgt);
1251
1252 Node* result = nullptr;
1253 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1254
1255 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1256 // Divide src size by 2 if String is UTF16 encoded
1257 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1258 }
1259 if (ae == StrIntrinsicNode::UU) {
1260 // Divide substring size by 2 if String is UTF16 encoded
1261 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1262 }
1263
1264 if (call_opt_stub) {
1265 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1266 StubRoutines::_string_indexof_array[ae],
1267 "stringIndexOf", TypePtr::BOTTOM, src_start,
1268 src_count, tgt_start, tgt_count);
1269 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1270 } else {
1271 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1272 result_rgn, result_phi, ae);
1273 }
1274 if (result != nullptr) {
1275 result_phi->init_req(3, result);
1276 result_rgn->init_req(3, control());
1277 }
1278 set_control(_gvn.transform(result_rgn));
1279 record_for_igvn(result_rgn);
1280 set_result(_gvn.transform(result_phi));
1281
1282 return true;
1283 }
1284
1285 //-----------------------------inline_string_indexOfI-----------------------
1286 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1287 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1288 return false;
1289 }
1290 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1291 return false;
1292 }
1293
1294 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1295 Node* src = argument(0); // byte[]
1296 Node* src_count = argument(1); // char count
1297 Node* tgt = argument(2); // byte[]
1298 Node* tgt_count = argument(3); // char count
1299 Node* from_index = argument(4); // char index
1300
1301 src = must_be_not_null(src, true);
1302 tgt = must_be_not_null(tgt, true);
1303
1304 // Multiply byte array index by 2 if String is UTF16 encoded
1305 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1306 src_count = _gvn.transform(new SubINode(src_count, from_index));
1307 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1308 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1309
1310 // Range checks
1311 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1312 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1313 if (stopped()) {
1314 return true;
1315 }
1316
1317 RegionNode* region = new RegionNode(5);
1318 Node* phi = new PhiNode(region, TypeInt::INT);
1319 Node* result = nullptr;
1320
1321 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1322
1323 if (call_opt_stub) {
1324 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1325 StubRoutines::_string_indexof_array[ae],
1326 "stringIndexOf", TypePtr::BOTTOM, src_start,
1327 src_count, tgt_start, tgt_count);
1328 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1329 } else {
1330 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1331 region, phi, ae);
1332 }
1333 if (result != nullptr) {
1334 // The result is index relative to from_index if substring was found, -1 otherwise.
1335 // Generate code which will fold into cmove.
1336 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1337 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1338
1339 Node* if_lt = generate_slow_guard(bol, nullptr);
1340 if (if_lt != nullptr) {
1341 // result == -1
1342 phi->init_req(3, result);
1343 region->init_req(3, if_lt);
1344 }
1345 if (!stopped()) {
1346 result = _gvn.transform(new AddINode(result, from_index));
1347 phi->init_req(4, result);
1348 region->init_req(4, control());
1349 }
1350 }
1351
1352 set_control(_gvn.transform(region));
1353 record_for_igvn(region);
1354 set_result(_gvn.transform(phi));
1355 clear_upper_avx();
1356
1357 return true;
1358 }
1359
1360 // Create StrIndexOfNode with fast path checks
1361 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1362 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1363 // Check for substr count > string count
1364 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1365 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1366 Node* if_gt = generate_slow_guard(bol, nullptr);
1367 if (if_gt != nullptr) {
1368 phi->init_req(1, intcon(-1));
1369 region->init_req(1, if_gt);
1370 }
1371 if (!stopped()) {
1372 // Check for substr count == 0
1373 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1374 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1375 Node* if_zero = generate_slow_guard(bol, nullptr);
1376 if (if_zero != nullptr) {
1377 phi->init_req(2, intcon(0));
1378 region->init_req(2, if_zero);
1379 }
1380 }
1381 if (!stopped()) {
1382 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1383 }
1384 return nullptr;
1385 }
1386
1387 //-----------------------------inline_string_indexOfChar-----------------------
1388 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1389 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1390 return false;
1391 }
1392 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1393 return false;
1394 }
1395 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1396 Node* src = argument(0); // byte[]
1397 Node* int_ch = argument(1);
1398 Node* from_index = argument(2);
1399 Node* max = argument(3);
1400
1401 src = must_be_not_null(src, true);
1402
1403 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1404 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1405 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1406
1407 // Range checks
1408 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1409
1410 // Check for int_ch >= 0
1411 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1412 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1413 {
1414 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1415 uncommon_trap(Deoptimization::Reason_intrinsic,
1416 Deoptimization::Action_maybe_recompile);
1417 }
1418 if (stopped()) {
1419 return true;
1420 }
1421
1422 RegionNode* region = new RegionNode(3);
1423 Node* phi = new PhiNode(region, TypeInt::INT);
1424
1425 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1426 C->set_has_split_ifs(true); // Has chance for split-if optimization
1427 _gvn.transform(result);
1428
1429 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1430 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1431
1432 Node* if_lt = generate_slow_guard(bol, nullptr);
1433 if (if_lt != nullptr) {
1434 // result == -1
1435 phi->init_req(2, result);
1436 region->init_req(2, if_lt);
1437 }
1438 if (!stopped()) {
1439 result = _gvn.transform(new AddINode(result, from_index));
1440 phi->init_req(1, result);
1441 region->init_req(1, control());
1442 }
1443 set_control(_gvn.transform(region));
1444 record_for_igvn(region);
1445 set_result(_gvn.transform(phi));
1446 clear_upper_avx();
1447
1448 return true;
1449 }
1450 //---------------------------inline_string_copy---------------------
1451 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1452 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1453 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1454 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1455 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1456 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1457 bool LibraryCallKit::inline_string_copy(bool compress) {
1458 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1459 return false;
1460 }
1461 int nargs = 5; // 2 oops, 3 ints
1462 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1463
1464 Node* src = argument(0);
1465 Node* src_offset = argument(1);
1466 Node* dst = argument(2);
1467 Node* dst_offset = argument(3);
1468 Node* length = argument(4);
1469
1470 // Check for allocation before we add nodes that would confuse
1471 // tightly_coupled_allocation()
1472 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1473
1474 // Figure out the size and type of the elements we will be copying.
1475 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1476 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1477 if (src_type == nullptr || dst_type == nullptr) {
1478 return false;
1479 }
1480 BasicType src_elem = src_type->elem()->array_element_basic_type();
1481 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1482 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1483 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1484 "Unsupported array types for inline_string_copy");
1485
1486 src = must_be_not_null(src, true);
1487 dst = must_be_not_null(dst, true);
1488
1489 // Convert char[] offsets to byte[] offsets
1490 bool convert_src = (compress && src_elem == T_BYTE);
1491 bool convert_dst = (!compress && dst_elem == T_BYTE);
1492 if (convert_src) {
1493 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1494 } else if (convert_dst) {
1495 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1496 }
1497
1498 // Range checks
1499 generate_string_range_check(src, src_offset, length, convert_src);
1500 generate_string_range_check(dst, dst_offset, length, convert_dst);
1501 if (stopped()) {
1502 return true;
1503 }
1504
1505 Node* src_start = array_element_address(src, src_offset, src_elem);
1506 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1507 // 'src_start' points to src array + scaled offset
1508 // 'dst_start' points to dst array + scaled offset
1509 Node* count = nullptr;
1510 if (compress) {
1511 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1512 } else {
1513 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1514 }
1515
1516 if (alloc != nullptr) {
1517 if (alloc->maybe_set_complete(&_gvn)) {
1518 // "You break it, you buy it."
1519 InitializeNode* init = alloc->initialization();
1520 assert(init->is_complete(), "we just did this");
1521 init->set_complete_with_arraycopy();
1522 assert(dst->is_CheckCastPP(), "sanity");
1523 assert(dst->in(0)->in(0) == init, "dest pinned");
1524 }
1525 // Do not let stores that initialize this object be reordered with
1526 // a subsequent store that would make this object accessible by
1527 // other threads.
1528 // Record what AllocateNode this StoreStore protects so that
1529 // escape analysis can go from the MemBarStoreStoreNode to the
1530 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1531 // based on the escape status of the AllocateNode.
1532 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1533 }
1534 if (compress) {
1535 set_result(_gvn.transform(count));
1536 }
1537 clear_upper_avx();
1538
1539 return true;
1540 }
1541
1542 #ifdef _LP64
1543 #define XTOP ,top() /*additional argument*/
1544 #else //_LP64
1545 #define XTOP /*no additional argument*/
1546 #endif //_LP64
1547
1548 //------------------------inline_string_toBytesU--------------------------
1549 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1550 bool LibraryCallKit::inline_string_toBytesU() {
1551 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1552 return false;
1553 }
1554 // Get the arguments.
1555 Node* value = argument(0);
1556 Node* offset = argument(1);
1557 Node* length = argument(2);
1558
1559 Node* newcopy = nullptr;
1560
1561 // Set the original stack and the reexecute bit for the interpreter to reexecute
1562 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1563 { PreserveReexecuteState preexecs(this);
1564 jvms()->set_should_reexecute(true);
1565
1566 // Check if a null path was taken unconditionally.
1567 value = null_check(value);
1568
1569 RegionNode* bailout = new RegionNode(1);
1570 record_for_igvn(bailout);
1571
1572 // Range checks
1573 generate_negative_guard(offset, bailout);
1574 generate_negative_guard(length, bailout);
1575 generate_limit_guard(offset, length, load_array_length(value), bailout);
1576 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1577 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1578
1579 if (bailout->req() > 1) {
1580 PreserveJVMState pjvms(this);
1581 set_control(_gvn.transform(bailout));
1582 uncommon_trap(Deoptimization::Reason_intrinsic,
1583 Deoptimization::Action_maybe_recompile);
1584 }
1585 if (stopped()) {
1586 return true;
1587 }
1588
1589 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1590 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1591 newcopy = new_array(klass_node, size, 0); // no arguments to push
1592 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1593 guarantee(alloc != nullptr, "created above");
1594
1595 // Calculate starting addresses.
1596 Node* src_start = array_element_address(value, offset, T_CHAR);
1597 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1598
1599 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1600 const TypeInt* toffset = gvn().type(offset)->is_int();
1601 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1602
1603 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1604 const char* copyfunc_name = "arraycopy";
1605 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1606 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1607 OptoRuntime::fast_arraycopy_Type(),
1608 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1609 src_start, dst_start, ConvI2X(length) XTOP);
1610 // Do not let reads from the cloned object float above the arraycopy.
1611 if (alloc->maybe_set_complete(&_gvn)) {
1612 // "You break it, you buy it."
1613 InitializeNode* init = alloc->initialization();
1614 assert(init->is_complete(), "we just did this");
1615 init->set_complete_with_arraycopy();
1616 assert(newcopy->is_CheckCastPP(), "sanity");
1617 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1618 }
1619 // Do not let stores that initialize this object be reordered with
1620 // a subsequent store that would make this object accessible by
1621 // other threads.
1622 // Record what AllocateNode this StoreStore protects so that
1623 // escape analysis can go from the MemBarStoreStoreNode to the
1624 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1625 // based on the escape status of the AllocateNode.
1626 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1627 } // original reexecute is set back here
1628
1629 C->set_has_split_ifs(true); // Has chance for split-if optimization
1630 if (!stopped()) {
1631 set_result(newcopy);
1632 }
1633 clear_upper_avx();
1634
1635 return true;
1636 }
1637
1638 //------------------------inline_string_getCharsU--------------------------
1639 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1640 bool LibraryCallKit::inline_string_getCharsU() {
1641 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1642 return false;
1643 }
1644
1645 // Get the arguments.
1646 Node* src = argument(0);
1647 Node* src_begin = argument(1);
1648 Node* src_end = argument(2); // exclusive offset (i < src_end)
1649 Node* dst = argument(3);
1650 Node* dst_begin = argument(4);
1651
1652 // Check for allocation before we add nodes that would confuse
1653 // tightly_coupled_allocation()
1654 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1655
1656 // Check if a null path was taken unconditionally.
1657 src = null_check(src);
1658 dst = null_check(dst);
1659 if (stopped()) {
1660 return true;
1661 }
1662
1663 // Get length and convert char[] offset to byte[] offset
1664 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1665 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1666
1667 // Range checks
1668 generate_string_range_check(src, src_begin, length, true);
1669 generate_string_range_check(dst, dst_begin, length, false);
1670 if (stopped()) {
1671 return true;
1672 }
1673
1674 if (!stopped()) {
1675 // Calculate starting addresses.
1676 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1677 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1678
1679 // Check if array addresses are aligned to HeapWordSize
1680 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1681 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1682 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1683 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1684
1685 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1686 const char* copyfunc_name = "arraycopy";
1687 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1688 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1689 OptoRuntime::fast_arraycopy_Type(),
1690 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1691 src_start, dst_start, ConvI2X(length) XTOP);
1692 // Do not let reads from the cloned object float above the arraycopy.
1693 if (alloc != nullptr) {
1694 if (alloc->maybe_set_complete(&_gvn)) {
1695 // "You break it, you buy it."
1696 InitializeNode* init = alloc->initialization();
1697 assert(init->is_complete(), "we just did this");
1698 init->set_complete_with_arraycopy();
1699 assert(dst->is_CheckCastPP(), "sanity");
1700 assert(dst->in(0)->in(0) == init, "dest pinned");
1701 }
1702 // Do not let stores that initialize this object be reordered with
1703 // a subsequent store that would make this object accessible by
1704 // other threads.
1705 // Record what AllocateNode this StoreStore protects so that
1706 // escape analysis can go from the MemBarStoreStoreNode to the
1707 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1708 // based on the escape status of the AllocateNode.
1709 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1710 } else {
1711 insert_mem_bar(Op_MemBarCPUOrder);
1712 }
1713 }
1714
1715 C->set_has_split_ifs(true); // Has chance for split-if optimization
1716 return true;
1717 }
1718
1719 //----------------------inline_string_char_access----------------------------
1720 // Store/Load char to/from byte[] array.
1721 // static void StringUTF16.putChar(byte[] val, int index, int c)
1722 // static char StringUTF16.getChar(byte[] val, int index)
1723 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1724 Node* value = argument(0);
1725 Node* index = argument(1);
1726 Node* ch = is_store ? argument(2) : nullptr;
1727
1728 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1729 // correctly requires matched array shapes.
1730 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1731 "sanity: byte[] and char[] bases agree");
1732 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1733 "sanity: byte[] and char[] scales agree");
1734
1735 // Bail when getChar over constants is requested: constant folding would
1736 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1737 // Java method would constant fold nicely instead.
1738 if (!is_store && value->is_Con() && index->is_Con()) {
1739 return false;
1740 }
1741
1742 // Save state and restore on bailout
1743 uint old_sp = sp();
1744 SafePointNode* old_map = clone_map();
1745
1746 value = must_be_not_null(value, true);
1747
1748 Node* adr = array_element_address(value, index, T_CHAR);
1749 if (adr->is_top()) {
1750 set_map(old_map);
1751 set_sp(old_sp);
1752 return false;
1753 }
1754 destruct_map_clone(old_map);
1755 if (is_store) {
1756 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1757 } else {
1758 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);
1759 set_result(ch);
1760 }
1761 return true;
1762 }
1763
1764
1765 //------------------------------inline_math-----------------------------------
1766 // public static double Math.abs(double)
1767 // public static double Math.sqrt(double)
1768 // public static double Math.log(double)
1769 // public static double Math.log10(double)
1770 // public static double Math.round(double)
1771 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1772 Node* arg = argument(0);
1773 Node* n = nullptr;
1774 switch (id) {
1775 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1776 case vmIntrinsics::_dsqrt:
1777 case vmIntrinsics::_dsqrt_strict:
1778 n = new SqrtDNode(C, control(), arg); break;
1779 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1780 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1781 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1782 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1783 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1784 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1785 default: fatal_unexpected_iid(id); break;
1786 }
1787 set_result(_gvn.transform(n));
1788 return true;
1789 }
1790
1791 //------------------------------inline_math-----------------------------------
1792 // public static float Math.abs(float)
1793 // public static int Math.abs(int)
1794 // public static long Math.abs(long)
1795 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1796 Node* arg = argument(0);
1797 Node* n = nullptr;
1798 switch (id) {
1799 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1800 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1801 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1802 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1803 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1804 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1805 default: fatal_unexpected_iid(id); break;
1806 }
1807 set_result(_gvn.transform(n));
1808 return true;
1809 }
1810
1811 //------------------------------runtime_math-----------------------------
1812 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1813 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1814 "must be (DD)D or (D)D type");
1815
1816 // Inputs
1817 Node* a = argument(0);
1818 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1819
1820 const TypePtr* no_memory_effects = nullptr;
1821 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1822 no_memory_effects,
1823 a, top(), b, b ? top() : nullptr);
1824 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1825 #ifdef ASSERT
1826 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1827 assert(value_top == top(), "second value must be top");
1828 #endif
1829
1830 set_result(value);
1831 return true;
1832 }
1833
1834 //------------------------------inline_math_pow-----------------------------
1835 bool LibraryCallKit::inline_math_pow() {
1836 Node* exp = argument(2);
1837 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1838 if (d != nullptr) {
1839 if (d->getd() == 2.0) {
1840 // Special case: pow(x, 2.0) => x * x
1841 Node* base = argument(0);
1842 set_result(_gvn.transform(new MulDNode(base, base)));
1843 return true;
1844 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1845 // Special case: pow(x, 0.5) => sqrt(x)
1846 Node* base = argument(0);
1847 Node* zero = _gvn.zerocon(T_DOUBLE);
1848
1849 RegionNode* region = new RegionNode(3);
1850 Node* phi = new PhiNode(region, Type::DOUBLE);
1851
1852 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1853 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1854 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1855 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1856 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1857
1858 Node* if_pow = generate_slow_guard(test, nullptr);
1859 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1860 phi->init_req(1, value_sqrt);
1861 region->init_req(1, control());
1862
1863 if (if_pow != nullptr) {
1864 set_control(if_pow);
1865 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1866 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1867 const TypePtr* no_memory_effects = nullptr;
1868 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1869 no_memory_effects, base, top(), exp, top());
1870 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1871 #ifdef ASSERT
1872 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1873 assert(value_top == top(), "second value must be top");
1874 #endif
1875 phi->init_req(2, value_pow);
1876 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1877 }
1878
1879 C->set_has_split_ifs(true); // Has chance for split-if optimization
1880 set_control(_gvn.transform(region));
1881 record_for_igvn(region);
1882 set_result(_gvn.transform(phi));
1883
1884 return true;
1885 }
1886 }
1887
1888 return StubRoutines::dpow() != nullptr ?
1889 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1890 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1891 }
1892
1893 //------------------------------inline_math_native-----------------------------
1894 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1895 switch (id) {
1896 case vmIntrinsics::_dsin:
1897 return StubRoutines::dsin() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1900 case vmIntrinsics::_dcos:
1901 return StubRoutines::dcos() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1904 case vmIntrinsics::_dtan:
1905 return StubRoutines::dtan() != nullptr ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1907 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1908 case vmIntrinsics::_dtanh:
1909 return StubRoutines::dtanh() != nullptr ?
1910 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1911 case vmIntrinsics::_dexp:
1912 return StubRoutines::dexp() != nullptr ?
1913 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1914 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1915 case vmIntrinsics::_dlog:
1916 return StubRoutines::dlog() != nullptr ?
1917 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1918 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1919 case vmIntrinsics::_dlog10:
1920 return StubRoutines::dlog10() != nullptr ?
1921 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1922 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1923
1924 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1925 case vmIntrinsics::_ceil:
1926 case vmIntrinsics::_floor:
1927 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1928
1929 case vmIntrinsics::_dsqrt:
1930 case vmIntrinsics::_dsqrt_strict:
1931 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1932 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1933 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1934 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1935 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1936
1937 case vmIntrinsics::_dpow: return inline_math_pow();
1938 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1939 case vmIntrinsics::_fcopySign: return inline_math(id);
1940 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1941 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1942 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1943
1944 // These intrinsics are not yet correctly implemented
1945 case vmIntrinsics::_datan2:
1946 return false;
1947
1948 default:
1949 fatal_unexpected_iid(id);
1950 return false;
1951 }
1952 }
1953
1954 //----------------------------inline_notify-----------------------------------*
1955 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1956 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1957 address func;
1958 if (id == vmIntrinsics::_notify) {
1959 func = OptoRuntime::monitor_notify_Java();
1960 } else {
1961 func = OptoRuntime::monitor_notifyAll_Java();
1962 }
1963 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1964 make_slow_call_ex(call, env()->Throwable_klass(), false);
1965 return true;
1966 }
1967
1968
1969 //----------------------------inline_min_max-----------------------------------
1970 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1971 Node* a = nullptr;
1972 Node* b = nullptr;
1973 Node* n = nullptr;
1974 switch (id) {
1975 case vmIntrinsics::_min:
1976 case vmIntrinsics::_max:
1977 case vmIntrinsics::_minF:
1978 case vmIntrinsics::_maxF:
1979 case vmIntrinsics::_minF_strict:
1980 case vmIntrinsics::_maxF_strict:
1981 case vmIntrinsics::_min_strict:
1982 case vmIntrinsics::_max_strict:
1983 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1984 a = argument(0);
1985 b = argument(1);
1986 break;
1987 case vmIntrinsics::_minD:
1988 case vmIntrinsics::_maxD:
1989 case vmIntrinsics::_minD_strict:
1990 case vmIntrinsics::_maxD_strict:
1991 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1992 a = argument(0);
1993 b = argument(2);
1994 break;
1995 case vmIntrinsics::_minL:
1996 case vmIntrinsics::_maxL:
1997 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1998 a = argument(0);
1999 b = argument(2);
2000 break;
2001 default:
2002 fatal_unexpected_iid(id);
2003 break;
2004 }
2005
2006 switch (id) {
2007 case vmIntrinsics::_min:
2008 case vmIntrinsics::_min_strict:
2009 n = new MinINode(a, b);
2010 break;
2011 case vmIntrinsics::_max:
2012 case vmIntrinsics::_max_strict:
2013 n = new MaxINode(a, b);
2014 break;
2015 case vmIntrinsics::_minF:
2016 case vmIntrinsics::_minF_strict:
2017 n = new MinFNode(a, b);
2018 break;
2019 case vmIntrinsics::_maxF:
2020 case vmIntrinsics::_maxF_strict:
2021 n = new MaxFNode(a, b);
2022 break;
2023 case vmIntrinsics::_minD:
2024 case vmIntrinsics::_minD_strict:
2025 n = new MinDNode(a, b);
2026 break;
2027 case vmIntrinsics::_maxD:
2028 case vmIntrinsics::_maxD_strict:
2029 n = new MaxDNode(a, b);
2030 break;
2031 case vmIntrinsics::_minL:
2032 n = new MinLNode(_gvn.C, a, b);
2033 break;
2034 case vmIntrinsics::_maxL:
2035 n = new MaxLNode(_gvn.C, a, b);
2036 break;
2037 default:
2038 fatal_unexpected_iid(id);
2039 break;
2040 }
2041
2042 set_result(_gvn.transform(n));
2043 return true;
2044 }
2045
2046 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2047 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2048 env()->ArithmeticException_instance())) {
2049 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2050 // so let's bail out intrinsic rather than risking deopting again.
2051 return false;
2052 }
2053
2054 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2055 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2056 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2057 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2058
2059 {
2060 PreserveJVMState pjvms(this);
2061 PreserveReexecuteState preexecs(this);
2062 jvms()->set_should_reexecute(true);
2063
2064 set_control(slow_path);
2065 set_i_o(i_o());
2066
2067 builtin_throw(Deoptimization::Reason_intrinsic,
2068 env()->ArithmeticException_instance(),
2069 /*allow_too_many_traps*/ false);
2070 }
2071
2072 set_control(fast_path);
2073 set_result(math);
2074 return true;
2075 }
2076
2077 template <typename OverflowOp>
2078 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2079 typedef typename OverflowOp::MathOp MathOp;
2080
2081 MathOp* mathOp = new MathOp(arg1, arg2);
2082 Node* operation = _gvn.transform( mathOp );
2083 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2084 return inline_math_mathExact(operation, ofcheck);
2085 }
2086
2087 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2088 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2089 }
2090
2091 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2092 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2093 }
2094
2095 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2096 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2097 }
2098
2099 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2100 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2101 }
2102
2103 bool LibraryCallKit::inline_math_negateExactI() {
2104 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2105 }
2106
2107 bool LibraryCallKit::inline_math_negateExactL() {
2108 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2109 }
2110
2111 bool LibraryCallKit::inline_math_multiplyExactI() {
2112 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2113 }
2114
2115 bool LibraryCallKit::inline_math_multiplyExactL() {
2116 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2117 }
2118
2119 bool LibraryCallKit::inline_math_multiplyHigh() {
2120 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2121 return true;
2122 }
2123
2124 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2125 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2126 return true;
2127 }
2128
2129 inline int
2130 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2131 const TypePtr* base_type = TypePtr::NULL_PTR;
2132 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2133 if (base_type == nullptr) {
2134 // Unknown type.
2135 return Type::AnyPtr;
2136 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2137 // Since this is a null+long form, we have to switch to a rawptr.
2138 base = _gvn.transform(new CastX2PNode(offset));
2139 offset = MakeConX(0);
2140 return Type::RawPtr;
2141 } else if (base_type->base() == Type::RawPtr) {
2142 return Type::RawPtr;
2143 } else if (base_type->isa_oopptr()) {
2144 // Base is never null => always a heap address.
2145 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2146 return Type::OopPtr;
2147 }
2148 // Offset is small => always a heap address.
2149 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2150 if (offset_type != nullptr &&
2151 base_type->offset() == 0 && // (should always be?)
2152 offset_type->_lo >= 0 &&
2153 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2154 return Type::OopPtr;
2155 } else if (type == T_OBJECT) {
2156 // off heap access to an oop doesn't make any sense. Has to be on
2157 // heap.
2158 return Type::OopPtr;
2159 }
2160 // Otherwise, it might either be oop+off or null+addr.
2161 return Type::AnyPtr;
2162 } else {
2163 // No information:
2164 return Type::AnyPtr;
2165 }
2166 }
2167
2168 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2169 Node* uncasted_base = base;
2170 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2171 if (kind == Type::RawPtr) {
2172 return basic_plus_adr(top(), uncasted_base, offset);
2173 } else if (kind == Type::AnyPtr) {
2174 assert(base == uncasted_base, "unexpected base change");
2175 if (can_cast) {
2176 if (!_gvn.type(base)->speculative_maybe_null() &&
2177 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2178 // According to profiling, this access is always on
2179 // heap. Casting the base to not null and thus avoiding membars
2180 // around the access should allow better optimizations
2181 Node* null_ctl = top();
2182 base = null_check_oop(base, &null_ctl, true, true, true);
2183 assert(null_ctl->is_top(), "no null control here");
2184 return basic_plus_adr(base, offset);
2185 } else if (_gvn.type(base)->speculative_always_null() &&
2186 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2187 // According to profiling, this access is always off
2188 // heap.
2189 base = null_assert(base);
2190 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2191 offset = MakeConX(0);
2192 return basic_plus_adr(top(), raw_base, offset);
2193 }
2194 }
2195 // We don't know if it's an on heap or off heap access. Fall back
2196 // to raw memory access.
2197 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2198 return basic_plus_adr(top(), raw, offset);
2199 } else {
2200 assert(base == uncasted_base, "unexpected base change");
2201 // We know it's an on heap access so base can't be null
2202 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2203 base = must_be_not_null(base, true);
2204 }
2205 return basic_plus_adr(base, offset);
2206 }
2207 }
2208
2209 //--------------------------inline_number_methods-----------------------------
2210 // inline int Integer.numberOfLeadingZeros(int)
2211 // inline int Long.numberOfLeadingZeros(long)
2212 //
2213 // inline int Integer.numberOfTrailingZeros(int)
2214 // inline int Long.numberOfTrailingZeros(long)
2215 //
2216 // inline int Integer.bitCount(int)
2217 // inline int Long.bitCount(long)
2218 //
2219 // inline char Character.reverseBytes(char)
2220 // inline short Short.reverseBytes(short)
2221 // inline int Integer.reverseBytes(int)
2222 // inline long Long.reverseBytes(long)
2223 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2224 Node* arg = argument(0);
2225 Node* n = nullptr;
2226 switch (id) {
2227 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2228 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2229 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2230 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2231 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2232 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2233 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2234 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2235 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2236 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2237 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2238 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2239 default: fatal_unexpected_iid(id); break;
2240 }
2241 set_result(_gvn.transform(n));
2242 return true;
2243 }
2244
2245 //--------------------------inline_bitshuffle_methods-----------------------------
2246 // inline int Integer.compress(int, int)
2247 // inline int Integer.expand(int, int)
2248 // inline long Long.compress(long, long)
2249 // inline long Long.expand(long, long)
2250 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2251 Node* n = nullptr;
2252 switch (id) {
2253 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2254 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2255 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2256 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2257 default: fatal_unexpected_iid(id); break;
2258 }
2259 set_result(_gvn.transform(n));
2260 return true;
2261 }
2262
2263 //--------------------------inline_number_methods-----------------------------
2264 // inline int Integer.compareUnsigned(int, int)
2265 // inline int Long.compareUnsigned(long, long)
2266 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2267 Node* arg1 = argument(0);
2268 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2269 Node* n = nullptr;
2270 switch (id) {
2271 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2272 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2273 default: fatal_unexpected_iid(id); break;
2274 }
2275 set_result(_gvn.transform(n));
2276 return true;
2277 }
2278
2279 //--------------------------inline_unsigned_divmod_methods-----------------------------
2280 // inline int Integer.divideUnsigned(int, int)
2281 // inline int Integer.remainderUnsigned(int, int)
2282 // inline long Long.divideUnsigned(long, long)
2283 // inline long Long.remainderUnsigned(long, long)
2284 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2285 Node* n = nullptr;
2286 switch (id) {
2287 case vmIntrinsics::_divideUnsigned_i: {
2288 zero_check_int(argument(1));
2289 // Compile-time detect of null-exception
2290 if (stopped()) {
2291 return true; // keep the graph constructed so far
2292 }
2293 n = new UDivINode(control(), argument(0), argument(1));
2294 break;
2295 }
2296 case vmIntrinsics::_divideUnsigned_l: {
2297 zero_check_long(argument(2));
2298 // Compile-time detect of null-exception
2299 if (stopped()) {
2300 return true; // keep the graph constructed so far
2301 }
2302 n = new UDivLNode(control(), argument(0), argument(2));
2303 break;
2304 }
2305 case vmIntrinsics::_remainderUnsigned_i: {
2306 zero_check_int(argument(1));
2307 // Compile-time detect of null-exception
2308 if (stopped()) {
2309 return true; // keep the graph constructed so far
2310 }
2311 n = new UModINode(control(), argument(0), argument(1));
2312 break;
2313 }
2314 case vmIntrinsics::_remainderUnsigned_l: {
2315 zero_check_long(argument(2));
2316 // Compile-time detect of null-exception
2317 if (stopped()) {
2318 return true; // keep the graph constructed so far
2319 }
2320 n = new UModLNode(control(), argument(0), argument(2));
2321 break;
2322 }
2323 default: fatal_unexpected_iid(id); break;
2324 }
2325 set_result(_gvn.transform(n));
2326 return true;
2327 }
2328
2329 //----------------------------inline_unsafe_access----------------------------
2330
2331 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2332 // Attempt to infer a sharper value type from the offset and base type.
2333 ciKlass* sharpened_klass = nullptr;
2334 bool null_free = false;
2335
2336 // See if it is an instance field, with an object type.
2337 if (alias_type->field() != nullptr) {
2338 if (alias_type->field()->type()->is_klass()) {
2339 sharpened_klass = alias_type->field()->type()->as_klass();
2340 null_free = alias_type->field()->is_null_free();
2341 }
2342 }
2343
2344 const TypeOopPtr* result = nullptr;
2345 // See if it is a narrow oop array.
2346 if (adr_type->isa_aryptr()) {
2347 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2348 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2349 null_free = adr_type->is_aryptr()->is_null_free();
2350 if (elem_type != nullptr && elem_type->is_loaded()) {
2351 // Sharpen the value type.
2352 result = elem_type;
2353 }
2354 }
2355 }
2356
2357 // The sharpened class might be unloaded if there is no class loader
2358 // contraint in place.
2359 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2360 // Sharpen the value type.
2361 result = TypeOopPtr::make_from_klass(sharpened_klass);
2362 if (null_free) {
2363 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2364 }
2365 }
2366 if (result != nullptr) {
2367 #ifndef PRODUCT
2368 if (C->print_intrinsics() || C->print_inlining()) {
2369 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2370 tty->print(" sharpened value: "); result->dump(); tty->cr();
2371 }
2372 #endif
2373 }
2374 return result;
2375 }
2376
2377 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2378 switch (kind) {
2379 case Relaxed:
2380 return MO_UNORDERED;
2381 case Opaque:
2382 return MO_RELAXED;
2383 case Acquire:
2384 return MO_ACQUIRE;
2385 case Release:
2386 return MO_RELEASE;
2387 case Volatile:
2388 return MO_SEQ_CST;
2389 default:
2390 ShouldNotReachHere();
2391 return 0;
2392 }
2393 }
2394
2395 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2396 if (callee()->is_static()) return false; // caller must have the capability!
2397 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2398 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2399 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2400 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2401
2402 if (is_reference_type(type)) {
2403 decorators |= ON_UNKNOWN_OOP_REF;
2404 }
2405
2406 if (unaligned) {
2407 decorators |= C2_UNALIGNED;
2408 }
2409
2410 #ifndef PRODUCT
2411 {
2412 ResourceMark rm;
2413 // Check the signatures.
2414 ciSignature* sig = callee()->signature();
2415 #ifdef ASSERT
2416 if (!is_store) {
2417 // Object getReference(Object base, int/long offset), etc.
2418 BasicType rtype = sig->return_type()->basic_type();
2419 assert(rtype == type, "getter must return the expected value");
2420 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2421 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2422 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2423 } else {
2424 // void putReference(Object base, int/long offset, Object x), etc.
2425 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2426 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2427 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2428 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2429 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2430 assert(vtype == type, "putter must accept the expected value");
2431 }
2432 #endif // ASSERT
2433 }
2434 #endif //PRODUCT
2435
2436 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2437
2438 Node* receiver = argument(0); // type: oop
2439
2440 // Build address expression.
2441 Node* heap_base_oop = top();
2442
2443 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2444 Node* base = argument(1); // type: oop
2445 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2446 Node* offset = argument(2); // type: long
2447 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2448 // to be plain byte offsets, which are also the same as those accepted
2449 // by oopDesc::field_addr.
2450 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2451 "fieldOffset must be byte-scaled");
2452
2453 ciInlineKlass* inline_klass = nullptr;
2454 if (is_flat) {
2455 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2456 if (cls == nullptr || cls->const_oop() == nullptr) {
2457 return false;
2458 }
2459 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2460 if (!mirror_type->is_inlinetype()) {
2461 return false;
2462 }
2463 inline_klass = mirror_type->as_inline_klass();
2464 }
2465
2466 if (base->is_InlineType()) {
2467 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2468 InlineTypeNode* vt = base->as_InlineType();
2469 if (offset->is_Con()) {
2470 long off = find_long_con(offset, 0);
2471 ciInlineKlass* vk = vt->type()->inline_klass();
2472 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2473 return false;
2474 }
2475
2476 ciField* field = vk->get_non_flat_field_by_offset(off);
2477 if (field != nullptr) {
2478 BasicType bt = type2field[field->type()->basic_type()];
2479 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2480 bt = T_OBJECT;
2481 }
2482 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2483 Node* value = vt->field_value_by_offset(off, false);
2484 if (value->is_InlineType()) {
2485 value = value->as_InlineType()->adjust_scalarization_depth(this);
2486 }
2487 set_result(value);
2488 return true;
2489 }
2490 }
2491 }
2492 {
2493 // Re-execute the unsafe access if allocation triggers deoptimization.
2494 PreserveReexecuteState preexecs(this);
2495 jvms()->set_should_reexecute(true);
2496 vt = vt->buffer(this);
2497 }
2498 base = vt->get_oop();
2499 }
2500
2501 // 32-bit machines ignore the high half!
2502 offset = ConvL2X(offset);
2503
2504 // Save state and restore on bailout
2505 uint old_sp = sp();
2506 SafePointNode* old_map = clone_map();
2507
2508 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2509 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2510
2511 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2512 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2513 decorators |= IN_NATIVE; // off-heap primitive access
2514 } else {
2515 set_map(old_map);
2516 set_sp(old_sp);
2517 return false; // off-heap oop accesses are not supported
2518 }
2519 } else {
2520 heap_base_oop = base; // on-heap or mixed access
2521 }
2522
2523 // Can base be null? Otherwise, always on-heap access.
2524 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2525
2526 if (!can_access_non_heap) {
2527 decorators |= IN_HEAP;
2528 }
2529
2530 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2531
2532 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2533 if (adr_type == TypePtr::NULL_PTR) {
2534 set_map(old_map);
2535 set_sp(old_sp);
2536 return false; // off-heap access with zero address
2537 }
2538
2539 // Try to categorize the address.
2540 Compile::AliasType* alias_type = C->alias_type(adr_type);
2541 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2542
2543 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2544 alias_type->adr_type() == TypeAryPtr::RANGE) {
2545 set_map(old_map);
2546 set_sp(old_sp);
2547 return false; // not supported
2548 }
2549
2550 bool mismatched = false;
2551 BasicType bt = T_ILLEGAL;
2552 ciField* field = nullptr;
2553 if (adr_type->isa_instptr()) {
2554 const TypeInstPtr* instptr = adr_type->is_instptr();
2555 ciInstanceKlass* k = instptr->instance_klass();
2556 int off = instptr->offset();
2557 if (instptr->const_oop() != nullptr &&
2558 k == ciEnv::current()->Class_klass() &&
2559 instptr->offset() >= (k->size_helper() * wordSize)) {
2560 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2561 field = k->get_field_by_offset(off, true);
2562 } else {
2563 field = k->get_non_flat_field_by_offset(off);
2564 }
2565 if (field != nullptr) {
2566 bt = type2field[field->type()->basic_type()];
2567 }
2568 if (bt != alias_type->basic_type()) {
2569 // Type mismatch. Is it an access to a nested flat field?
2570 field = k->get_field_by_offset(off, false);
2571 if (field != nullptr) {
2572 bt = type2field[field->type()->basic_type()];
2573 }
2574 }
2575 assert(bt == alias_type->basic_type() || is_flat, "should match");
2576 } else {
2577 bt = alias_type->basic_type();
2578 }
2579
2580 if (bt != T_ILLEGAL) {
2581 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2582 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2583 // Alias type doesn't differentiate between byte[] and boolean[]).
2584 // Use address type to get the element type.
2585 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2586 }
2587 if (is_reference_type(bt, true)) {
2588 // accessing an array field with getReference is not a mismatch
2589 bt = T_OBJECT;
2590 }
2591 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2592 // Don't intrinsify mismatched object accesses
2593 set_map(old_map);
2594 set_sp(old_sp);
2595 return false;
2596 }
2597 mismatched = (bt != type);
2598 } else if (alias_type->adr_type()->isa_oopptr()) {
2599 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2600 }
2601
2602 if (is_flat) {
2603 if (adr_type->isa_instptr()) {
2604 if (field == nullptr || field->type() != inline_klass) {
2605 mismatched = true;
2606 }
2607 } else if (adr_type->isa_aryptr()) {
2608 const Type* elem = adr_type->is_aryptr()->elem();
2609 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2610 mismatched = true;
2611 }
2612 } else {
2613 mismatched = true;
2614 }
2615 if (is_store) {
2616 const Type* val_t = _gvn.type(val);
2617 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2618 set_map(old_map);
2619 set_sp(old_sp);
2620 return false;
2621 }
2622 }
2623 }
2624
2625 destruct_map_clone(old_map);
2626 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2627
2628 if (mismatched) {
2629 decorators |= C2_MISMATCHED;
2630 }
2631
2632 // First guess at the value type.
2633 const Type *value_type = Type::get_const_basic_type(type);
2634
2635 // Figure out the memory ordering.
2636 decorators |= mo_decorator_for_access_kind(kind);
2637
2638 if (!is_store) {
2639 if (type == T_OBJECT && !is_flat) {
2640 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2641 if (tjp != nullptr) {
2642 value_type = tjp;
2643 }
2644 }
2645 }
2646
2647 receiver = null_check(receiver);
2648 if (stopped()) {
2649 return true;
2650 }
2651 // Heap pointers get a null-check from the interpreter,
2652 // as a courtesy. However, this is not guaranteed by Unsafe,
2653 // and it is not possible to fully distinguish unintended nulls
2654 // from intended ones in this API.
2655
2656 if (!is_store) {
2657 Node* p = nullptr;
2658 // Try to constant fold a load from a constant field
2659
2660 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2661 // final or stable field
2662 p = make_constant_from_field(field, heap_base_oop);
2663 }
2664
2665 if (p == nullptr) { // Could not constant fold the load
2666 if (is_flat) {
2667 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, adr_type, false, false, true);
2668 } else {
2669 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2670 const TypeOopPtr* ptr = value_type->make_oopptr();
2671 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2672 // Load a non-flattened inline type from memory
2673 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2674 }
2675 }
2676 // Normalize the value returned by getBoolean in the following cases
2677 if (type == T_BOOLEAN &&
2678 (mismatched ||
2679 heap_base_oop == top() || // - heap_base_oop is null or
2680 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2681 // and the unsafe access is made to large offset
2682 // (i.e., larger than the maximum offset necessary for any
2683 // field access)
2684 ) {
2685 IdealKit ideal = IdealKit(this);
2686 #define __ ideal.
2687 IdealVariable normalized_result(ideal);
2688 __ declarations_done();
2689 __ set(normalized_result, p);
2690 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2691 __ set(normalized_result, ideal.ConI(1));
2692 ideal.end_if();
2693 final_sync(ideal);
2694 p = __ value(normalized_result);
2695 #undef __
2696 }
2697 }
2698 if (type == T_ADDRESS) {
2699 p = gvn().transform(new CastP2XNode(nullptr, p));
2700 p = ConvX2UL(p);
2701 }
2702 // The load node has the control of the preceding MemBarCPUOrder. All
2703 // following nodes will have the control of the MemBarCPUOrder inserted at
2704 // the end of this method. So, pushing the load onto the stack at a later
2705 // point is fine.
2706 set_result(p);
2707 } else {
2708 if (bt == T_ADDRESS) {
2709 // Repackage the long as a pointer.
2710 val = ConvL2X(val);
2711 val = gvn().transform(new CastX2PNode(val));
2712 }
2713 if (is_flat) {
2714 val->as_InlineType()->store_flat(this, base, adr, false, false, true, decorators);
2715 } else {
2716 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2717 }
2718 }
2719
2720 return true;
2721 }
2722
2723 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2724 #ifdef ASSERT
2725 {
2726 ResourceMark rm;
2727 // Check the signatures.
2728 ciSignature* sig = callee()->signature();
2729 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2730 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2731 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2732 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2733 if (is_store) {
2734 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2735 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2736 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2737 } else {
2738 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2739 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2740 }
2741 }
2742 #endif // ASSERT
2743
2744 assert(kind == Relaxed, "Only plain accesses for now");
2745 if (callee()->is_static()) {
2746 // caller must have the capability!
2747 return false;
2748 }
2749 C->set_has_unsafe_access(true);
2750
2751 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2752 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2753 // parameter valueType is not a constant
2754 return false;
2755 }
2756 ciInlineKlass* value_klass = value_klass_node->const_oop()->as_instance()->java_mirror_type()->as_inline_klass();
2757
2758 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2759 if (layout_type == nullptr || !layout_type->is_con()) {
2760 // parameter layoutKind is not a constant
2761 return false;
2762 }
2763 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2764 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2765 "invalid layoutKind %d", layout_type->get_con());
2766 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2767 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NON_ATOMIC_FLAT ||
2768 layout == LayoutKind::ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2769 "unexpected layoutKind %d", layout_type->get_con());
2770
2771 null_check(argument(0));
2772 if (stopped()) {
2773 return true;
2774 }
2775
2776 Node* base = must_be_not_null(argument(1), true);
2777 Node* offset = argument(2);
2778 const Type* base_type = _gvn.type(base);
2779
2780 Node* ptr;
2781 bool immutable_memory = false;
2782 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2783 if (base_type->isa_instptr()) {
2784 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2785 if (offset_type == nullptr || !offset_type->is_con()) {
2786 // Offset into a non-array should be a constant
2787 decorators |= C2_MISMATCHED;
2788 } else {
2789 int offset_con = checked_cast<int>(offset_type->get_con());
2790 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2791 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2792 if (field == nullptr) {
2793 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2794 decorators |= C2_MISMATCHED;
2795 } else {
2796 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2797 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2798 immutable_memory = field->is_strict() && field->is_final();
2799
2800 if (base->is_InlineType()) {
2801 assert(!is_store, "Cannot store into a non-larval value object");
2802 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2803 return true;
2804 }
2805 }
2806 }
2807
2808 if (base->is_InlineType()) {
2809 assert(!is_store, "Cannot store into a non-larval value object");
2810 base = base->as_InlineType()->buffer(this, true);
2811 }
2812 ptr = basic_plus_adr(base, ConvL2X(offset));
2813 } else if (base_type->isa_aryptr()) {
2814 decorators |= IS_ARRAY;
2815 if (layout == LayoutKind::REFERENCE) {
2816 if (!base_type->is_aryptr()->is_not_flat()) {
2817 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2818 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::StrongDependency));
2819 replace_in_map(base, new_base);
2820 base = new_base;
2821 }
2822 ptr = basic_plus_adr(base, ConvL2X(offset));
2823 } else {
2824 // Flat array must have an exact type
2825 bool is_null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2826 bool is_atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2827 Node* new_base = cast_to_flat_array(base, value_klass, is_null_free, !is_null_free, is_atomic);
2828 replace_in_map(base, new_base);
2829 base = new_base;
2830 ptr = basic_plus_adr(base, ConvL2X(offset));
2831 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2832 if (ptr_type->field_offset().get() != 0) {
2833 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::StrongDependency));
2834 }
2835 }
2836 } else {
2837 decorators |= C2_MISMATCHED;
2838 ptr = basic_plus_adr(base, ConvL2X(offset));
2839 }
2840
2841 if (is_store) {
2842 Node* value = argument(6);
2843 const Type* value_type = _gvn.type(value);
2844 if (!value_type->is_inlinetypeptr()) {
2845 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2846 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::StrongDependency));
2847 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2848 replace_in_map(value, new_value);
2849 value = new_value;
2850 }
2851
2852 assert(value_type->inline_klass() == value_klass, "value is of type %s while valueType is %s", value_type->inline_klass()->name()->as_utf8(), value_klass->name()->as_utf8());
2853 if (layout == LayoutKind::REFERENCE) {
2854 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2855 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2856 } else {
2857 bool atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2858 bool null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2859 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2860 }
2861
2862 return true;
2863 } else {
2864 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2865 InlineTypeNode* result;
2866 if (layout == LayoutKind::REFERENCE) {
2867 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2868 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2869 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2870 } else {
2871 bool atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2872 bool null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2873 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2874 }
2875
2876 set_result(result);
2877 return true;
2878 }
2879 }
2880
2881 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2882 Node* receiver = argument(0);
2883 Node* value = argument(1);
2884
2885 const Type* type = gvn().type(value);
2886 if (!type->is_inlinetypeptr()) {
2887 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2888 return false;
2889 }
2890
2891 null_check(receiver);
2892 if (stopped()) {
2893 return true;
2894 }
2895
2896 value = null_check(value);
2897 if (stopped()) {
2898 return true;
2899 }
2900
2901 ciInlineKlass* vk = type->inline_klass();
2902 Node* klass = makecon(TypeKlassPtr::make(vk));
2903 Node* obj = new_instance(klass);
2904 AllocateNode::Ideal_allocation(obj)->_larval = true;
2905
2906 assert(value->is_InlineType(), "must be an InlineTypeNode");
2907 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2908 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2909
2910 set_result(obj);
2911 return true;
2912 }
2913
2914 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2915 Node* receiver = argument(0);
2916 Node* buffer = argument(1);
2917
2918 const Type* type = gvn().type(buffer);
2919 if (!type->is_inlinetypeptr()) {
2920 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2921 return false;
2922 }
2923
2924 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2925 if (alloc == nullptr) {
2926 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2927 return false;
2928 }
2929
2930 null_check(receiver);
2931 if (stopped()) {
2932 return true;
2933 }
2934
2935 // Unset the larval bit in the object header
2936 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2937 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2938 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2939
2940 // We must ensure that the buffer is properly published
2941 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2942 assert(!type->maybe_null(), "result of an allocation should not be null");
2943 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
2944 return true;
2945 }
2946
2947 //----------------------------inline_unsafe_load_store----------------------------
2948 // This method serves a couple of different customers (depending on LoadStoreKind):
2949 //
2950 // LS_cmp_swap:
2951 //
2952 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2953 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2954 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2955 //
2956 // LS_cmp_swap_weak:
2957 //
2958 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2959 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2960 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2961 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2962 //
2963 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2964 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2965 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2966 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2967 //
2968 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2969 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2970 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2971 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2972 //
2973 // LS_cmp_exchange:
2974 //
2975 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2976 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2977 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2978 //
2979 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2980 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2981 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2982 //
2983 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2984 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2985 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2986 //
2987 // LS_get_add:
2988 //
2989 // int getAndAddInt( Object o, long offset, int delta)
2990 // long getAndAddLong(Object o, long offset, long delta)
2991 //
2992 // LS_get_set:
2993 //
2994 // int getAndSet(Object o, long offset, int newValue)
2995 // long getAndSet(Object o, long offset, long newValue)
2996 // Object getAndSet(Object o, long offset, Object newValue)
2997 //
2998 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2999 // This basic scheme here is the same as inline_unsafe_access, but
3000 // differs in enough details that combining them would make the code
3001 // overly confusing. (This is a true fact! I originally combined
3002 // them, but even I was confused by it!) As much code/comments as
3003 // possible are retained from inline_unsafe_access though to make
3004 // the correspondences clearer. - dl
3005
3006 if (callee()->is_static()) return false; // caller must have the capability!
3007
3008 DecoratorSet decorators = C2_UNSAFE_ACCESS;
3009 decorators |= mo_decorator_for_access_kind(access_kind);
3010
3011 #ifndef PRODUCT
3012 BasicType rtype;
3013 {
3014 ResourceMark rm;
3015 // Check the signatures.
3016 ciSignature* sig = callee()->signature();
3017 rtype = sig->return_type()->basic_type();
3018 switch(kind) {
3019 case LS_get_add:
3020 case LS_get_set: {
3021 // Check the signatures.
3022 #ifdef ASSERT
3023 assert(rtype == type, "get and set must return the expected type");
3024 assert(sig->count() == 3, "get and set has 3 arguments");
3025 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
3026 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
3027 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
3028 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
3029 #endif // ASSERT
3030 break;
3031 }
3032 case LS_cmp_swap:
3033 case LS_cmp_swap_weak: {
3034 // Check the signatures.
3035 #ifdef ASSERT
3036 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3037 assert(sig->count() == 4, "CAS has 4 arguments");
3038 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3039 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3040 #endif // ASSERT
3041 break;
3042 }
3043 case LS_cmp_exchange: {
3044 // Check the signatures.
3045 #ifdef ASSERT
3046 assert(rtype == type, "CAS must return the expected type");
3047 assert(sig->count() == 4, "CAS has 4 arguments");
3048 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3049 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3050 #endif // ASSERT
3051 break;
3052 }
3053 default:
3054 ShouldNotReachHere();
3055 }
3056 }
3057 #endif //PRODUCT
3058
3059 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3060
3061 // Get arguments:
3062 Node* receiver = nullptr;
3063 Node* base = nullptr;
3064 Node* offset = nullptr;
3065 Node* oldval = nullptr;
3066 Node* newval = nullptr;
3067 switch(kind) {
3068 case LS_cmp_swap:
3069 case LS_cmp_swap_weak:
3070 case LS_cmp_exchange: {
3071 const bool two_slot_type = type2size[type] == 2;
3072 receiver = argument(0); // type: oop
3073 base = argument(1); // type: oop
3074 offset = argument(2); // type: long
3075 oldval = argument(4); // type: oop, int, or long
3076 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3077 break;
3078 }
3079 case LS_get_add:
3080 case LS_get_set: {
3081 receiver = argument(0); // type: oop
3082 base = argument(1); // type: oop
3083 offset = argument(2); // type: long
3084 oldval = nullptr;
3085 newval = argument(4); // type: oop, int, or long
3086 break;
3087 }
3088 default:
3089 ShouldNotReachHere();
3090 }
3091
3092 // Build field offset expression.
3093 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3094 // to be plain byte offsets, which are also the same as those accepted
3095 // by oopDesc::field_addr.
3096 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3097 // 32-bit machines ignore the high half of long offsets
3098 offset = ConvL2X(offset);
3099 // Save state and restore on bailout
3100 uint old_sp = sp();
3101 SafePointNode* old_map = clone_map();
3102 Node* adr = make_unsafe_address(base, offset,type, false);
3103 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3104
3105 Compile::AliasType* alias_type = C->alias_type(adr_type);
3106 BasicType bt = alias_type->basic_type();
3107 if (bt != T_ILLEGAL &&
3108 (is_reference_type(bt) != (type == T_OBJECT))) {
3109 // Don't intrinsify mismatched object accesses.
3110 set_map(old_map);
3111 set_sp(old_sp);
3112 return false;
3113 }
3114
3115 destruct_map_clone(old_map);
3116
3117 // For CAS, unlike inline_unsafe_access, there seems no point in
3118 // trying to refine types. Just use the coarse types here.
3119 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3120 const Type *value_type = Type::get_const_basic_type(type);
3121
3122 switch (kind) {
3123 case LS_get_set:
3124 case LS_cmp_exchange: {
3125 if (type == T_OBJECT) {
3126 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3127 if (tjp != nullptr) {
3128 value_type = tjp;
3129 }
3130 }
3131 break;
3132 }
3133 case LS_cmp_swap:
3134 case LS_cmp_swap_weak:
3135 case LS_get_add:
3136 break;
3137 default:
3138 ShouldNotReachHere();
3139 }
3140
3141 // Null check receiver.
3142 receiver = null_check(receiver);
3143 if (stopped()) {
3144 return true;
3145 }
3146
3147 int alias_idx = C->get_alias_index(adr_type);
3148
3149 if (is_reference_type(type)) {
3150 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3151
3152 if (oldval != nullptr && oldval->is_InlineType()) {
3153 // Re-execute the unsafe access if allocation triggers deoptimization.
3154 PreserveReexecuteState preexecs(this);
3155 jvms()->set_should_reexecute(true);
3156 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3157 }
3158 if (newval != nullptr && newval->is_InlineType()) {
3159 // Re-execute the unsafe access if allocation triggers deoptimization.
3160 PreserveReexecuteState preexecs(this);
3161 jvms()->set_should_reexecute(true);
3162 newval = newval->as_InlineType()->buffer(this)->get_oop();
3163 }
3164
3165 // Transformation of a value which could be null pointer (CastPP #null)
3166 // could be delayed during Parse (for example, in adjust_map_after_if()).
3167 // Execute transformation here to avoid barrier generation in such case.
3168 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3169 newval = _gvn.makecon(TypePtr::NULL_PTR);
3170
3171 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3172 // Refine the value to a null constant, when it is known to be null
3173 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3174 }
3175 }
3176
3177 Node* result = nullptr;
3178 switch (kind) {
3179 case LS_cmp_exchange: {
3180 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3181 oldval, newval, value_type, type, decorators);
3182 break;
3183 }
3184 case LS_cmp_swap_weak:
3185 decorators |= C2_WEAK_CMPXCHG;
3186 case LS_cmp_swap: {
3187 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3188 oldval, newval, value_type, type, decorators);
3189 break;
3190 }
3191 case LS_get_set: {
3192 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3193 newval, value_type, type, decorators);
3194 break;
3195 }
3196 case LS_get_add: {
3197 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3198 newval, value_type, type, decorators);
3199 break;
3200 }
3201 default:
3202 ShouldNotReachHere();
3203 }
3204
3205 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3206 set_result(result);
3207 return true;
3208 }
3209
3210 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3211 // Regardless of form, don't allow previous ld/st to move down,
3212 // then issue acquire, release, or volatile mem_bar.
3213 insert_mem_bar(Op_MemBarCPUOrder);
3214 switch(id) {
3215 case vmIntrinsics::_loadFence:
3216 insert_mem_bar(Op_LoadFence);
3217 return true;
3218 case vmIntrinsics::_storeFence:
3219 insert_mem_bar(Op_StoreFence);
3220 return true;
3221 case vmIntrinsics::_storeStoreFence:
3222 insert_mem_bar(Op_StoreStoreFence);
3223 return true;
3224 case vmIntrinsics::_fullFence:
3225 insert_mem_bar(Op_MemBarVolatile);
3226 return true;
3227 default:
3228 fatal_unexpected_iid(id);
3229 return false;
3230 }
3231 }
3232
3233 bool LibraryCallKit::inline_onspinwait() {
3234 insert_mem_bar(Op_OnSpinWait);
3235 return true;
3236 }
3237
3238 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3239 if (!kls->is_Con()) {
3240 return true;
3241 }
3242 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3243 if (klsptr == nullptr) {
3244 return true;
3245 }
3246 ciInstanceKlass* ik = klsptr->instance_klass();
3247 // don't need a guard for a klass that is already initialized
3248 return !ik->is_initialized();
3249 }
3250
3251 //----------------------------inline_unsafe_writeback0-------------------------
3252 // public native void Unsafe.writeback0(long address)
3253 bool LibraryCallKit::inline_unsafe_writeback0() {
3254 if (!Matcher::has_match_rule(Op_CacheWB)) {
3255 return false;
3256 }
3257 #ifndef PRODUCT
3258 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3259 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3260 ciSignature* sig = callee()->signature();
3261 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3262 #endif
3263 null_check_receiver(); // null-check, then ignore
3264 Node *addr = argument(1);
3265 addr = new CastX2PNode(addr);
3266 addr = _gvn.transform(addr);
3267 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3268 flush = _gvn.transform(flush);
3269 set_memory(flush, TypeRawPtr::BOTTOM);
3270 return true;
3271 }
3272
3273 //----------------------------inline_unsafe_writeback0-------------------------
3274 // public native void Unsafe.writeback0(long address)
3275 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3276 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3277 return false;
3278 }
3279 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3280 return false;
3281 }
3282 #ifndef PRODUCT
3283 assert(Matcher::has_match_rule(Op_CacheWB),
3284 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3285 : "found match rule for CacheWBPostSync but not CacheWB"));
3286
3287 #endif
3288 null_check_receiver(); // null-check, then ignore
3289 Node *sync;
3290 if (is_pre) {
3291 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3292 } else {
3293 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3294 }
3295 sync = _gvn.transform(sync);
3296 set_memory(sync, TypeRawPtr::BOTTOM);
3297 return true;
3298 }
3299
3300 //----------------------------inline_unsafe_allocate---------------------------
3301 // public native Object Unsafe.allocateInstance(Class<?> cls);
3302 bool LibraryCallKit::inline_unsafe_allocate() {
3303
3304 #if INCLUDE_JVMTI
3305 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3306 return false;
3307 }
3308 #endif //INCLUDE_JVMTI
3309
3310 if (callee()->is_static()) return false; // caller must have the capability!
3311
3312 null_check_receiver(); // null-check, then ignore
3313 Node* cls = null_check(argument(1));
3314 if (stopped()) return true;
3315
3316 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3317 kls = null_check(kls);
3318 if (stopped()) return true; // argument was like int.class
3319
3320 #if INCLUDE_JVMTI
3321 // Don't try to access new allocated obj in the intrinsic.
3322 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3323 // Deoptimize and allocate in interpreter instead.
3324 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3325 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3326 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3327 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3328 {
3329 BuildCutout unless(this, tst, PROB_MAX);
3330 uncommon_trap(Deoptimization::Reason_intrinsic,
3331 Deoptimization::Action_make_not_entrant);
3332 }
3333 if (stopped()) {
3334 return true;
3335 }
3336 #endif //INCLUDE_JVMTI
3337
3338 Node* test = nullptr;
3339 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3340 // Note: The argument might still be an illegal value like
3341 // Serializable.class or Object[].class. The runtime will handle it.
3342 // But we must make an explicit check for initialization.
3343 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3344 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3345 // can generate code to load it as unsigned byte.
3346 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3347 Node* bits = intcon(InstanceKlass::fully_initialized);
3348 test = _gvn.transform(new SubINode(inst, bits));
3349 // The 'test' is non-zero if we need to take a slow path.
3350 }
3351 Node* obj = nullptr;
3352 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3353 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3354 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3355 } else {
3356 obj = new_instance(kls, test);
3357 }
3358 set_result(obj);
3359 return true;
3360 }
3361
3362 //------------------------inline_native_time_funcs--------------
3363 // inline code for System.currentTimeMillis() and System.nanoTime()
3364 // these have the same type and signature
3365 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3366 const TypeFunc* tf = OptoRuntime::void_long_Type();
3367 const TypePtr* no_memory_effects = nullptr;
3368 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3369 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3370 #ifdef ASSERT
3371 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3372 assert(value_top == top(), "second value must be top");
3373 #endif
3374 set_result(value);
3375 return true;
3376 }
3377
3378
3379 #if INCLUDE_JVMTI
3380
3381 // When notifications are disabled then just update the VTMS transition bit and return.
3382 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3383 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3384 if (!DoJVMTIVirtualThreadTransitions) {
3385 return true;
3386 }
3387 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3388 IdealKit ideal(this);
3389
3390 Node* ONE = ideal.ConI(1);
3391 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3392 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3393 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3394
3395 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3396 sync_kit(ideal);
3397 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3398 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3399 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3400 ideal.sync_kit(this);
3401 } ideal.else_(); {
3402 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3403 Node* thread = ideal.thread();
3404 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3405 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3406
3407 sync_kit(ideal);
3408 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3409 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3410
3411 ideal.sync_kit(this);
3412 } ideal.end_if();
3413 final_sync(ideal);
3414
3415 return true;
3416 }
3417
3418 // Always update the is_disable_suspend bit.
3419 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3420 if (!DoJVMTIVirtualThreadTransitions) {
3421 return true;
3422 }
3423 IdealKit ideal(this);
3424
3425 {
3426 // unconditionally update the is_disable_suspend bit in current JavaThread
3427 Node* thread = ideal.thread();
3428 Node* arg = _gvn.transform(argument(0)); // argument for notification
3429 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3430 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3431
3432 sync_kit(ideal);
3433 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3434 ideal.sync_kit(this);
3435 }
3436 final_sync(ideal);
3437
3438 return true;
3439 }
3440
3441 #endif // INCLUDE_JVMTI
3442
3443 #ifdef JFR_HAVE_INTRINSICS
3444
3445 /**
3446 * if oop->klass != null
3447 * // normal class
3448 * epoch = _epoch_state ? 2 : 1
3449 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3450 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3451 * }
3452 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3453 * else
3454 * // primitive class
3455 * if oop->array_klass != null
3456 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3457 * else
3458 * id = LAST_TYPE_ID + 1 // void class path
3459 * if (!signaled)
3460 * signaled = true
3461 */
3462 bool LibraryCallKit::inline_native_classID() {
3463 Node* cls = argument(0);
3464
3465 IdealKit ideal(this);
3466 #define __ ideal.
3467 IdealVariable result(ideal); __ declarations_done();
3468 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3469 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3470 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3471
3472
3473 __ if_then(kls, BoolTest::ne, null()); {
3474 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3475 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3476
3477 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3478 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3479 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3480 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3481 mask = _gvn.transform(new OrLNode(mask, epoch));
3482 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3483
3484 float unlikely = PROB_UNLIKELY(0.999);
3485 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3486 sync_kit(ideal);
3487 make_runtime_call(RC_LEAF,
3488 OptoRuntime::class_id_load_barrier_Type(),
3489 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3490 "class id load barrier",
3491 TypePtr::BOTTOM,
3492 kls);
3493 ideal.sync_kit(this);
3494 } __ end_if();
3495
3496 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3497 } __ else_(); {
3498 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3499 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3500 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3501 __ if_then(array_kls, BoolTest::ne, null()); {
3502 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3503 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3504 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3505 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3506 } __ else_(); {
3507 // void class case
3508 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3509 } __ end_if();
3510
3511 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3512 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3513 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3514 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3515 } __ end_if();
3516 } __ end_if();
3517
3518 final_sync(ideal);
3519 set_result(ideal.value(result));
3520 #undef __
3521 return true;
3522 }
3523
3524 //------------------------inline_native_jvm_commit------------------
3525 bool LibraryCallKit::inline_native_jvm_commit() {
3526 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3527
3528 // Save input memory and i_o state.
3529 Node* input_memory_state = reset_memory();
3530 set_all_memory(input_memory_state);
3531 Node* input_io_state = i_o();
3532
3533 // TLS.
3534 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3535 // Jfr java buffer.
3536 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3537 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3538 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3539
3540 // Load the current value of the notified field in the JfrThreadLocal.
3541 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3542 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3543
3544 // Test for notification.
3545 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3546 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3547 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3548
3549 // True branch, is notified.
3550 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3551 set_control(is_notified);
3552
3553 // Reset notified state.
3554 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3555 Node* notified_reset_memory = reset_memory();
3556
3557 // 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.
3558 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3559 // Convert the machine-word to a long.
3560 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3561
3562 // False branch, not notified.
3563 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3564 set_control(not_notified);
3565 set_all_memory(input_memory_state);
3566
3567 // Arg is the next position as a long.
3568 Node* arg = argument(0);
3569 // Convert long to machine-word.
3570 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3571
3572 // Store the next_position to the underlying jfr java buffer.
3573 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3574
3575 Node* commit_memory = reset_memory();
3576 set_all_memory(commit_memory);
3577
3578 // 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.
3579 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3580 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3581 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3582
3583 // And flags with lease constant.
3584 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3585
3586 // Branch on lease to conditionalize returning the leased java buffer.
3587 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3588 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3589 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3590
3591 // False branch, not a lease.
3592 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3593
3594 // True branch, is lease.
3595 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3596 set_control(is_lease);
3597
3598 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3599 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3600 OptoRuntime::void_void_Type(),
3601 SharedRuntime::jfr_return_lease(),
3602 "return_lease", TypePtr::BOTTOM);
3603 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3604
3605 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3606 record_for_igvn(lease_compare_rgn);
3607 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3608 record_for_igvn(lease_compare_mem);
3609 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3610 record_for_igvn(lease_compare_io);
3611 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3612 record_for_igvn(lease_result_value);
3613
3614 // Update control and phi nodes.
3615 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3616 lease_compare_rgn->init_req(_false_path, not_lease);
3617
3618 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3619 lease_compare_mem->init_req(_false_path, commit_memory);
3620
3621 lease_compare_io->init_req(_true_path, i_o());
3622 lease_compare_io->init_req(_false_path, input_io_state);
3623
3624 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3625 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3626
3627 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3628 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3629 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3630 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3631
3632 // Update control and phi nodes.
3633 result_rgn->init_req(_true_path, is_notified);
3634 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3635
3636 result_mem->init_req(_true_path, notified_reset_memory);
3637 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3638
3639 result_io->init_req(_true_path, input_io_state);
3640 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3641
3642 result_value->init_req(_true_path, current_pos);
3643 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3644
3645 // Set output state.
3646 set_control(_gvn.transform(result_rgn));
3647 set_all_memory(_gvn.transform(result_mem));
3648 set_i_o(_gvn.transform(result_io));
3649 set_result(result_rgn, result_value);
3650 return true;
3651 }
3652
3653 /*
3654 * The intrinsic is a model of this pseudo-code:
3655 *
3656 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3657 * jobject h_event_writer = tl->java_event_writer();
3658 * if (h_event_writer == nullptr) {
3659 * return nullptr;
3660 * }
3661 * oop threadObj = Thread::threadObj();
3662 * oop vthread = java_lang_Thread::vthread(threadObj);
3663 * traceid tid;
3664 * bool pinVirtualThread;
3665 * bool excluded;
3666 * if (vthread != threadObj) { // i.e. current thread is virtual
3667 * tid = java_lang_Thread::tid(vthread);
3668 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3669 * pinVirtualThread = VMContinuations;
3670 * excluded = vthread_epoch_raw & excluded_mask;
3671 * if (!excluded) {
3672 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3673 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3674 * if (vthread_epoch != current_epoch) {
3675 * write_checkpoint();
3676 * }
3677 * }
3678 * } else {
3679 * tid = java_lang_Thread::tid(threadObj);
3680 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3681 * pinVirtualThread = false;
3682 * excluded = thread_epoch_raw & excluded_mask;
3683 * }
3684 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3685 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3686 * if (tid_in_event_writer != tid) {
3687 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3688 * setField(event_writer, "excluded", excluded);
3689 * setField(event_writer, "threadID", tid);
3690 * }
3691 * return event_writer
3692 */
3693 bool LibraryCallKit::inline_native_getEventWriter() {
3694 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3695
3696 // Save input memory and i_o state.
3697 Node* input_memory_state = reset_memory();
3698 set_all_memory(input_memory_state);
3699 Node* input_io_state = i_o();
3700
3701 // The most significant bit of the u2 is used to denote thread exclusion
3702 Node* excluded_shift = _gvn.intcon(15);
3703 Node* excluded_mask = _gvn.intcon(1 << 15);
3704 // The epoch generation is the range [1-32767]
3705 Node* epoch_mask = _gvn.intcon(32767);
3706
3707 // TLS
3708 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3709
3710 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3711 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3712
3713 // Load the eventwriter jobject handle.
3714 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3715
3716 // Null check the jobject handle.
3717 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3718 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3719 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3720
3721 // False path, jobj is null.
3722 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3723
3724 // True path, jobj is not null.
3725 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3726
3727 set_control(jobj_is_not_null);
3728
3729 // Load the threadObj for the CarrierThread.
3730 Node* threadObj = generate_current_thread(tls_ptr);
3731
3732 // Load the vthread.
3733 Node* vthread = generate_virtual_thread(tls_ptr);
3734
3735 // If vthread != threadObj, this is a virtual thread.
3736 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3737 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3738 IfNode* iff_vthread_not_equal_threadObj =
3739 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3740
3741 // False branch, fallback to threadObj.
3742 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3743 set_control(vthread_equal_threadObj);
3744
3745 // Load the tid field from the vthread object.
3746 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3747
3748 // Load the raw epoch value from the threadObj.
3749 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3750 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3751 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3752 TypeInt::CHAR, T_CHAR,
3753 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3754
3755 // Mask off the excluded information from the epoch.
3756 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3757
3758 // True branch, this is a virtual thread.
3759 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3760 set_control(vthread_not_equal_threadObj);
3761
3762 // Load the tid field from the vthread object.
3763 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3764
3765 // Continuation support determines if a virtual thread should be pinned.
3766 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3767 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3768
3769 // Load the raw epoch value from the vthread.
3770 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3771 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3772 TypeInt::CHAR, T_CHAR,
3773 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3774
3775 // Mask off the excluded information from the epoch.
3776 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3777
3778 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3779 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3780 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3781 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3782
3783 // False branch, vthread is excluded, no need to write epoch info.
3784 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3785
3786 // True branch, vthread is included, update epoch info.
3787 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3788 set_control(included);
3789
3790 // Get epoch value.
3791 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3792
3793 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3794 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3795 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3796
3797 // Compare the epoch in the vthread to the current epoch generation.
3798 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3799 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3800 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3801
3802 // False path, epoch is equal, checkpoint information is valid.
3803 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3804
3805 // True path, epoch is not equal, write a checkpoint for the vthread.
3806 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3807
3808 set_control(epoch_is_not_equal);
3809
3810 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3811 // The call also updates the native thread local thread id and the vthread with the current epoch.
3812 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3813 OptoRuntime::jfr_write_checkpoint_Type(),
3814 SharedRuntime::jfr_write_checkpoint(),
3815 "write_checkpoint", TypePtr::BOTTOM);
3816 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3817
3818 // vthread epoch != current epoch
3819 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3820 record_for_igvn(epoch_compare_rgn);
3821 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3822 record_for_igvn(epoch_compare_mem);
3823 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3824 record_for_igvn(epoch_compare_io);
3825
3826 // Update control and phi nodes.
3827 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3828 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3829 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3830 epoch_compare_mem->init_req(_false_path, input_memory_state);
3831 epoch_compare_io->init_req(_true_path, i_o());
3832 epoch_compare_io->init_req(_false_path, input_io_state);
3833
3834 // excluded != true
3835 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3836 record_for_igvn(exclude_compare_rgn);
3837 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3838 record_for_igvn(exclude_compare_mem);
3839 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3840 record_for_igvn(exclude_compare_io);
3841
3842 // Update control and phi nodes.
3843 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3844 exclude_compare_rgn->init_req(_false_path, excluded);
3845 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3846 exclude_compare_mem->init_req(_false_path, input_memory_state);
3847 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3848 exclude_compare_io->init_req(_false_path, input_io_state);
3849
3850 // vthread != threadObj
3851 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3852 record_for_igvn(vthread_compare_rgn);
3853 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3854 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3855 record_for_igvn(vthread_compare_io);
3856 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3857 record_for_igvn(tid);
3858 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3859 record_for_igvn(exclusion);
3860 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3861 record_for_igvn(pinVirtualThread);
3862
3863 // Update control and phi nodes.
3864 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3865 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3866 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3867 vthread_compare_mem->init_req(_false_path, input_memory_state);
3868 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3869 vthread_compare_io->init_req(_false_path, input_io_state);
3870 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3871 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3872 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3873 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3874 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3875 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3876
3877 // Update branch state.
3878 set_control(_gvn.transform(vthread_compare_rgn));
3879 set_all_memory(_gvn.transform(vthread_compare_mem));
3880 set_i_o(_gvn.transform(vthread_compare_io));
3881
3882 // Load the event writer oop by dereferencing the jobject handle.
3883 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3884 assert(klass_EventWriter->is_loaded(), "invariant");
3885 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3886 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3887 const TypeOopPtr* const xtype = aklass->as_instance_type();
3888 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3889 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3890
3891 // Load the current thread id from the event writer object.
3892 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3893 // Get the field offset to, conditionally, store an updated tid value later.
3894 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3895 // Get the field offset to, conditionally, store an updated exclusion value later.
3896 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3897 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3898 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3899
3900 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3901 record_for_igvn(event_writer_tid_compare_rgn);
3902 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3903 record_for_igvn(event_writer_tid_compare_mem);
3904 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3905 record_for_igvn(event_writer_tid_compare_io);
3906
3907 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3908 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3909 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3910 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3911
3912 // False path, tids are the same.
3913 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3914
3915 // True path, tid is not equal, need to update the tid in the event writer.
3916 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3917 record_for_igvn(tid_is_not_equal);
3918
3919 // Store the pin state to the event writer.
3920 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3921
3922 // Store the exclusion state to the event writer.
3923 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3924 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3925
3926 // Store the tid to the event writer.
3927 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3928
3929 // Update control and phi nodes.
3930 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3931 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3932 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3933 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3934 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3935 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3936
3937 // Result of top level CFG, Memory, IO and Value.
3938 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3939 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3940 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3941 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3942
3943 // Result control.
3944 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3945 result_rgn->init_req(_false_path, jobj_is_null);
3946
3947 // Result memory.
3948 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3949 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3950
3951 // Result IO.
3952 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3953 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3954
3955 // Result value.
3956 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3957 result_value->init_req(_false_path, null()); // return null
3958
3959 // Set output state.
3960 set_control(_gvn.transform(result_rgn));
3961 set_all_memory(_gvn.transform(result_mem));
3962 set_i_o(_gvn.transform(result_io));
3963 set_result(result_rgn, result_value);
3964 return true;
3965 }
3966
3967 /*
3968 * The intrinsic is a model of this pseudo-code:
3969 *
3970 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3971 * if (carrierThread != thread) { // is virtual thread
3972 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3973 * bool excluded = vthread_epoch_raw & excluded_mask;
3974 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3975 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3976 * if (!excluded) {
3977 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3978 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3979 * }
3980 * Atomic::release_store(&tl->_vthread, true);
3981 * return;
3982 * }
3983 * Atomic::release_store(&tl->_vthread, false);
3984 */
3985 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3986 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3987
3988 Node* input_memory_state = reset_memory();
3989 set_all_memory(input_memory_state);
3990
3991 // The most significant bit of the u2 is used to denote thread exclusion
3992 Node* excluded_mask = _gvn.intcon(1 << 15);
3993 // The epoch generation is the range [1-32767]
3994 Node* epoch_mask = _gvn.intcon(32767);
3995
3996 Node* const carrierThread = generate_current_thread(jt);
3997 // If thread != carrierThread, this is a virtual thread.
3998 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3999 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4000 IfNode* iff_thread_not_equal_carrierThread =
4001 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4002
4003 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4004
4005 // False branch, is carrierThread.
4006 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4007 // Store release
4008 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4009
4010 set_all_memory(input_memory_state);
4011
4012 // True branch, is virtual thread.
4013 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4014 set_control(thread_not_equal_carrierThread);
4015
4016 // Load the raw epoch value from the vthread.
4017 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4018 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4019 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4020
4021 // Mask off the excluded information from the epoch.
4022 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4023
4024 // Load the tid field from the thread.
4025 Node* tid = load_field_from_object(thread, "tid", "J");
4026
4027 // Store the vthread tid to the jfr thread local.
4028 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4029 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4030
4031 // Branch is_excluded to conditionalize updating the epoch .
4032 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4033 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4034 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4035
4036 // True branch, vthread is excluded, no need to write epoch info.
4037 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4038 set_control(excluded);
4039 Node* vthread_is_excluded = _gvn.intcon(1);
4040
4041 // False branch, vthread is included, update epoch info.
4042 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4043 set_control(included);
4044 Node* vthread_is_included = _gvn.intcon(0);
4045
4046 // Get epoch value.
4047 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4048
4049 // Store the vthread epoch to the jfr thread local.
4050 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4051 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4052
4053 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4054 record_for_igvn(excluded_rgn);
4055 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4056 record_for_igvn(excluded_mem);
4057 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4058 record_for_igvn(exclusion);
4059
4060 // Merge the excluded control and memory.
4061 excluded_rgn->init_req(_true_path, excluded);
4062 excluded_rgn->init_req(_false_path, included);
4063 excluded_mem->init_req(_true_path, tid_memory);
4064 excluded_mem->init_req(_false_path, included_memory);
4065 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4066 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4067
4068 // Set intermediate state.
4069 set_control(_gvn.transform(excluded_rgn));
4070 set_all_memory(excluded_mem);
4071
4072 // Store the vthread exclusion state to the jfr thread local.
4073 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4074 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4075
4076 // Store release
4077 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4078
4079 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4080 record_for_igvn(thread_compare_rgn);
4081 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4082 record_for_igvn(thread_compare_mem);
4083 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4084 record_for_igvn(vthread);
4085
4086 // Merge the thread_compare control and memory.
4087 thread_compare_rgn->init_req(_true_path, control());
4088 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4089 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4090 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4091
4092 // Set output state.
4093 set_control(_gvn.transform(thread_compare_rgn));
4094 set_all_memory(_gvn.transform(thread_compare_mem));
4095 }
4096
4097 #endif // JFR_HAVE_INTRINSICS
4098
4099 //------------------------inline_native_currentCarrierThread------------------
4100 bool LibraryCallKit::inline_native_currentCarrierThread() {
4101 Node* junk = nullptr;
4102 set_result(generate_current_thread(junk));
4103 return true;
4104 }
4105
4106 //------------------------inline_native_currentThread------------------
4107 bool LibraryCallKit::inline_native_currentThread() {
4108 Node* junk = nullptr;
4109 set_result(generate_virtual_thread(junk));
4110 return true;
4111 }
4112
4113 //------------------------inline_native_setVthread------------------
4114 bool LibraryCallKit::inline_native_setCurrentThread() {
4115 assert(C->method()->changes_current_thread(),
4116 "method changes current Thread but is not annotated ChangesCurrentThread");
4117 Node* arr = argument(1);
4118 Node* thread = _gvn.transform(new ThreadLocalNode());
4119 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4120 Node* thread_obj_handle
4121 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4122 thread_obj_handle = _gvn.transform(thread_obj_handle);
4123 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4124 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4125
4126 // Change the _monitor_owner_id of the JavaThread
4127 Node* tid = load_field_from_object(arr, "tid", "J");
4128 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4129 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4130
4131 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4132 return true;
4133 }
4134
4135 const Type* LibraryCallKit::scopedValueCache_type() {
4136 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4137 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4138 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
4139
4140 // Because we create the scopedValue cache lazily we have to make the
4141 // type of the result BotPTR.
4142 bool xk = etype->klass_is_exact();
4143 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4144 return objects_type;
4145 }
4146
4147 Node* LibraryCallKit::scopedValueCache_helper() {
4148 Node* thread = _gvn.transform(new ThreadLocalNode());
4149 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4150 // We cannot use immutable_memory() because we might flip onto a
4151 // different carrier thread, at which point we'll need to use that
4152 // carrier thread's cache.
4153 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4154 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4155 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4156 }
4157
4158 //------------------------inline_native_scopedValueCache------------------
4159 bool LibraryCallKit::inline_native_scopedValueCache() {
4160 Node* cache_obj_handle = scopedValueCache_helper();
4161 const Type* objects_type = scopedValueCache_type();
4162 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4163
4164 return true;
4165 }
4166
4167 //------------------------inline_native_setScopedValueCache------------------
4168 bool LibraryCallKit::inline_native_setScopedValueCache() {
4169 Node* arr = argument(0);
4170 Node* cache_obj_handle = scopedValueCache_helper();
4171 const Type* objects_type = scopedValueCache_type();
4172
4173 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4174 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4175
4176 return true;
4177 }
4178
4179 //------------------------inline_native_Continuation_pin and unpin-----------
4180
4181 // Shared implementation routine for both pin and unpin.
4182 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4183 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4184
4185 // Save input memory.
4186 Node* input_memory_state = reset_memory();
4187 set_all_memory(input_memory_state);
4188
4189 // TLS
4190 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4191 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4192 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4193
4194 // Null check the last continuation object.
4195 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4196 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4197 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4198
4199 // False path, last continuation is null.
4200 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4201
4202 // True path, last continuation is not null.
4203 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4204
4205 set_control(continuation_is_not_null);
4206
4207 // Load the pin count from the last continuation.
4208 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4209 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4210
4211 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4212 Node* pin_count_rhs;
4213 if (unpin) {
4214 pin_count_rhs = _gvn.intcon(0);
4215 } else {
4216 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4217 }
4218 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4219 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4220 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4221
4222 // True branch, pin count over/underflow.
4223 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4224 {
4225 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4226 // which will throw IllegalStateException for pin count over/underflow.
4227 // No memory changed so far - we can use memory create by reset_memory()
4228 // at the beginning of this intrinsic. No need to call reset_memory() again.
4229 PreserveJVMState pjvms(this);
4230 set_control(pin_count_over_underflow);
4231 uncommon_trap(Deoptimization::Reason_intrinsic,
4232 Deoptimization::Action_none);
4233 assert(stopped(), "invariant");
4234 }
4235
4236 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4237 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4238 set_control(valid_pin_count);
4239
4240 Node* next_pin_count;
4241 if (unpin) {
4242 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4243 } else {
4244 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4245 }
4246
4247 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4248
4249 // Result of top level CFG and Memory.
4250 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4251 record_for_igvn(result_rgn);
4252 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4253 record_for_igvn(result_mem);
4254
4255 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4256 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4257 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4258 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4259
4260 // Set output state.
4261 set_control(_gvn.transform(result_rgn));
4262 set_all_memory(_gvn.transform(result_mem));
4263
4264 return true;
4265 }
4266
4267 //-----------------------load_klass_from_mirror_common-------------------------
4268 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4269 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4270 // and branch to the given path on the region.
4271 // If never_see_null, take an uncommon trap on null, so we can optimistically
4272 // compile for the non-null case.
4273 // If the region is null, force never_see_null = true.
4274 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4275 bool never_see_null,
4276 RegionNode* region,
4277 int null_path,
4278 int offset) {
4279 if (region == nullptr) never_see_null = true;
4280 Node* p = basic_plus_adr(mirror, offset);
4281 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4282 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4283 Node* null_ctl = top();
4284 kls = null_check_oop(kls, &null_ctl, never_see_null);
4285 if (region != nullptr) {
4286 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4287 region->init_req(null_path, null_ctl);
4288 } else {
4289 assert(null_ctl == top(), "no loose ends");
4290 }
4291 return kls;
4292 }
4293
4294 //--------------------(inline_native_Class_query helpers)---------------------
4295 // Use this for JVM_ACC_INTERFACE.
4296 // Fall through if (mods & mask) == bits, take the guard otherwise.
4297 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4298 ByteSize offset, const Type* type, BasicType bt) {
4299 // Branch around if the given klass has the given modifier bit set.
4300 // Like generate_guard, adds a new path onto the region.
4301 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4302 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4303 Node* mask = intcon(modifier_mask);
4304 Node* bits = intcon(modifier_bits);
4305 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4306 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4307 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4308 return generate_fair_guard(bol, region);
4309 }
4310
4311 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4312 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4313 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4314 }
4315
4316 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4317 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4318 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4319 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4320 }
4321
4322 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4323 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4324 }
4325
4326 //-------------------------inline_native_Class_query-------------------
4327 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4328 const Type* return_type = TypeInt::BOOL;
4329 Node* prim_return_value = top(); // what happens if it's a primitive class?
4330 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4331 bool expect_prim = false; // most of these guys expect to work on refs
4332
4333 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4334
4335 Node* mirror = argument(0);
4336 Node* obj = top();
4337
4338 switch (id) {
4339 case vmIntrinsics::_isInstance:
4340 // nothing is an instance of a primitive type
4341 prim_return_value = intcon(0);
4342 obj = argument(1);
4343 break;
4344 case vmIntrinsics::_isHidden:
4345 prim_return_value = intcon(0);
4346 break;
4347 case vmIntrinsics::_getSuperclass:
4348 prim_return_value = null();
4349 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4350 break;
4351 case vmIntrinsics::_getClassAccessFlags:
4352 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4353 return_type = TypeInt::CHAR;
4354 break;
4355 default:
4356 fatal_unexpected_iid(id);
4357 break;
4358 }
4359
4360 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4361 if (mirror_con == nullptr) return false; // cannot happen?
4362
4363 #ifndef PRODUCT
4364 if (C->print_intrinsics() || C->print_inlining()) {
4365 ciType* k = mirror_con->java_mirror_type();
4366 if (k) {
4367 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4368 k->print_name();
4369 tty->cr();
4370 }
4371 }
4372 #endif
4373
4374 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4375 RegionNode* region = new RegionNode(PATH_LIMIT);
4376 record_for_igvn(region);
4377 PhiNode* phi = new PhiNode(region, return_type);
4378
4379 // The mirror will never be null of Reflection.getClassAccessFlags, however
4380 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4381 // if it is. See bug 4774291.
4382
4383 // For Reflection.getClassAccessFlags(), the null check occurs in
4384 // the wrong place; see inline_unsafe_access(), above, for a similar
4385 // situation.
4386 mirror = null_check(mirror);
4387 // If mirror or obj is dead, only null-path is taken.
4388 if (stopped()) return true;
4389
4390 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4391
4392 // Now load the mirror's klass metaobject, and null-check it.
4393 // Side-effects region with the control path if the klass is null.
4394 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4395 // If kls is null, we have a primitive mirror.
4396 phi->init_req(_prim_path, prim_return_value);
4397 if (stopped()) { set_result(region, phi); return true; }
4398 bool safe_for_replace = (region->in(_prim_path) == top());
4399
4400 Node* p; // handy temp
4401 Node* null_ctl;
4402
4403 // Now that we have the non-null klass, we can perform the real query.
4404 // For constant classes, the query will constant-fold in LoadNode::Value.
4405 Node* query_value = top();
4406 switch (id) {
4407 case vmIntrinsics::_isInstance:
4408 // nothing is an instance of a primitive type
4409 query_value = gen_instanceof(obj, kls, safe_for_replace);
4410 break;
4411
4412 case vmIntrinsics::_isHidden:
4413 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4414 if (generate_hidden_class_guard(kls, region) != nullptr)
4415 // A guard was added. If the guard is taken, it was an hidden class.
4416 phi->add_req(intcon(1));
4417 // If we fall through, it's a plain class.
4418 query_value = intcon(0);
4419 break;
4420
4421
4422 case vmIntrinsics::_getSuperclass:
4423 // The rules here are somewhat unfortunate, but we can still do better
4424 // with random logic than with a JNI call.
4425 // Interfaces store null or Object as _super, but must report null.
4426 // Arrays store an intermediate super as _super, but must report Object.
4427 // Other types can report the actual _super.
4428 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4429 if (generate_interface_guard(kls, region) != nullptr)
4430 // A guard was added. If the guard is taken, it was an interface.
4431 phi->add_req(null());
4432 if (generate_array_guard(kls, region) != nullptr)
4433 // A guard was added. If the guard is taken, it was an array.
4434 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4435 // If we fall through, it's a plain class. Get its _super.
4436 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4437 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4438 null_ctl = top();
4439 kls = null_check_oop(kls, &null_ctl);
4440 if (null_ctl != top()) {
4441 // If the guard is taken, Object.superClass is null (both klass and mirror).
4442 region->add_req(null_ctl);
4443 phi ->add_req(null());
4444 }
4445 if (!stopped()) {
4446 query_value = load_mirror_from_klass(kls);
4447 }
4448 break;
4449
4450 case vmIntrinsics::_getClassAccessFlags:
4451 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4452 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4453 break;
4454
4455 default:
4456 fatal_unexpected_iid(id);
4457 break;
4458 }
4459
4460 // Fall-through is the normal case of a query to a real class.
4461 phi->init_req(1, query_value);
4462 region->init_req(1, control());
4463
4464 C->set_has_split_ifs(true); // Has chance for split-if optimization
4465 set_result(region, phi);
4466 return true;
4467 }
4468
4469
4470 //-------------------------inline_Class_cast-------------------
4471 bool LibraryCallKit::inline_Class_cast() {
4472 Node* mirror = argument(0); // Class
4473 Node* obj = argument(1);
4474 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4475 if (mirror_con == nullptr) {
4476 return false; // dead path (mirror->is_top()).
4477 }
4478 if (obj == nullptr || obj->is_top()) {
4479 return false; // dead path
4480 }
4481 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4482
4483 // First, see if Class.cast() can be folded statically.
4484 // java_mirror_type() returns non-null for compile-time Class constants.
4485 bool is_null_free_array = false;
4486 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4487 if (tm != nullptr && tm->is_klass() &&
4488 tp != nullptr) {
4489 if (!tp->is_loaded()) {
4490 // Don't use intrinsic when class is not loaded.
4491 return false;
4492 } else {
4493 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4494 if (is_null_free_array) {
4495 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4496 }
4497 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4498 if (static_res == Compile::SSC_always_true) {
4499 // isInstance() is true - fold the code.
4500 set_result(obj);
4501 return true;
4502 } else if (static_res == Compile::SSC_always_false) {
4503 // Don't use intrinsic, have to throw ClassCastException.
4504 // If the reference is null, the non-intrinsic bytecode will
4505 // be optimized appropriately.
4506 return false;
4507 }
4508 }
4509 }
4510
4511 // Bailout intrinsic and do normal inlining if exception path is frequent.
4512 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4513 return false;
4514 }
4515
4516 // Generate dynamic checks.
4517 // Class.cast() is java implementation of _checkcast bytecode.
4518 // Do checkcast (Parse::do_checkcast()) optimizations here.
4519
4520 mirror = null_check(mirror);
4521 // If mirror is dead, only null-path is taken.
4522 if (stopped()) {
4523 return true;
4524 }
4525
4526 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4527 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4528 RegionNode* region = new RegionNode(PATH_LIMIT);
4529 record_for_igvn(region);
4530
4531 // Now load the mirror's klass metaobject, and null-check it.
4532 // If kls is null, we have a primitive mirror and
4533 // nothing is an instance of a primitive type.
4534 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4535
4536 Node* res = top();
4537 Node* io = i_o();
4538 Node* mem = merged_memory();
4539 if (!stopped()) {
4540
4541 Node* bad_type_ctrl = top();
4542 // Do checkcast optimizations.
4543 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4544 region->init_req(_bad_type_path, bad_type_ctrl);
4545 }
4546 if (region->in(_prim_path) != top() ||
4547 region->in(_bad_type_path) != top() ||
4548 region->in(_npe_path) != top()) {
4549 // Let Interpreter throw ClassCastException.
4550 PreserveJVMState pjvms(this);
4551 set_control(_gvn.transform(region));
4552 // Set IO and memory because gen_checkcast may override them when buffering inline types
4553 set_i_o(io);
4554 set_all_memory(mem);
4555 uncommon_trap(Deoptimization::Reason_intrinsic,
4556 Deoptimization::Action_maybe_recompile);
4557 }
4558 if (!stopped()) {
4559 set_result(res);
4560 }
4561 return true;
4562 }
4563
4564
4565 //--------------------------inline_native_subtype_check------------------------
4566 // This intrinsic takes the JNI calls out of the heart of
4567 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4568 bool LibraryCallKit::inline_native_subtype_check() {
4569 // Pull both arguments off the stack.
4570 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4571 args[0] = argument(0);
4572 args[1] = argument(1);
4573 Node* klasses[2]; // corresponding Klasses: superk, subk
4574 klasses[0] = klasses[1] = top();
4575
4576 enum {
4577 // A full decision tree on {superc is prim, subc is prim}:
4578 _prim_0_path = 1, // {P,N} => false
4579 // {P,P} & superc!=subc => false
4580 _prim_same_path, // {P,P} & superc==subc => true
4581 _prim_1_path, // {N,P} => false
4582 _ref_subtype_path, // {N,N} & subtype check wins => true
4583 _both_ref_path, // {N,N} & subtype check loses => false
4584 PATH_LIMIT
4585 };
4586
4587 RegionNode* region = new RegionNode(PATH_LIMIT);
4588 RegionNode* prim_region = new RegionNode(2);
4589 Node* phi = new PhiNode(region, TypeInt::BOOL);
4590 record_for_igvn(region);
4591 record_for_igvn(prim_region);
4592
4593 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4594 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4595 int class_klass_offset = java_lang_Class::klass_offset();
4596
4597 // First null-check both mirrors and load each mirror's klass metaobject.
4598 int which_arg;
4599 for (which_arg = 0; which_arg <= 1; which_arg++) {
4600 Node* arg = args[which_arg];
4601 arg = null_check(arg);
4602 if (stopped()) break;
4603 args[which_arg] = arg;
4604
4605 Node* p = basic_plus_adr(arg, class_klass_offset);
4606 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4607 klasses[which_arg] = _gvn.transform(kls);
4608 }
4609
4610 // Having loaded both klasses, test each for null.
4611 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4612 for (which_arg = 0; which_arg <= 1; which_arg++) {
4613 Node* kls = klasses[which_arg];
4614 Node* null_ctl = top();
4615 kls = null_check_oop(kls, &null_ctl, never_see_null);
4616 if (which_arg == 0) {
4617 prim_region->init_req(1, null_ctl);
4618 } else {
4619 region->init_req(_prim_1_path, null_ctl);
4620 }
4621 if (stopped()) break;
4622 klasses[which_arg] = kls;
4623 }
4624
4625 if (!stopped()) {
4626 // now we have two reference types, in klasses[0..1]
4627 Node* subk = klasses[1]; // the argument to isAssignableFrom
4628 Node* superk = klasses[0]; // the receiver
4629 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4630 region->set_req(_ref_subtype_path, control());
4631 }
4632
4633 // If both operands are primitive (both klasses null), then
4634 // we must return true when they are identical primitives.
4635 // It is convenient to test this after the first null klass check.
4636 // This path is also used if superc is a value mirror.
4637 set_control(_gvn.transform(prim_region));
4638 if (!stopped()) {
4639 // Since superc is primitive, make a guard for the superc==subc case.
4640 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4641 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4642 generate_fair_guard(bol_eq, region);
4643 if (region->req() == PATH_LIMIT+1) {
4644 // A guard was added. If the added guard is taken, superc==subc.
4645 region->swap_edges(PATH_LIMIT, _prim_same_path);
4646 region->del_req(PATH_LIMIT);
4647 }
4648 region->set_req(_prim_0_path, control()); // Not equal after all.
4649 }
4650
4651 // these are the only paths that produce 'true':
4652 phi->set_req(_prim_same_path, intcon(1));
4653 phi->set_req(_ref_subtype_path, intcon(1));
4654
4655 // pull together the cases:
4656 assert(region->req() == PATH_LIMIT, "sane region");
4657 for (uint i = 1; i < region->req(); i++) {
4658 Node* ctl = region->in(i);
4659 if (ctl == nullptr || ctl == top()) {
4660 region->set_req(i, top());
4661 phi ->set_req(i, top());
4662 } else if (phi->in(i) == nullptr) {
4663 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4664 }
4665 }
4666
4667 set_control(_gvn.transform(region));
4668 set_result(_gvn.transform(phi));
4669 return true;
4670 }
4671
4672 //---------------------generate_array_guard_common------------------------
4673 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4674
4675 if (stopped()) {
4676 return nullptr;
4677 }
4678
4679 // Like generate_guard, adds a new path onto the region.
4680 jint layout_con = 0;
4681 Node* layout_val = get_layout_helper(kls, layout_con);
4682 if (layout_val == nullptr) {
4683 bool query = 0;
4684 switch(kind) {
4685 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4686 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4687 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4688 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4689 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4690 default:
4691 ShouldNotReachHere();
4692 }
4693 if (!query) {
4694 return nullptr; // never a branch
4695 } else { // always a branch
4696 Node* always_branch = control();
4697 if (region != nullptr)
4698 region->add_req(always_branch);
4699 set_control(top());
4700 return always_branch;
4701 }
4702 }
4703 unsigned int value = 0;
4704 BoolTest::mask btest = BoolTest::illegal;
4705 switch(kind) {
4706 case ObjectArray:
4707 case NonObjectArray: {
4708 value = Klass::_lh_array_tag_obj_value;
4709 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4710 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4711 break;
4712 }
4713 case TypeArray: {
4714 value = Klass::_lh_array_tag_type_value;
4715 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4716 btest = BoolTest::eq;
4717 break;
4718 }
4719 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4720 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4721 default:
4722 ShouldNotReachHere();
4723 }
4724 // Now test the correct condition.
4725 jint nval = (jint)value;
4726 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4727 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4728 Node* ctrl = generate_fair_guard(bol, region);
4729 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4730 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4731 // Keep track of the fact that 'obj' is an array to prevent
4732 // array specific accesses from floating above the guard.
4733 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4734 }
4735 return ctrl;
4736 }
4737
4738 // public static native Object[] newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4739 // public static native Object[] newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4740 // public static native Object[] newNullableAtomicArray(Class<?> componentType, int length);
4741 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4742 assert(null_free || atomic, "nullable implies atomic");
4743 Node* componentType = argument(0);
4744 Node* length = argument(1);
4745 Node* init_val = null_free ? argument(2) : nullptr;
4746
4747 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4748 if (tp != nullptr) {
4749 ciInstanceKlass* ik = tp->instance_klass();
4750 if (ik == C->env()->Class_klass()) {
4751 ciType* t = tp->java_mirror_type();
4752 if (t != nullptr && t->is_inlinetype()) {
4753 ciInlineKlass* vk = t->as_inline_klass();
4754 bool flat = vk->maybe_flat_in_array();
4755 if (flat && atomic) {
4756 // Only flat if we have a corresponding atomic layout
4757 flat = null_free ? vk->has_atomic_layout() : vk->has_nullable_atomic_layout();
4758 }
4759 // TODO 8350865 refactor
4760 if (flat && !atomic) {
4761 flat = vk->has_non_atomic_layout();
4762 }
4763
4764 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4765 if (UseZGC && null_free && !flat) {
4766 return false;
4767 }
4768
4769 ciArrayKlass* array_klass = ciArrayKlass::make(t, flat, null_free, atomic);
4770 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4771 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4772 if (null_free) {
4773 if (init_val->is_InlineType()) {
4774 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4775 // Zeroing is enough because the init value is the all-zero value
4776 init_val = nullptr;
4777 } else {
4778 init_val = init_val->as_InlineType()->buffer(this);
4779 }
4780 }
4781 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4782 }
4783 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4784 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4785 assert(arytype->is_null_free() == null_free, "inconsistency");
4786 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4787 assert(arytype->is_flat() == flat, "inconsistency");
4788 assert(arytype->is_aryptr()->is_not_flat() == !flat, "inconsistency");
4789 set_result(obj);
4790 return true;
4791 }
4792 }
4793 }
4794 }
4795 return false;
4796 }
4797
4798 //-----------------------inline_native_newArray--------------------------
4799 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4800 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4801 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4802 Node* mirror;
4803 Node* count_val;
4804 if (uninitialized) {
4805 null_check_receiver();
4806 mirror = argument(1);
4807 count_val = argument(2);
4808 } else {
4809 mirror = argument(0);
4810 count_val = argument(1);
4811 }
4812
4813 mirror = null_check(mirror);
4814 // If mirror or obj is dead, only null-path is taken.
4815 if (stopped()) return true;
4816
4817 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4818 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4819 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4820 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4821 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4822
4823 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4824 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4825 result_reg, _slow_path);
4826 Node* normal_ctl = control();
4827 Node* no_array_ctl = result_reg->in(_slow_path);
4828
4829 // Generate code for the slow case. We make a call to newArray().
4830 set_control(no_array_ctl);
4831 if (!stopped()) {
4832 // Either the input type is void.class, or else the
4833 // array klass has not yet been cached. Either the
4834 // ensuing call will throw an exception, or else it
4835 // will cache the array klass for next time.
4836 PreserveJVMState pjvms(this);
4837 CallJavaNode* slow_call = nullptr;
4838 if (uninitialized) {
4839 // Generate optimized virtual call (holder class 'Unsafe' is final)
4840 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4841 } else {
4842 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4843 }
4844 Node* slow_result = set_results_for_java_call(slow_call);
4845 // this->control() comes from set_results_for_java_call
4846 result_reg->set_req(_slow_path, control());
4847 result_val->set_req(_slow_path, slow_result);
4848 result_io ->set_req(_slow_path, i_o());
4849 result_mem->set_req(_slow_path, reset_memory());
4850 }
4851
4852 set_control(normal_ctl);
4853 if (!stopped()) {
4854 // Normal case: The array type has been cached in the java.lang.Class.
4855 // The following call works fine even if the array type is polymorphic.
4856 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4857 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4858 result_reg->init_req(_normal_path, control());
4859 result_val->init_req(_normal_path, obj);
4860 result_io ->init_req(_normal_path, i_o());
4861 result_mem->init_req(_normal_path, reset_memory());
4862
4863 if (uninitialized) {
4864 // Mark the allocation so that zeroing is skipped
4865 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4866 alloc->maybe_set_complete(&_gvn);
4867 }
4868 }
4869
4870 // Return the combined state.
4871 set_i_o( _gvn.transform(result_io) );
4872 set_all_memory( _gvn.transform(result_mem));
4873
4874 C->set_has_split_ifs(true); // Has chance for split-if optimization
4875 set_result(result_reg, result_val);
4876 return true;
4877 }
4878
4879 //----------------------inline_native_getLength--------------------------
4880 // public static native int java.lang.reflect.Array.getLength(Object array);
4881 bool LibraryCallKit::inline_native_getLength() {
4882 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4883
4884 Node* array = null_check(argument(0));
4885 // If array is dead, only null-path is taken.
4886 if (stopped()) return true;
4887
4888 // Deoptimize if it is a non-array.
4889 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4890
4891 if (non_array != nullptr) {
4892 PreserveJVMState pjvms(this);
4893 set_control(non_array);
4894 uncommon_trap(Deoptimization::Reason_intrinsic,
4895 Deoptimization::Action_maybe_recompile);
4896 }
4897
4898 // If control is dead, only non-array-path is taken.
4899 if (stopped()) return true;
4900
4901 // The works fine even if the array type is polymorphic.
4902 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4903 Node* result = load_array_length(array);
4904
4905 C->set_has_split_ifs(true); // Has chance for split-if optimization
4906 set_result(result);
4907 return true;
4908 }
4909
4910 //------------------------inline_array_copyOf----------------------------
4911 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4912 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4913 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4914 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4915
4916 // Get the arguments.
4917 Node* original = argument(0);
4918 Node* start = is_copyOfRange? argument(1): intcon(0);
4919 Node* end = is_copyOfRange? argument(2): argument(1);
4920 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4921
4922 Node* newcopy = nullptr;
4923
4924 // Set the original stack and the reexecute bit for the interpreter to reexecute
4925 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4926 { PreserveReexecuteState preexecs(this);
4927 jvms()->set_should_reexecute(true);
4928
4929 array_type_mirror = null_check(array_type_mirror);
4930 original = null_check(original);
4931
4932 // Check if a null path was taken unconditionally.
4933 if (stopped()) return true;
4934
4935 Node* orig_length = load_array_length(original);
4936
4937 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4938 klass_node = null_check(klass_node);
4939
4940 RegionNode* bailout = new RegionNode(1);
4941 record_for_igvn(bailout);
4942
4943 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4944 // Bail out if that is so.
4945 // Inline type array may have object field that would require a
4946 // write barrier. Conservatively, go to slow path.
4947 // TODO 8251971: Optimize for the case when flat src/dst are later found
4948 // to not contain oops (i.e., move this check to the macro expansion phase).
4949 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4950 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4951 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4952 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4953 // Can src array be flat and contain oops?
4954 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4955 // Can dest array be flat and contain oops?
4956 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4957 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4958 if (not_objArray != nullptr) {
4959 // Improve the klass node's type from the new optimistic assumption:
4960 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4961 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4962 Node* cast = new CastPPNode(control(), klass_node, akls);
4963 klass_node = _gvn.transform(cast);
4964 }
4965
4966 // Bail out if either start or end is negative.
4967 generate_negative_guard(start, bailout, &start);
4968 generate_negative_guard(end, bailout, &end);
4969
4970 Node* length = end;
4971 if (_gvn.type(start) != TypeInt::ZERO) {
4972 length = _gvn.transform(new SubINode(end, start));
4973 }
4974
4975 // Bail out if length is negative (i.e., if start > end).
4976 // Without this the new_array would throw
4977 // NegativeArraySizeException but IllegalArgumentException is what
4978 // should be thrown
4979 generate_negative_guard(length, bailout, &length);
4980
4981 // Handle inline type arrays
4982 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4983 if (!stopped()) {
4984 // TODO JDK-8329224
4985 if (!orig_t->is_null_free()) {
4986 // Not statically known to be null free, add a check
4987 generate_fair_guard(null_free_array_test(original), bailout);
4988 }
4989 orig_t = _gvn.type(original)->isa_aryptr();
4990 if (orig_t != nullptr && orig_t->is_flat()) {
4991 // Src is flat, check that dest is flat as well
4992 if (exclude_flat) {
4993 // Dest can't be flat, bail out
4994 bailout->add_req(control());
4995 set_control(top());
4996 } else {
4997 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4998 }
4999 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5000 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5001 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5002 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5003 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5004 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5005 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5006 if (orig_t != nullptr) {
5007 orig_t = orig_t->cast_to_not_flat();
5008 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5009 }
5010 }
5011 if (!can_validate) {
5012 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5013 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5014 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5015 generate_fair_guard(flat_array_test(klass_node), bailout);
5016 generate_fair_guard(null_free_array_test(original), bailout);
5017 }
5018 }
5019
5020 // Bail out if start is larger than the original length
5021 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5022 generate_negative_guard(orig_tail, bailout, &orig_tail);
5023
5024 if (bailout->req() > 1) {
5025 PreserveJVMState pjvms(this);
5026 set_control(_gvn.transform(bailout));
5027 uncommon_trap(Deoptimization::Reason_intrinsic,
5028 Deoptimization::Action_maybe_recompile);
5029 }
5030
5031 if (!stopped()) {
5032 // How many elements will we copy from the original?
5033 // The answer is MinI(orig_tail, length).
5034 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5035
5036 // Generate a direct call to the right arraycopy function(s).
5037 // We know the copy is disjoint but we might not know if the
5038 // oop stores need checking.
5039 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5040 // This will fail a store-check if x contains any non-nulls.
5041
5042 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5043 // loads/stores but it is legal only if we're sure the
5044 // Arrays.copyOf would succeed. So we need all input arguments
5045 // to the copyOf to be validated, including that the copy to the
5046 // new array won't trigger an ArrayStoreException. That subtype
5047 // check can be optimized if we know something on the type of
5048 // the input array from type speculation.
5049 if (_gvn.type(klass_node)->singleton()) {
5050 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5051 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5052
5053 int test = C->static_subtype_check(superk, subk);
5054 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5055 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5056 if (t_original->speculative_type() != nullptr) {
5057 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5058 }
5059 }
5060 }
5061
5062 bool validated = false;
5063 // Reason_class_check rather than Reason_intrinsic because we
5064 // want to intrinsify even if this traps.
5065 if (can_validate) {
5066 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5067
5068 if (not_subtype_ctrl != top()) {
5069 PreserveJVMState pjvms(this);
5070 set_control(not_subtype_ctrl);
5071 uncommon_trap(Deoptimization::Reason_class_check,
5072 Deoptimization::Action_make_not_entrant);
5073 assert(stopped(), "Should be stopped");
5074 }
5075 validated = true;
5076 }
5077
5078 if (!stopped()) {
5079 newcopy = new_array(klass_node, length, 0); // no arguments to push
5080
5081 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5082 load_object_klass(original), klass_node);
5083 if (!is_copyOfRange) {
5084 ac->set_copyof(validated);
5085 } else {
5086 ac->set_copyofrange(validated);
5087 }
5088 Node* n = _gvn.transform(ac);
5089 if (n == ac) {
5090 ac->connect_outputs(this);
5091 } else {
5092 assert(validated, "shouldn't transform if all arguments not validated");
5093 set_all_memory(n);
5094 }
5095 }
5096 }
5097 } // original reexecute is set back here
5098
5099 C->set_has_split_ifs(true); // Has chance for split-if optimization
5100 if (!stopped()) {
5101 set_result(newcopy);
5102 }
5103 return true;
5104 }
5105
5106
5107 //----------------------generate_virtual_guard---------------------------
5108 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5109 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5110 RegionNode* slow_region) {
5111 ciMethod* method = callee();
5112 int vtable_index = method->vtable_index();
5113 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5114 "bad index %d", vtable_index);
5115 // Get the Method* out of the appropriate vtable entry.
5116 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5117 vtable_index*vtableEntry::size_in_bytes() +
5118 in_bytes(vtableEntry::method_offset());
5119 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5120 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5121
5122 // Compare the target method with the expected method (e.g., Object.hashCode).
5123 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5124
5125 Node* native_call = makecon(native_call_addr);
5126 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5127 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5128
5129 return generate_slow_guard(test_native, slow_region);
5130 }
5131
5132 //-----------------------generate_method_call----------------------------
5133 // Use generate_method_call to make a slow-call to the real
5134 // method if the fast path fails. An alternative would be to
5135 // use a stub like OptoRuntime::slow_arraycopy_Java.
5136 // This only works for expanding the current library call,
5137 // not another intrinsic. (E.g., don't use this for making an
5138 // arraycopy call inside of the copyOf intrinsic.)
5139 CallJavaNode*
5140 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5141 // When compiling the intrinsic method itself, do not use this technique.
5142 guarantee(callee() != C->method(), "cannot make slow-call to self");
5143
5144 ciMethod* method = callee();
5145 // ensure the JVMS we have will be correct for this call
5146 guarantee(method_id == method->intrinsic_id(), "must match");
5147
5148 const TypeFunc* tf = TypeFunc::make(method);
5149 if (res_not_null) {
5150 assert(tf->return_type() == T_OBJECT, "");
5151 const TypeTuple* range = tf->range_cc();
5152 const Type** fields = TypeTuple::fields(range->cnt());
5153 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5154 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5155 tf = TypeFunc::make(tf->domain_cc(), new_range);
5156 }
5157 CallJavaNode* slow_call;
5158 if (is_static) {
5159 assert(!is_virtual, "");
5160 slow_call = new CallStaticJavaNode(C, tf,
5161 SharedRuntime::get_resolve_static_call_stub(), method);
5162 } else if (is_virtual) {
5163 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5164 int vtable_index = Method::invalid_vtable_index;
5165 if (UseInlineCaches) {
5166 // Suppress the vtable call
5167 } else {
5168 // hashCode and clone are not a miranda methods,
5169 // so the vtable index is fixed.
5170 // No need to use the linkResolver to get it.
5171 vtable_index = method->vtable_index();
5172 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5173 "bad index %d", vtable_index);
5174 }
5175 slow_call = new CallDynamicJavaNode(tf,
5176 SharedRuntime::get_resolve_virtual_call_stub(),
5177 method, vtable_index);
5178 } else { // neither virtual nor static: opt_virtual
5179 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5180 slow_call = new CallStaticJavaNode(C, tf,
5181 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5182 slow_call->set_optimized_virtual(true);
5183 }
5184 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5185 // To be able to issue a direct call (optimized virtual or virtual)
5186 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5187 // about the method being invoked should be attached to the call site to
5188 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5189 slow_call->set_override_symbolic_info(true);
5190 }
5191 set_arguments_for_java_call(slow_call);
5192 set_edges_for_java_call(slow_call);
5193 return slow_call;
5194 }
5195
5196
5197 /**
5198 * Build special case code for calls to hashCode on an object. This call may
5199 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5200 * slightly different code.
5201 */
5202 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5203 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5204 assert(!(is_virtual && is_static), "either virtual, special, or static");
5205
5206 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5207
5208 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5209 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5210 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5211 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5212 Node* obj = argument(0);
5213
5214 // Don't intrinsify hashcode on inline types for now.
5215 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
5216 if (gvn().type(obj)->is_inlinetypeptr()) {
5217 return false;
5218 }
5219
5220 if (!is_static) {
5221 // Check for hashing null object
5222 obj = null_check_receiver();
5223 if (stopped()) return true; // unconditionally null
5224 result_reg->init_req(_null_path, top());
5225 result_val->init_req(_null_path, top());
5226 } else {
5227 // Do a null check, and return zero if null.
5228 // System.identityHashCode(null) == 0
5229 Node* null_ctl = top();
5230 obj = null_check_oop(obj, &null_ctl);
5231 result_reg->init_req(_null_path, null_ctl);
5232 result_val->init_req(_null_path, _gvn.intcon(0));
5233 }
5234
5235 // Unconditionally null? Then return right away.
5236 if (stopped()) {
5237 set_control( result_reg->in(_null_path));
5238 if (!stopped())
5239 set_result(result_val->in(_null_path));
5240 return true;
5241 }
5242
5243 // We only go to the fast case code if we pass a number of guards. The
5244 // paths which do not pass are accumulated in the slow_region.
5245 RegionNode* slow_region = new RegionNode(1);
5246 record_for_igvn(slow_region);
5247
5248 // If this is a virtual call, we generate a funny guard. We pull out
5249 // the vtable entry corresponding to hashCode() from the target object.
5250 // If the target method which we are calling happens to be the native
5251 // Object hashCode() method, we pass the guard. We do not need this
5252 // guard for non-virtual calls -- the caller is known to be the native
5253 // Object hashCode().
5254 if (is_virtual) {
5255 // After null check, get the object's klass.
5256 Node* obj_klass = load_object_klass(obj);
5257 generate_virtual_guard(obj_klass, slow_region);
5258 }
5259
5260 // Get the header out of the object, use LoadMarkNode when available
5261 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5262 // The control of the load must be null. Otherwise, the load can move before
5263 // the null check after castPP removal.
5264 Node* no_ctrl = nullptr;
5265 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5266
5267 if (!UseObjectMonitorTable) {
5268 // Test the header to see if it is safe to read w.r.t. locking.
5269 // This also serves as guard against inline types
5270 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5271 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5272 if (LockingMode == LM_LIGHTWEIGHT) {
5273 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5274 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5275 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5276
5277 generate_slow_guard(test_monitor, slow_region);
5278 } else {
5279 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5280 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5281 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5282
5283 generate_slow_guard(test_not_unlocked, slow_region);
5284 }
5285 }
5286
5287 // Get the hash value and check to see that it has been properly assigned.
5288 // We depend on hash_mask being at most 32 bits and avoid the use of
5289 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5290 // vm: see markWord.hpp.
5291 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5292 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5293 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5294 // This hack lets the hash bits live anywhere in the mark object now, as long
5295 // as the shift drops the relevant bits into the low 32 bits. Note that
5296 // Java spec says that HashCode is an int so there's no point in capturing
5297 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5298 hshifted_header = ConvX2I(hshifted_header);
5299 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5300
5301 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5302 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5303 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5304
5305 generate_slow_guard(test_assigned, slow_region);
5306
5307 Node* init_mem = reset_memory();
5308 // fill in the rest of the null path:
5309 result_io ->init_req(_null_path, i_o());
5310 result_mem->init_req(_null_path, init_mem);
5311
5312 result_val->init_req(_fast_path, hash_val);
5313 result_reg->init_req(_fast_path, control());
5314 result_io ->init_req(_fast_path, i_o());
5315 result_mem->init_req(_fast_path, init_mem);
5316
5317 // Generate code for the slow case. We make a call to hashCode().
5318 set_control(_gvn.transform(slow_region));
5319 if (!stopped()) {
5320 // No need for PreserveJVMState, because we're using up the present state.
5321 set_all_memory(init_mem);
5322 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5323 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5324 Node* slow_result = set_results_for_java_call(slow_call);
5325 // this->control() comes from set_results_for_java_call
5326 result_reg->init_req(_slow_path, control());
5327 result_val->init_req(_slow_path, slow_result);
5328 result_io ->set_req(_slow_path, i_o());
5329 result_mem ->set_req(_slow_path, reset_memory());
5330 }
5331
5332 // Return the combined state.
5333 set_i_o( _gvn.transform(result_io) );
5334 set_all_memory( _gvn.transform(result_mem));
5335
5336 set_result(result_reg, result_val);
5337 return true;
5338 }
5339
5340 //---------------------------inline_native_getClass----------------------------
5341 // public final native Class<?> java.lang.Object.getClass();
5342 //
5343 // Build special case code for calls to getClass on an object.
5344 bool LibraryCallKit::inline_native_getClass() {
5345 Node* obj = argument(0);
5346 if (obj->is_InlineType()) {
5347 const Type* t = _gvn.type(obj);
5348 if (t->maybe_null()) {
5349 null_check(obj);
5350 }
5351 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5352 return true;
5353 }
5354 obj = null_check_receiver();
5355 if (stopped()) return true;
5356 set_result(load_mirror_from_klass(load_object_klass(obj)));
5357 return true;
5358 }
5359
5360 //-----------------inline_native_Reflection_getCallerClass---------------------
5361 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5362 //
5363 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5364 //
5365 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5366 // in that it must skip particular security frames and checks for
5367 // caller sensitive methods.
5368 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5369 #ifndef PRODUCT
5370 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5371 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5372 }
5373 #endif
5374
5375 if (!jvms()->has_method()) {
5376 #ifndef PRODUCT
5377 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5378 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5379 }
5380 #endif
5381 return false;
5382 }
5383
5384 // Walk back up the JVM state to find the caller at the required
5385 // depth.
5386 JVMState* caller_jvms = jvms();
5387
5388 // Cf. JVM_GetCallerClass
5389 // NOTE: Start the loop at depth 1 because the current JVM state does
5390 // not include the Reflection.getCallerClass() frame.
5391 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5392 ciMethod* m = caller_jvms->method();
5393 switch (n) {
5394 case 0:
5395 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5396 break;
5397 case 1:
5398 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5399 if (!m->caller_sensitive()) {
5400 #ifndef PRODUCT
5401 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5402 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5403 }
5404 #endif
5405 return false; // bail-out; let JVM_GetCallerClass do the work
5406 }
5407 break;
5408 default:
5409 if (!m->is_ignored_by_security_stack_walk()) {
5410 // We have reached the desired frame; return the holder class.
5411 // Acquire method holder as java.lang.Class and push as constant.
5412 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5413 ciInstance* caller_mirror = caller_klass->java_mirror();
5414 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5415
5416 #ifndef PRODUCT
5417 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5418 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());
5419 tty->print_cr(" JVM state at this point:");
5420 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5421 ciMethod* m = jvms()->of_depth(i)->method();
5422 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5423 }
5424 }
5425 #endif
5426 return true;
5427 }
5428 break;
5429 }
5430 }
5431
5432 #ifndef PRODUCT
5433 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5434 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5435 tty->print_cr(" JVM state at this point:");
5436 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5437 ciMethod* m = jvms()->of_depth(i)->method();
5438 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5439 }
5440 }
5441 #endif
5442
5443 return false; // bail-out; let JVM_GetCallerClass do the work
5444 }
5445
5446 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5447 Node* arg = argument(0);
5448 Node* result = nullptr;
5449
5450 switch (id) {
5451 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5452 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5453 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5454 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5455 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5456 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5457
5458 case vmIntrinsics::_doubleToLongBits: {
5459 // two paths (plus control) merge in a wood
5460 RegionNode *r = new RegionNode(3);
5461 Node *phi = new PhiNode(r, TypeLong::LONG);
5462
5463 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5464 // Build the boolean node
5465 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5466
5467 // Branch either way.
5468 // NaN case is less traveled, which makes all the difference.
5469 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5470 Node *opt_isnan = _gvn.transform(ifisnan);
5471 assert( opt_isnan->is_If(), "Expect an IfNode");
5472 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5473 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5474
5475 set_control(iftrue);
5476
5477 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5478 Node *slow_result = longcon(nan_bits); // return NaN
5479 phi->init_req(1, _gvn.transform( slow_result ));
5480 r->init_req(1, iftrue);
5481
5482 // Else fall through
5483 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5484 set_control(iffalse);
5485
5486 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5487 r->init_req(2, iffalse);
5488
5489 // Post merge
5490 set_control(_gvn.transform(r));
5491 record_for_igvn(r);
5492
5493 C->set_has_split_ifs(true); // Has chance for split-if optimization
5494 result = phi;
5495 assert(result->bottom_type()->isa_long(), "must be");
5496 break;
5497 }
5498
5499 case vmIntrinsics::_floatToIntBits: {
5500 // two paths (plus control) merge in a wood
5501 RegionNode *r = new RegionNode(3);
5502 Node *phi = new PhiNode(r, TypeInt::INT);
5503
5504 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5505 // Build the boolean node
5506 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5507
5508 // Branch either way.
5509 // NaN case is less traveled, which makes all the difference.
5510 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5511 Node *opt_isnan = _gvn.transform(ifisnan);
5512 assert( opt_isnan->is_If(), "Expect an IfNode");
5513 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5514 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5515
5516 set_control(iftrue);
5517
5518 static const jint nan_bits = 0x7fc00000;
5519 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5520 phi->init_req(1, _gvn.transform( slow_result ));
5521 r->init_req(1, iftrue);
5522
5523 // Else fall through
5524 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5525 set_control(iffalse);
5526
5527 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5528 r->init_req(2, iffalse);
5529
5530 // Post merge
5531 set_control(_gvn.transform(r));
5532 record_for_igvn(r);
5533
5534 C->set_has_split_ifs(true); // Has chance for split-if optimization
5535 result = phi;
5536 assert(result->bottom_type()->isa_int(), "must be");
5537 break;
5538 }
5539
5540 default:
5541 fatal_unexpected_iid(id);
5542 break;
5543 }
5544 set_result(_gvn.transform(result));
5545 return true;
5546 }
5547
5548 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5549 Node* arg = argument(0);
5550 Node* result = nullptr;
5551
5552 switch (id) {
5553 case vmIntrinsics::_floatIsInfinite:
5554 result = new IsInfiniteFNode(arg);
5555 break;
5556 case vmIntrinsics::_floatIsFinite:
5557 result = new IsFiniteFNode(arg);
5558 break;
5559 case vmIntrinsics::_doubleIsInfinite:
5560 result = new IsInfiniteDNode(arg);
5561 break;
5562 case vmIntrinsics::_doubleIsFinite:
5563 result = new IsFiniteDNode(arg);
5564 break;
5565 default:
5566 fatal_unexpected_iid(id);
5567 break;
5568 }
5569 set_result(_gvn.transform(result));
5570 return true;
5571 }
5572
5573 //----------------------inline_unsafe_copyMemory-------------------------
5574 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5575
5576 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5577 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5578 const Type* base_t = gvn.type(base);
5579
5580 bool in_native = (base_t == TypePtr::NULL_PTR);
5581 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5582 bool is_mixed = !in_heap && !in_native;
5583
5584 if (is_mixed) {
5585 return true; // mixed accesses can touch both on-heap and off-heap memory
5586 }
5587 if (in_heap) {
5588 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5589 if (!is_prim_array) {
5590 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5591 // there's not enough type information available to determine proper memory slice for it.
5592 return true;
5593 }
5594 }
5595 return false;
5596 }
5597
5598 bool LibraryCallKit::inline_unsafe_copyMemory() {
5599 if (callee()->is_static()) return false; // caller must have the capability!
5600 null_check_receiver(); // null-check receiver
5601 if (stopped()) return true;
5602
5603 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5604
5605 Node* src_base = argument(1); // type: oop
5606 Node* src_off = ConvL2X(argument(2)); // type: long
5607 Node* dst_base = argument(4); // type: oop
5608 Node* dst_off = ConvL2X(argument(5)); // type: long
5609 Node* size = ConvL2X(argument(7)); // type: long
5610
5611 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5612 "fieldOffset must be byte-scaled");
5613
5614 Node* src_addr = make_unsafe_address(src_base, src_off);
5615 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5616
5617 Node* thread = _gvn.transform(new ThreadLocalNode());
5618 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5619 BasicType doing_unsafe_access_bt = T_BYTE;
5620 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5621
5622 // update volatile field
5623 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5624
5625 int flags = RC_LEAF | RC_NO_FP;
5626
5627 const TypePtr* dst_type = TypePtr::BOTTOM;
5628
5629 // Adjust memory effects of the runtime call based on input values.
5630 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5631 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5632 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5633
5634 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5635 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5636 flags |= RC_NARROW_MEM; // narrow in memory
5637 }
5638 }
5639
5640 // Call it. Note that the length argument is not scaled.
5641 make_runtime_call(flags,
5642 OptoRuntime::fast_arraycopy_Type(),
5643 StubRoutines::unsafe_arraycopy(),
5644 "unsafe_arraycopy",
5645 dst_type,
5646 src_addr, dst_addr, size XTOP);
5647
5648 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5649
5650 return true;
5651 }
5652
5653 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5654 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5655 bool LibraryCallKit::inline_unsafe_setMemory() {
5656 if (callee()->is_static()) return false; // caller must have the capability!
5657 null_check_receiver(); // null-check receiver
5658 if (stopped()) return true;
5659
5660 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5661
5662 Node* dst_base = argument(1); // type: oop
5663 Node* dst_off = ConvL2X(argument(2)); // type: long
5664 Node* size = ConvL2X(argument(4)); // type: long
5665 Node* byte = argument(6); // type: byte
5666
5667 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5668 "fieldOffset must be byte-scaled");
5669
5670 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5671
5672 Node* thread = _gvn.transform(new ThreadLocalNode());
5673 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5674 BasicType doing_unsafe_access_bt = T_BYTE;
5675 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5676
5677 // update volatile field
5678 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5679
5680 int flags = RC_LEAF | RC_NO_FP;
5681
5682 const TypePtr* dst_type = TypePtr::BOTTOM;
5683
5684 // Adjust memory effects of the runtime call based on input values.
5685 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5686 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5687
5688 flags |= RC_NARROW_MEM; // narrow in memory
5689 }
5690
5691 // Call it. Note that the length argument is not scaled.
5692 make_runtime_call(flags,
5693 OptoRuntime::unsafe_setmemory_Type(),
5694 StubRoutines::unsafe_setmemory(),
5695 "unsafe_setmemory",
5696 dst_type,
5697 dst_addr, size XTOP, byte);
5698
5699 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5700
5701 return true;
5702 }
5703
5704 #undef XTOP
5705
5706 //----------------------inline_unsafe_isFlatArray------------------------
5707 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5708 // This intrinsic exploits assumptions made by the native implementation
5709 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5710 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5711 Node* cls = argument(1);
5712 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5713 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p,
5714 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5715 Node* result = flat_array_test(kls);
5716 set_result(result);
5717 return true;
5718 }
5719
5720 //------------------------clone_coping-----------------------------------
5721 // Helper function for inline_native_clone.
5722 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5723 assert(obj_size != nullptr, "");
5724 Node* raw_obj = alloc_obj->in(1);
5725 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5726
5727 AllocateNode* alloc = nullptr;
5728 if (ReduceBulkZeroing &&
5729 // If we are implementing an array clone without knowing its source type
5730 // (can happen when compiling the array-guarded branch of a reflective
5731 // Object.clone() invocation), initialize the array within the allocation.
5732 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5733 // to a runtime clone call that assumes fully initialized source arrays.
5734 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5735 // We will be completely responsible for initializing this object -
5736 // mark Initialize node as complete.
5737 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5738 // The object was just allocated - there should be no any stores!
5739 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5740 // Mark as complete_with_arraycopy so that on AllocateNode
5741 // expansion, we know this AllocateNode is initialized by an array
5742 // copy and a StoreStore barrier exists after the array copy.
5743 alloc->initialization()->set_complete_with_arraycopy();
5744 }
5745
5746 Node* size = _gvn.transform(obj_size);
5747 access_clone(obj, alloc_obj, size, is_array);
5748
5749 // Do not let reads from the cloned object float above the arraycopy.
5750 if (alloc != nullptr) {
5751 // Do not let stores that initialize this object be reordered with
5752 // a subsequent store that would make this object accessible by
5753 // other threads.
5754 // Record what AllocateNode this StoreStore protects so that
5755 // escape analysis can go from the MemBarStoreStoreNode to the
5756 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5757 // based on the escape status of the AllocateNode.
5758 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5759 } else {
5760 insert_mem_bar(Op_MemBarCPUOrder);
5761 }
5762 }
5763
5764 //------------------------inline_native_clone----------------------------
5765 // protected native Object java.lang.Object.clone();
5766 //
5767 // Here are the simple edge cases:
5768 // null receiver => normal trap
5769 // virtual and clone was overridden => slow path to out-of-line clone
5770 // not cloneable or finalizer => slow path to out-of-line Object.clone
5771 //
5772 // The general case has two steps, allocation and copying.
5773 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5774 //
5775 // Copying also has two cases, oop arrays and everything else.
5776 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5777 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5778 //
5779 // These steps fold up nicely if and when the cloned object's klass
5780 // can be sharply typed as an object array, a type array, or an instance.
5781 //
5782 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5783 PhiNode* result_val;
5784
5785 // Set the reexecute bit for the interpreter to reexecute
5786 // the bytecode that invokes Object.clone if deoptimization happens.
5787 { PreserveReexecuteState preexecs(this);
5788 jvms()->set_should_reexecute(true);
5789
5790 Node* obj = argument(0);
5791 obj = null_check_receiver();
5792 if (stopped()) return true;
5793
5794 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5795 if (obj_type->is_inlinetypeptr()) {
5796 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5797 // no identity.
5798 set_result(obj);
5799 return true;
5800 }
5801
5802 // If we are going to clone an instance, we need its exact type to
5803 // know the number and types of fields to convert the clone to
5804 // loads/stores. Maybe a speculative type can help us.
5805 if (!obj_type->klass_is_exact() &&
5806 obj_type->speculative_type() != nullptr &&
5807 obj_type->speculative_type()->is_instance_klass() &&
5808 !obj_type->speculative_type()->is_inlinetype()) {
5809 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5810 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5811 !spec_ik->has_injected_fields()) {
5812 if (!obj_type->isa_instptr() ||
5813 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5814 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5815 }
5816 }
5817 }
5818
5819 // Conservatively insert a memory barrier on all memory slices.
5820 // Do not let writes into the original float below the clone.
5821 insert_mem_bar(Op_MemBarCPUOrder);
5822
5823 // paths into result_reg:
5824 enum {
5825 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5826 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5827 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5828 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5829 PATH_LIMIT
5830 };
5831 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5832 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5833 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5834 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5835 record_for_igvn(result_reg);
5836
5837 // TODO 8350865 For arrays, this might be folded and then not account for atomic arrays
5838 Node* obj_klass = load_object_klass(obj);
5839 // We only go to the fast case code if we pass a number of guards.
5840 // The paths which do not pass are accumulated in the slow_region.
5841 RegionNode* slow_region = new RegionNode(1);
5842 record_for_igvn(slow_region);
5843
5844 Node* array_obj = obj;
5845 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5846 if (array_ctl != nullptr) {
5847 // It's an array.
5848 PreserveJVMState pjvms(this);
5849 set_control(array_ctl);
5850
5851 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5852 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5853 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5854 obj_type->can_be_inline_array() &&
5855 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5856 // Flat inline type array may have object field that would require a
5857 // write barrier. Conservatively, go to slow path.
5858 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5859 }
5860
5861 if (!stopped()) {
5862 Node* obj_length = load_array_length(array_obj);
5863 Node* array_size = nullptr; // Size of the array without object alignment padding.
5864 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5865
5866 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5867 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5868 // If it is an oop array, it requires very special treatment,
5869 // because gc barriers are required when accessing the array.
5870 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5871 if (is_obja != nullptr) {
5872 PreserveJVMState pjvms2(this);
5873 set_control(is_obja);
5874 // Generate a direct call to the right arraycopy function(s).
5875 // Clones are always tightly coupled.
5876 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5877 ac->set_clone_oop_array();
5878 Node* n = _gvn.transform(ac);
5879 assert(n == ac, "cannot disappear");
5880 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5881
5882 result_reg->init_req(_objArray_path, control());
5883 result_val->init_req(_objArray_path, alloc_obj);
5884 result_i_o ->set_req(_objArray_path, i_o());
5885 result_mem ->set_req(_objArray_path, reset_memory());
5886 }
5887 }
5888 // Otherwise, there are no barriers to worry about.
5889 // (We can dispense with card marks if we know the allocation
5890 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5891 // causes the non-eden paths to take compensating steps to
5892 // simulate a fresh allocation, so that no further
5893 // card marks are required in compiled code to initialize
5894 // the object.)
5895
5896 if (!stopped()) {
5897 copy_to_clone(obj, alloc_obj, array_size, true);
5898
5899 // Present the results of the copy.
5900 result_reg->init_req(_array_path, control());
5901 result_val->init_req(_array_path, alloc_obj);
5902 result_i_o ->set_req(_array_path, i_o());
5903 result_mem ->set_req(_array_path, reset_memory());
5904 }
5905 }
5906 }
5907
5908 if (!stopped()) {
5909 // It's an instance (we did array above). Make the slow-path tests.
5910 // If this is a virtual call, we generate a funny guard. We grab
5911 // the vtable entry corresponding to clone() from the target object.
5912 // If the target method which we are calling happens to be the
5913 // Object clone() method, we pass the guard. We do not need this
5914 // guard for non-virtual calls; the caller is known to be the native
5915 // Object clone().
5916 if (is_virtual) {
5917 generate_virtual_guard(obj_klass, slow_region);
5918 }
5919
5920 // The object must be easily cloneable and must not have a finalizer.
5921 // Both of these conditions may be checked in a single test.
5922 // We could optimize the test further, but we don't care.
5923 generate_misc_flags_guard(obj_klass,
5924 // Test both conditions:
5925 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5926 // Must be cloneable but not finalizer:
5927 KlassFlags::_misc_is_cloneable_fast,
5928 slow_region);
5929 }
5930
5931 if (!stopped()) {
5932 // It's an instance, and it passed the slow-path tests.
5933 PreserveJVMState pjvms(this);
5934 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5935 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5936 // is reexecuted if deoptimization occurs and there could be problems when merging
5937 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5938 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5939
5940 copy_to_clone(obj, alloc_obj, obj_size, false);
5941
5942 // Present the results of the slow call.
5943 result_reg->init_req(_instance_path, control());
5944 result_val->init_req(_instance_path, alloc_obj);
5945 result_i_o ->set_req(_instance_path, i_o());
5946 result_mem ->set_req(_instance_path, reset_memory());
5947 }
5948
5949 // Generate code for the slow case. We make a call to clone().
5950 set_control(_gvn.transform(slow_region));
5951 if (!stopped()) {
5952 PreserveJVMState pjvms(this);
5953 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5954 // We need to deoptimize on exception (see comment above)
5955 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5956 // this->control() comes from set_results_for_java_call
5957 result_reg->init_req(_slow_path, control());
5958 result_val->init_req(_slow_path, slow_result);
5959 result_i_o ->set_req(_slow_path, i_o());
5960 result_mem ->set_req(_slow_path, reset_memory());
5961 }
5962
5963 // Return the combined state.
5964 set_control( _gvn.transform(result_reg));
5965 set_i_o( _gvn.transform(result_i_o));
5966 set_all_memory( _gvn.transform(result_mem));
5967 } // original reexecute is set back here
5968
5969 set_result(_gvn.transform(result_val));
5970 return true;
5971 }
5972
5973 // If we have a tightly coupled allocation, the arraycopy may take care
5974 // of the array initialization. If one of the guards we insert between
5975 // the allocation and the arraycopy causes a deoptimization, an
5976 // uninitialized array will escape the compiled method. To prevent that
5977 // we set the JVM state for uncommon traps between the allocation and
5978 // the arraycopy to the state before the allocation so, in case of
5979 // deoptimization, we'll reexecute the allocation and the
5980 // initialization.
5981 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5982 if (alloc != nullptr) {
5983 ciMethod* trap_method = alloc->jvms()->method();
5984 int trap_bci = alloc->jvms()->bci();
5985
5986 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5987 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5988 // Make sure there's no store between the allocation and the
5989 // arraycopy otherwise visible side effects could be rexecuted
5990 // in case of deoptimization and cause incorrect execution.
5991 bool no_interfering_store = true;
5992 Node* mem = alloc->in(TypeFunc::Memory);
5993 if (mem->is_MergeMem()) {
5994 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5995 Node* n = mms.memory();
5996 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5997 assert(n->is_Store(), "what else?");
5998 no_interfering_store = false;
5999 break;
6000 }
6001 }
6002 } else {
6003 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6004 Node* n = mms.memory();
6005 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6006 assert(n->is_Store(), "what else?");
6007 no_interfering_store = false;
6008 break;
6009 }
6010 }
6011 }
6012
6013 if (no_interfering_store) {
6014 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6015
6016 JVMState* saved_jvms = jvms();
6017 saved_reexecute_sp = _reexecute_sp;
6018
6019 set_jvms(sfpt->jvms());
6020 _reexecute_sp = jvms()->sp();
6021
6022 return saved_jvms;
6023 }
6024 }
6025 }
6026 return nullptr;
6027 }
6028
6029 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6030 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6031 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6032 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6033 uint size = alloc->req();
6034 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6035 old_jvms->set_map(sfpt);
6036 for (uint i = 0; i < size; i++) {
6037 sfpt->init_req(i, alloc->in(i));
6038 }
6039 int adjustment = 1;
6040 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6041 if (ary_klass_ptr->is_null_free()) {
6042 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6043 // also requires the componentType and initVal on stack for re-execution.
6044 // Re-create and push the componentType.
6045 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6046 ciInstance* instance = klass->component_mirror_instance();
6047 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6048 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6049 adjustment++;
6050 }
6051 // re-push array length for deoptimization
6052 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6053 if (ary_klass_ptr->is_null_free()) {
6054 // Re-create and push the initVal.
6055 Node* init_val = alloc->in(AllocateNode::InitValue);
6056 if (init_val == nullptr) {
6057 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6058 } else if (UseCompressedOops) {
6059 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6060 }
6061 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6062 adjustment++;
6063 }
6064 old_jvms->set_sp(old_jvms->sp() + adjustment);
6065 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6066 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6067 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6068 old_jvms->set_should_reexecute(true);
6069
6070 sfpt->set_i_o(map()->i_o());
6071 sfpt->set_memory(map()->memory());
6072 sfpt->set_control(map()->control());
6073 return sfpt;
6074 }
6075
6076 // In case of a deoptimization, we restart execution at the
6077 // allocation, allocating a new array. We would leave an uninitialized
6078 // array in the heap that GCs wouldn't expect. Move the allocation
6079 // after the traps so we don't allocate the array if we
6080 // deoptimize. This is possible because tightly_coupled_allocation()
6081 // guarantees there's no observer of the allocated array at this point
6082 // and the control flow is simple enough.
6083 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6084 int saved_reexecute_sp, uint new_idx) {
6085 if (saved_jvms_before_guards != nullptr && !stopped()) {
6086 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6087
6088 assert(alloc != nullptr, "only with a tightly coupled allocation");
6089 // restore JVM state to the state at the arraycopy
6090 saved_jvms_before_guards->map()->set_control(map()->control());
6091 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6092 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6093 // If we've improved the types of some nodes (null check) while
6094 // emitting the guards, propagate them to the current state
6095 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6096 set_jvms(saved_jvms_before_guards);
6097 _reexecute_sp = saved_reexecute_sp;
6098
6099 // Remove the allocation from above the guards
6100 CallProjections* callprojs = alloc->extract_projections(true);
6101 InitializeNode* init = alloc->initialization();
6102 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6103 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6104 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
6105
6106 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6107 // the allocation (i.e. is only valid if the allocation succeeds):
6108 // 1) replace CastIINode with AllocateArrayNode's length here
6109 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6110 //
6111 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6112 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6113 Node* init_control = init->proj_out(TypeFunc::Control);
6114 Node* alloc_length = alloc->Ideal_length();
6115 #ifdef ASSERT
6116 Node* prev_cast = nullptr;
6117 #endif
6118 for (uint i = 0; i < init_control->outcnt(); i++) {
6119 Node* init_out = init_control->raw_out(i);
6120 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6121 #ifdef ASSERT
6122 if (prev_cast == nullptr) {
6123 prev_cast = init_out;
6124 } else {
6125 if (prev_cast->cmp(*init_out) == false) {
6126 prev_cast->dump();
6127 init_out->dump();
6128 assert(false, "not equal CastIINode");
6129 }
6130 }
6131 #endif
6132 C->gvn_replace_by(init_out, alloc_length);
6133 }
6134 }
6135 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6136
6137 // move the allocation here (after the guards)
6138 _gvn.hash_delete(alloc);
6139 alloc->set_req(TypeFunc::Control, control());
6140 alloc->set_req(TypeFunc::I_O, i_o());
6141 Node *mem = reset_memory();
6142 set_all_memory(mem);
6143 alloc->set_req(TypeFunc::Memory, mem);
6144 set_control(init->proj_out_or_null(TypeFunc::Control));
6145 set_i_o(callprojs->fallthrough_ioproj);
6146
6147 // Update memory as done in GraphKit::set_output_for_allocation()
6148 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6149 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6150 if (ary_type->isa_aryptr() && length_type != nullptr) {
6151 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6152 }
6153 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6154 int elemidx = C->get_alias_index(telemref);
6155 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
6156 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
6157
6158 Node* allocx = _gvn.transform(alloc);
6159 assert(allocx == alloc, "where has the allocation gone?");
6160 assert(dest->is_CheckCastPP(), "not an allocation result?");
6161
6162 _gvn.hash_delete(dest);
6163 dest->set_req(0, control());
6164 Node* destx = _gvn.transform(dest);
6165 assert(destx == dest, "where has the allocation result gone?");
6166
6167 array_ideal_length(alloc, ary_type, true);
6168 }
6169 }
6170
6171 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6172 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6173 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6174 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6175 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6176 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6177 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6178 JVMState* saved_jvms_before_guards) {
6179 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6180 // There is at least one unrelated uncommon trap which needs to be replaced.
6181 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6182
6183 JVMState* saved_jvms = jvms();
6184 const int saved_reexecute_sp = _reexecute_sp;
6185 set_jvms(sfpt->jvms());
6186 _reexecute_sp = jvms()->sp();
6187
6188 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6189
6190 // Restore state
6191 set_jvms(saved_jvms);
6192 _reexecute_sp = saved_reexecute_sp;
6193 }
6194 }
6195
6196 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6197 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6198 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6199 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6200 while (if_proj->is_IfProj()) {
6201 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6202 if (uncommon_trap != nullptr) {
6203 create_new_uncommon_trap(uncommon_trap);
6204 }
6205 assert(if_proj->in(0)->is_If(), "must be If");
6206 if_proj = if_proj->in(0)->in(0);
6207 }
6208 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6209 "must have reached control projection of init node");
6210 }
6211
6212 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6213 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6214 assert(trap_request != 0, "no valid UCT trap request");
6215 PreserveJVMState pjvms(this);
6216 set_control(uncommon_trap_call->in(0));
6217 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6218 Deoptimization::trap_request_action(trap_request));
6219 assert(stopped(), "Should be stopped");
6220 _gvn.hash_delete(uncommon_trap_call);
6221 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6222 }
6223
6224 // Common checks for array sorting intrinsics arguments.
6225 // Returns `true` if checks passed.
6226 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6227 // check address of the class
6228 if (elementType == nullptr || elementType->is_top()) {
6229 return false; // dead path
6230 }
6231 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6232 if (elem_klass == nullptr) {
6233 return false; // dead path
6234 }
6235 // java_mirror_type() returns non-null for compile-time Class constants only
6236 ciType* elem_type = elem_klass->java_mirror_type();
6237 if (elem_type == nullptr) {
6238 return false;
6239 }
6240 bt = elem_type->basic_type();
6241 // Disable the intrinsic if the CPU does not support SIMD sort
6242 if (!Matcher::supports_simd_sort(bt)) {
6243 return false;
6244 }
6245 // check address of the array
6246 if (obj == nullptr || obj->is_top()) {
6247 return false; // dead path
6248 }
6249 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6250 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6251 return false; // failed input validation
6252 }
6253 return true;
6254 }
6255
6256 //------------------------------inline_array_partition-----------------------
6257 bool LibraryCallKit::inline_array_partition() {
6258 address stubAddr = StubRoutines::select_array_partition_function();
6259 if (stubAddr == nullptr) {
6260 return false; // Intrinsic's stub is not implemented on this platform
6261 }
6262 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6263
6264 // no receiver because it is a static method
6265 Node* elementType = argument(0);
6266 Node* obj = argument(1);
6267 Node* offset = argument(2); // long
6268 Node* fromIndex = argument(4);
6269 Node* toIndex = argument(5);
6270 Node* indexPivot1 = argument(6);
6271 Node* indexPivot2 = argument(7);
6272 // PartitionOperation: argument(8) is ignored
6273
6274 Node* pivotIndices = nullptr;
6275 BasicType bt = T_ILLEGAL;
6276
6277 if (!check_array_sort_arguments(elementType, obj, bt)) {
6278 return false;
6279 }
6280 null_check(obj);
6281 // If obj is dead, only null-path is taken.
6282 if (stopped()) {
6283 return true;
6284 }
6285 // Set the original stack and the reexecute bit for the interpreter to reexecute
6286 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6287 { PreserveReexecuteState preexecs(this);
6288 jvms()->set_should_reexecute(true);
6289
6290 Node* obj_adr = make_unsafe_address(obj, offset);
6291
6292 // create the pivotIndices array of type int and size = 2
6293 Node* size = intcon(2);
6294 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6295 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6296 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6297 guarantee(alloc != nullptr, "created above");
6298 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6299
6300 // pass the basic type enum to the stub
6301 Node* elemType = intcon(bt);
6302
6303 // Call the stub
6304 const char *stubName = "array_partition_stub";
6305 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6306 stubAddr, stubName, TypePtr::BOTTOM,
6307 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6308 indexPivot1, indexPivot2);
6309
6310 } // original reexecute is set back here
6311
6312 if (!stopped()) {
6313 set_result(pivotIndices);
6314 }
6315
6316 return true;
6317 }
6318
6319
6320 //------------------------------inline_array_sort-----------------------
6321 bool LibraryCallKit::inline_array_sort() {
6322 address stubAddr = StubRoutines::select_arraysort_function();
6323 if (stubAddr == nullptr) {
6324 return false; // Intrinsic's stub is not implemented on this platform
6325 }
6326 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6327
6328 // no receiver because it is a static method
6329 Node* elementType = argument(0);
6330 Node* obj = argument(1);
6331 Node* offset = argument(2); // long
6332 Node* fromIndex = argument(4);
6333 Node* toIndex = argument(5);
6334 // SortOperation: argument(6) is ignored
6335
6336 BasicType bt = T_ILLEGAL;
6337
6338 if (!check_array_sort_arguments(elementType, obj, bt)) {
6339 return false;
6340 }
6341 null_check(obj);
6342 // If obj is dead, only null-path is taken.
6343 if (stopped()) {
6344 return true;
6345 }
6346 Node* obj_adr = make_unsafe_address(obj, offset);
6347
6348 // pass the basic type enum to the stub
6349 Node* elemType = intcon(bt);
6350
6351 // Call the stub.
6352 const char *stubName = "arraysort_stub";
6353 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6354 stubAddr, stubName, TypePtr::BOTTOM,
6355 obj_adr, elemType, fromIndex, toIndex);
6356
6357 return true;
6358 }
6359
6360
6361 //------------------------------inline_arraycopy-----------------------
6362 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6363 // Object dest, int destPos,
6364 // int length);
6365 bool LibraryCallKit::inline_arraycopy() {
6366 // Get the arguments.
6367 Node* src = argument(0); // type: oop
6368 Node* src_offset = argument(1); // type: int
6369 Node* dest = argument(2); // type: oop
6370 Node* dest_offset = argument(3); // type: int
6371 Node* length = argument(4); // type: int
6372
6373 uint new_idx = C->unique();
6374
6375 // Check for allocation before we add nodes that would confuse
6376 // tightly_coupled_allocation()
6377 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6378
6379 int saved_reexecute_sp = -1;
6380 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6381 // See arraycopy_restore_alloc_state() comment
6382 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6383 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6384 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6385 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6386
6387 // The following tests must be performed
6388 // (1) src and dest are arrays.
6389 // (2) src and dest arrays must have elements of the same BasicType
6390 // (3) src and dest must not be null.
6391 // (4) src_offset must not be negative.
6392 // (5) dest_offset must not be negative.
6393 // (6) length must not be negative.
6394 // (7) src_offset + length must not exceed length of src.
6395 // (8) dest_offset + length must not exceed length of dest.
6396 // (9) each element of an oop array must be assignable
6397
6398 // (3) src and dest must not be null.
6399 // always do this here because we need the JVM state for uncommon traps
6400 Node* null_ctl = top();
6401 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6402 assert(null_ctl->is_top(), "no null control here");
6403 dest = null_check(dest, T_ARRAY);
6404
6405 if (!can_emit_guards) {
6406 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6407 // guards but the arraycopy node could still take advantage of a
6408 // tightly allocated allocation. tightly_coupled_allocation() is
6409 // called again to make sure it takes the null check above into
6410 // account: the null check is mandatory and if it caused an
6411 // uncommon trap to be emitted then the allocation can't be
6412 // considered tightly coupled in this context.
6413 alloc = tightly_coupled_allocation(dest);
6414 }
6415
6416 bool validated = false;
6417
6418 const Type* src_type = _gvn.type(src);
6419 const Type* dest_type = _gvn.type(dest);
6420 const TypeAryPtr* top_src = src_type->isa_aryptr();
6421 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6422
6423 // Do we have the type of src?
6424 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6425 // Do we have the type of dest?
6426 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6427 // Is the type for src from speculation?
6428 bool src_spec = false;
6429 // Is the type for dest from speculation?
6430 bool dest_spec = false;
6431
6432 if ((!has_src || !has_dest) && can_emit_guards) {
6433 // We don't have sufficient type information, let's see if
6434 // speculative types can help. We need to have types for both src
6435 // and dest so that it pays off.
6436
6437 // Do we already have or could we have type information for src
6438 bool could_have_src = has_src;
6439 // Do we already have or could we have type information for dest
6440 bool could_have_dest = has_dest;
6441
6442 ciKlass* src_k = nullptr;
6443 if (!has_src) {
6444 src_k = src_type->speculative_type_not_null();
6445 if (src_k != nullptr && src_k->is_array_klass()) {
6446 could_have_src = true;
6447 }
6448 }
6449
6450 ciKlass* dest_k = nullptr;
6451 if (!has_dest) {
6452 dest_k = dest_type->speculative_type_not_null();
6453 if (dest_k != nullptr && dest_k->is_array_klass()) {
6454 could_have_dest = true;
6455 }
6456 }
6457
6458 if (could_have_src && could_have_dest) {
6459 // This is going to pay off so emit the required guards
6460 if (!has_src) {
6461 src = maybe_cast_profiled_obj(src, src_k, true);
6462 src_type = _gvn.type(src);
6463 top_src = src_type->isa_aryptr();
6464 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6465 src_spec = true;
6466 }
6467 if (!has_dest) {
6468 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6469 dest_type = _gvn.type(dest);
6470 top_dest = dest_type->isa_aryptr();
6471 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6472 dest_spec = true;
6473 }
6474 }
6475 }
6476
6477 if (has_src && has_dest && can_emit_guards) {
6478 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6479 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6480 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6481 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6482
6483 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6484 // If both arrays are object arrays then having the exact types
6485 // for both will remove the need for a subtype check at runtime
6486 // before the call and may make it possible to pick a faster copy
6487 // routine (without a subtype check on every element)
6488 // Do we have the exact type of src?
6489 bool could_have_src = src_spec;
6490 // Do we have the exact type of dest?
6491 bool could_have_dest = dest_spec;
6492 ciKlass* src_k = nullptr;
6493 ciKlass* dest_k = nullptr;
6494 if (!src_spec) {
6495 src_k = src_type->speculative_type_not_null();
6496 if (src_k != nullptr && src_k->is_array_klass()) {
6497 could_have_src = true;
6498 }
6499 }
6500 if (!dest_spec) {
6501 dest_k = dest_type->speculative_type_not_null();
6502 if (dest_k != nullptr && dest_k->is_array_klass()) {
6503 could_have_dest = true;
6504 }
6505 }
6506 if (could_have_src && could_have_dest) {
6507 // If we can have both exact types, emit the missing guards
6508 if (could_have_src && !src_spec) {
6509 src = maybe_cast_profiled_obj(src, src_k, true);
6510 src_type = _gvn.type(src);
6511 top_src = src_type->isa_aryptr();
6512 }
6513 if (could_have_dest && !dest_spec) {
6514 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6515 dest_type = _gvn.type(dest);
6516 top_dest = dest_type->isa_aryptr();
6517 }
6518 }
6519 }
6520 }
6521
6522 ciMethod* trap_method = method();
6523 int trap_bci = bci();
6524 if (saved_jvms_before_guards != nullptr) {
6525 trap_method = alloc->jvms()->method();
6526 trap_bci = alloc->jvms()->bci();
6527 }
6528
6529 bool negative_length_guard_generated = false;
6530
6531 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6532 can_emit_guards && !src->is_top() && !dest->is_top()) {
6533 // validate arguments: enables transformation the ArrayCopyNode
6534 validated = true;
6535
6536 RegionNode* slow_region = new RegionNode(1);
6537 record_for_igvn(slow_region);
6538
6539 // (1) src and dest are arrays.
6540 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6541 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6542
6543 // (2) src and dest arrays must have elements of the same BasicType
6544 // done at macro expansion or at Ideal transformation time
6545
6546 // (4) src_offset must not be negative.
6547 generate_negative_guard(src_offset, slow_region);
6548
6549 // (5) dest_offset must not be negative.
6550 generate_negative_guard(dest_offset, slow_region);
6551
6552 // (7) src_offset + length must not exceed length of src.
6553 generate_limit_guard(src_offset, length,
6554 load_array_length(src),
6555 slow_region);
6556
6557 // (8) dest_offset + length must not exceed length of dest.
6558 generate_limit_guard(dest_offset, length,
6559 load_array_length(dest),
6560 slow_region);
6561
6562 // (6) length must not be negative.
6563 // This is also checked in generate_arraycopy() during macro expansion, but
6564 // we also have to check it here for the case where the ArrayCopyNode will
6565 // be eliminated by Escape Analysis.
6566 if (EliminateAllocations) {
6567 generate_negative_guard(length, slow_region);
6568 negative_length_guard_generated = true;
6569 }
6570
6571 // (9) each element of an oop array must be assignable
6572 Node* dest_klass = load_object_klass(dest);
6573 if (src != dest) {
6574 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6575 slow_region->add_req(not_subtype_ctrl);
6576 }
6577
6578 // TODO 8350865 Fix below logic. Also handle atomicity.
6579 generate_fair_guard(flat_array_test(src), slow_region);
6580 generate_fair_guard(flat_array_test(dest), slow_region);
6581
6582 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6583 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6584 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6585 src_type = _gvn.type(src);
6586 top_src = src_type->isa_aryptr();
6587
6588 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6589 if (!stopped() && UseArrayFlattening) {
6590 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6591 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6592 if (top_src != nullptr && top_src->is_flat()) {
6593 // Src is flat, check that dest is flat as well
6594 if (top_dest != nullptr && !top_dest->is_flat()) {
6595 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6596 // Since dest is flat and src <: dest, dest must have the same type as src.
6597 top_dest = top_src->cast_to_exactness(false);
6598 assert(top_dest->is_flat(), "dest must be flat");
6599 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6600 }
6601 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6602 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6603 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6604 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6605 generate_fair_guard(flat_array_test(src), slow_region);
6606 if (top_src != nullptr) {
6607 top_src = top_src->cast_to_not_flat();
6608 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6609 }
6610 }
6611 }
6612
6613 {
6614 PreserveJVMState pjvms(this);
6615 set_control(_gvn.transform(slow_region));
6616 uncommon_trap(Deoptimization::Reason_intrinsic,
6617 Deoptimization::Action_make_not_entrant);
6618 assert(stopped(), "Should be stopped");
6619 }
6620 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6621 }
6622
6623 if (stopped()) {
6624 return true;
6625 }
6626
6627 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6628 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6629 // so the compiler has a chance to eliminate them: during macro expansion,
6630 // we have to set their control (CastPP nodes are eliminated).
6631 load_object_klass(src), load_object_klass(dest),
6632 load_array_length(src), load_array_length(dest));
6633
6634 ac->set_arraycopy(validated);
6635
6636 Node* n = _gvn.transform(ac);
6637 if (n == ac) {
6638 ac->connect_outputs(this);
6639 } else {
6640 assert(validated, "shouldn't transform if all arguments not validated");
6641 set_all_memory(n);
6642 }
6643 clear_upper_avx();
6644
6645
6646 return true;
6647 }
6648
6649
6650 // Helper function which determines if an arraycopy immediately follows
6651 // an allocation, with no intervening tests or other escapes for the object.
6652 AllocateArrayNode*
6653 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6654 if (stopped()) return nullptr; // no fast path
6655 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6656
6657 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6658 if (alloc == nullptr) return nullptr;
6659
6660 Node* rawmem = memory(Compile::AliasIdxRaw);
6661 // Is the allocation's memory state untouched?
6662 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6663 // Bail out if there have been raw-memory effects since the allocation.
6664 // (Example: There might have been a call or safepoint.)
6665 return nullptr;
6666 }
6667 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6668 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6669 return nullptr;
6670 }
6671
6672 // There must be no unexpected observers of this allocation.
6673 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6674 Node* obs = ptr->fast_out(i);
6675 if (obs != this->map()) {
6676 return nullptr;
6677 }
6678 }
6679
6680 // This arraycopy must unconditionally follow the allocation of the ptr.
6681 Node* alloc_ctl = ptr->in(0);
6682 Node* ctl = control();
6683 while (ctl != alloc_ctl) {
6684 // There may be guards which feed into the slow_region.
6685 // Any other control flow means that we might not get a chance
6686 // to finish initializing the allocated object.
6687 // Various low-level checks bottom out in uncommon traps. These
6688 // are considered safe since we've already checked above that
6689 // there is no unexpected observer of this allocation.
6690 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6691 assert(ctl->in(0)->is_If(), "must be If");
6692 ctl = ctl->in(0)->in(0);
6693 } else {
6694 return nullptr;
6695 }
6696 }
6697
6698 // If we get this far, we have an allocation which immediately
6699 // precedes the arraycopy, and we can take over zeroing the new object.
6700 // The arraycopy will finish the initialization, and provide
6701 // a new control state to which we will anchor the destination pointer.
6702
6703 return alloc;
6704 }
6705
6706 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6707 if (node->is_IfProj()) {
6708 Node* other_proj = node->as_IfProj()->other_if_proj();
6709 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6710 Node* obs = other_proj->fast_out(j);
6711 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6712 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6713 return obs->as_CallStaticJava();
6714 }
6715 }
6716 }
6717 return nullptr;
6718 }
6719
6720 //-------------inline_encodeISOArray-----------------------------------
6721 // encode char[] to byte[] in ISO_8859_1 or ASCII
6722 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6723 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6724 // no receiver since it is static method
6725 Node *src = argument(0);
6726 Node *src_offset = argument(1);
6727 Node *dst = argument(2);
6728 Node *dst_offset = argument(3);
6729 Node *length = argument(4);
6730
6731 src = must_be_not_null(src, true);
6732 dst = must_be_not_null(dst, true);
6733
6734 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6735 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6736 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6737 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6738 // failed array check
6739 return false;
6740 }
6741
6742 // Figure out the size and type of the elements we will be copying.
6743 BasicType src_elem = src_type->elem()->array_element_basic_type();
6744 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6745 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6746 return false;
6747 }
6748
6749 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6750 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6751 // 'src_start' points to src array + scaled offset
6752 // 'dst_start' points to dst array + scaled offset
6753
6754 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6755 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6756 enc = _gvn.transform(enc);
6757 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6758 set_memory(res_mem, mtype);
6759 set_result(enc);
6760 clear_upper_avx();
6761
6762 return true;
6763 }
6764
6765 //-------------inline_multiplyToLen-----------------------------------
6766 bool LibraryCallKit::inline_multiplyToLen() {
6767 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6768
6769 address stubAddr = StubRoutines::multiplyToLen();
6770 if (stubAddr == nullptr) {
6771 return false; // Intrinsic's stub is not implemented on this platform
6772 }
6773 const char* stubName = "multiplyToLen";
6774
6775 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6776
6777 // no receiver because it is a static method
6778 Node* x = argument(0);
6779 Node* xlen = argument(1);
6780 Node* y = argument(2);
6781 Node* ylen = argument(3);
6782 Node* z = argument(4);
6783
6784 x = must_be_not_null(x, true);
6785 y = must_be_not_null(y, true);
6786
6787 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6788 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6789 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6790 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6791 // failed array check
6792 return false;
6793 }
6794
6795 BasicType x_elem = x_type->elem()->array_element_basic_type();
6796 BasicType y_elem = y_type->elem()->array_element_basic_type();
6797 if (x_elem != T_INT || y_elem != T_INT) {
6798 return false;
6799 }
6800
6801 Node* x_start = array_element_address(x, intcon(0), x_elem);
6802 Node* y_start = array_element_address(y, intcon(0), y_elem);
6803 // 'x_start' points to x array + scaled xlen
6804 // 'y_start' points to y array + scaled ylen
6805
6806 Node* z_start = array_element_address(z, intcon(0), T_INT);
6807
6808 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6809 OptoRuntime::multiplyToLen_Type(),
6810 stubAddr, stubName, TypePtr::BOTTOM,
6811 x_start, xlen, y_start, ylen, z_start);
6812
6813 C->set_has_split_ifs(true); // Has chance for split-if optimization
6814 set_result(z);
6815 return true;
6816 }
6817
6818 //-------------inline_squareToLen------------------------------------
6819 bool LibraryCallKit::inline_squareToLen() {
6820 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6821
6822 address stubAddr = StubRoutines::squareToLen();
6823 if (stubAddr == nullptr) {
6824 return false; // Intrinsic's stub is not implemented on this platform
6825 }
6826 const char* stubName = "squareToLen";
6827
6828 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6829
6830 Node* x = argument(0);
6831 Node* len = argument(1);
6832 Node* z = argument(2);
6833 Node* zlen = argument(3);
6834
6835 x = must_be_not_null(x, true);
6836 z = must_be_not_null(z, true);
6837
6838 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6839 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6840 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6841 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6842 // failed array check
6843 return false;
6844 }
6845
6846 BasicType x_elem = x_type->elem()->array_element_basic_type();
6847 BasicType z_elem = z_type->elem()->array_element_basic_type();
6848 if (x_elem != T_INT || z_elem != T_INT) {
6849 return false;
6850 }
6851
6852
6853 Node* x_start = array_element_address(x, intcon(0), x_elem);
6854 Node* z_start = array_element_address(z, intcon(0), z_elem);
6855
6856 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6857 OptoRuntime::squareToLen_Type(),
6858 stubAddr, stubName, TypePtr::BOTTOM,
6859 x_start, len, z_start, zlen);
6860
6861 set_result(z);
6862 return true;
6863 }
6864
6865 //-------------inline_mulAdd------------------------------------------
6866 bool LibraryCallKit::inline_mulAdd() {
6867 assert(UseMulAddIntrinsic, "not implemented on this platform");
6868
6869 address stubAddr = StubRoutines::mulAdd();
6870 if (stubAddr == nullptr) {
6871 return false; // Intrinsic's stub is not implemented on this platform
6872 }
6873 const char* stubName = "mulAdd";
6874
6875 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6876
6877 Node* out = argument(0);
6878 Node* in = argument(1);
6879 Node* offset = argument(2);
6880 Node* len = argument(3);
6881 Node* k = argument(4);
6882
6883 in = must_be_not_null(in, true);
6884 out = must_be_not_null(out, true);
6885
6886 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6887 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6888 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6889 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6890 // failed array check
6891 return false;
6892 }
6893
6894 BasicType out_elem = out_type->elem()->array_element_basic_type();
6895 BasicType in_elem = in_type->elem()->array_element_basic_type();
6896 if (out_elem != T_INT || in_elem != T_INT) {
6897 return false;
6898 }
6899
6900 Node* outlen = load_array_length(out);
6901 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6902 Node* out_start = array_element_address(out, intcon(0), out_elem);
6903 Node* in_start = array_element_address(in, intcon(0), in_elem);
6904
6905 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6906 OptoRuntime::mulAdd_Type(),
6907 stubAddr, stubName, TypePtr::BOTTOM,
6908 out_start,in_start, new_offset, len, k);
6909 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6910 set_result(result);
6911 return true;
6912 }
6913
6914 //-------------inline_montgomeryMultiply-----------------------------------
6915 bool LibraryCallKit::inline_montgomeryMultiply() {
6916 address stubAddr = StubRoutines::montgomeryMultiply();
6917 if (stubAddr == nullptr) {
6918 return false; // Intrinsic's stub is not implemented on this platform
6919 }
6920
6921 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6922 const char* stubName = "montgomery_multiply";
6923
6924 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6925
6926 Node* a = argument(0);
6927 Node* b = argument(1);
6928 Node* n = argument(2);
6929 Node* len = argument(3);
6930 Node* inv = argument(4);
6931 Node* m = argument(6);
6932
6933 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6934 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6935 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6936 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6937 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6938 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6939 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6940 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6941 // failed array check
6942 return false;
6943 }
6944
6945 BasicType a_elem = a_type->elem()->array_element_basic_type();
6946 BasicType b_elem = b_type->elem()->array_element_basic_type();
6947 BasicType n_elem = n_type->elem()->array_element_basic_type();
6948 BasicType m_elem = m_type->elem()->array_element_basic_type();
6949 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6950 return false;
6951 }
6952
6953 // Make the call
6954 {
6955 Node* a_start = array_element_address(a, intcon(0), a_elem);
6956 Node* b_start = array_element_address(b, intcon(0), b_elem);
6957 Node* n_start = array_element_address(n, intcon(0), n_elem);
6958 Node* m_start = array_element_address(m, intcon(0), m_elem);
6959
6960 Node* call = make_runtime_call(RC_LEAF,
6961 OptoRuntime::montgomeryMultiply_Type(),
6962 stubAddr, stubName, TypePtr::BOTTOM,
6963 a_start, b_start, n_start, len, inv, top(),
6964 m_start);
6965 set_result(m);
6966 }
6967
6968 return true;
6969 }
6970
6971 bool LibraryCallKit::inline_montgomerySquare() {
6972 address stubAddr = StubRoutines::montgomerySquare();
6973 if (stubAddr == nullptr) {
6974 return false; // Intrinsic's stub is not implemented on this platform
6975 }
6976
6977 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6978 const char* stubName = "montgomery_square";
6979
6980 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6981
6982 Node* a = argument(0);
6983 Node* n = argument(1);
6984 Node* len = argument(2);
6985 Node* inv = argument(3);
6986 Node* m = argument(5);
6987
6988 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6989 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6990 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6991 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6992 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6993 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6994 // failed array check
6995 return false;
6996 }
6997
6998 BasicType a_elem = a_type->elem()->array_element_basic_type();
6999 BasicType n_elem = n_type->elem()->array_element_basic_type();
7000 BasicType m_elem = m_type->elem()->array_element_basic_type();
7001 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7002 return false;
7003 }
7004
7005 // Make the call
7006 {
7007 Node* a_start = array_element_address(a, intcon(0), a_elem);
7008 Node* n_start = array_element_address(n, intcon(0), n_elem);
7009 Node* m_start = array_element_address(m, intcon(0), m_elem);
7010
7011 Node* call = make_runtime_call(RC_LEAF,
7012 OptoRuntime::montgomerySquare_Type(),
7013 stubAddr, stubName, TypePtr::BOTTOM,
7014 a_start, n_start, len, inv, top(),
7015 m_start);
7016 set_result(m);
7017 }
7018
7019 return true;
7020 }
7021
7022 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7023 address stubAddr = nullptr;
7024 const char* stubName = nullptr;
7025
7026 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7027 if (stubAddr == nullptr) {
7028 return false; // Intrinsic's stub is not implemented on this platform
7029 }
7030
7031 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7032
7033 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7034
7035 Node* newArr = argument(0);
7036 Node* oldArr = argument(1);
7037 Node* newIdx = argument(2);
7038 Node* shiftCount = argument(3);
7039 Node* numIter = argument(4);
7040
7041 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7042 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7043 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7044 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7045 return false;
7046 }
7047
7048 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7049 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7050 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7051 return false;
7052 }
7053
7054 // Make the call
7055 {
7056 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7057 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7058
7059 Node* call = make_runtime_call(RC_LEAF,
7060 OptoRuntime::bigIntegerShift_Type(),
7061 stubAddr,
7062 stubName,
7063 TypePtr::BOTTOM,
7064 newArr_start,
7065 oldArr_start,
7066 newIdx,
7067 shiftCount,
7068 numIter);
7069 }
7070
7071 return true;
7072 }
7073
7074 //-------------inline_vectorizedMismatch------------------------------
7075 bool LibraryCallKit::inline_vectorizedMismatch() {
7076 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7077
7078 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7079 Node* obja = argument(0); // Object
7080 Node* aoffset = argument(1); // long
7081 Node* objb = argument(3); // Object
7082 Node* boffset = argument(4); // long
7083 Node* length = argument(6); // int
7084 Node* scale = argument(7); // int
7085
7086 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7087 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7088 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7089 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7090 scale == top()) {
7091 return false; // failed input validation
7092 }
7093
7094 Node* obja_adr = make_unsafe_address(obja, aoffset);
7095 Node* objb_adr = make_unsafe_address(objb, boffset);
7096
7097 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7098 //
7099 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7100 // if (length <= inline_limit) {
7101 // inline_path:
7102 // vmask = VectorMaskGen length
7103 // vload1 = LoadVectorMasked obja, vmask
7104 // vload2 = LoadVectorMasked objb, vmask
7105 // result1 = VectorCmpMasked vload1, vload2, vmask
7106 // } else {
7107 // call_stub_path:
7108 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7109 // }
7110 // exit_block:
7111 // return Phi(result1, result2);
7112 //
7113 enum { inline_path = 1, // input is small enough to process it all at once
7114 stub_path = 2, // input is too large; call into the VM
7115 PATH_LIMIT = 3
7116 };
7117
7118 Node* exit_block = new RegionNode(PATH_LIMIT);
7119 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7120 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7121
7122 Node* call_stub_path = control();
7123
7124 BasicType elem_bt = T_ILLEGAL;
7125
7126 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7127 if (scale_t->is_con()) {
7128 switch (scale_t->get_con()) {
7129 case 0: elem_bt = T_BYTE; break;
7130 case 1: elem_bt = T_SHORT; break;
7131 case 2: elem_bt = T_INT; break;
7132 case 3: elem_bt = T_LONG; break;
7133
7134 default: elem_bt = T_ILLEGAL; break; // not supported
7135 }
7136 }
7137
7138 int inline_limit = 0;
7139 bool do_partial_inline = false;
7140
7141 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7142 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7143 do_partial_inline = inline_limit >= 16;
7144 }
7145
7146 if (do_partial_inline) {
7147 assert(elem_bt != T_ILLEGAL, "sanity");
7148
7149 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7150 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7151 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7152
7153 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7154 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7155 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7156
7157 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7158
7159 if (!stopped()) {
7160 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7161
7162 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7163 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7164 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7165 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7166
7167 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7168 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7169 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7170 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7171
7172 exit_block->init_req(inline_path, control());
7173 memory_phi->init_req(inline_path, map()->memory());
7174 result_phi->init_req(inline_path, result);
7175
7176 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7177 clear_upper_avx();
7178 }
7179 }
7180 }
7181
7182 if (call_stub_path != nullptr) {
7183 set_control(call_stub_path);
7184
7185 Node* call = make_runtime_call(RC_LEAF,
7186 OptoRuntime::vectorizedMismatch_Type(),
7187 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7188 obja_adr, objb_adr, length, scale);
7189
7190 exit_block->init_req(stub_path, control());
7191 memory_phi->init_req(stub_path, map()->memory());
7192 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7193 }
7194
7195 exit_block = _gvn.transform(exit_block);
7196 memory_phi = _gvn.transform(memory_phi);
7197 result_phi = _gvn.transform(result_phi);
7198
7199 set_control(exit_block);
7200 set_all_memory(memory_phi);
7201 set_result(result_phi);
7202
7203 return true;
7204 }
7205
7206 //------------------------------inline_vectorizedHashcode----------------------------
7207 bool LibraryCallKit::inline_vectorizedHashCode() {
7208 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7209
7210 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7211 Node* array = argument(0);
7212 Node* offset = argument(1);
7213 Node* length = argument(2);
7214 Node* initialValue = argument(3);
7215 Node* basic_type = argument(4);
7216
7217 if (basic_type == top()) {
7218 return false; // failed input validation
7219 }
7220
7221 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7222 if (!basic_type_t->is_con()) {
7223 return false; // Only intrinsify if mode argument is constant
7224 }
7225
7226 array = must_be_not_null(array, true);
7227
7228 BasicType bt = (BasicType)basic_type_t->get_con();
7229
7230 // Resolve address of first element
7231 Node* array_start = array_element_address(array, offset, bt);
7232
7233 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7234 array_start, length, initialValue, basic_type)));
7235 clear_upper_avx();
7236
7237 return true;
7238 }
7239
7240 /**
7241 * Calculate CRC32 for byte.
7242 * int java.util.zip.CRC32.update(int crc, int b)
7243 */
7244 bool LibraryCallKit::inline_updateCRC32() {
7245 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7246 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7247 // no receiver since it is static method
7248 Node* crc = argument(0); // type: int
7249 Node* b = argument(1); // type: int
7250
7251 /*
7252 * int c = ~ crc;
7253 * b = timesXtoThe32[(b ^ c) & 0xFF];
7254 * b = b ^ (c >>> 8);
7255 * crc = ~b;
7256 */
7257
7258 Node* M1 = intcon(-1);
7259 crc = _gvn.transform(new XorINode(crc, M1));
7260 Node* result = _gvn.transform(new XorINode(crc, b));
7261 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7262
7263 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7264 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7265 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7266 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7267
7268 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7269 result = _gvn.transform(new XorINode(crc, result));
7270 result = _gvn.transform(new XorINode(result, M1));
7271 set_result(result);
7272 return true;
7273 }
7274
7275 /**
7276 * Calculate CRC32 for byte[] array.
7277 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7278 */
7279 bool LibraryCallKit::inline_updateBytesCRC32() {
7280 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7281 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7282 // no receiver since it is static method
7283 Node* crc = argument(0); // type: int
7284 Node* src = argument(1); // type: oop
7285 Node* offset = argument(2); // type: int
7286 Node* length = argument(3); // type: int
7287
7288 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7289 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7290 // failed array check
7291 return false;
7292 }
7293
7294 // Figure out the size and type of the elements we will be copying.
7295 BasicType src_elem = src_type->elem()->array_element_basic_type();
7296 if (src_elem != T_BYTE) {
7297 return false;
7298 }
7299
7300 // 'src_start' points to src array + scaled offset
7301 src = must_be_not_null(src, true);
7302 Node* src_start = array_element_address(src, offset, src_elem);
7303
7304 // We assume that range check is done by caller.
7305 // TODO: generate range check (offset+length < src.length) in debug VM.
7306
7307 // Call the stub.
7308 address stubAddr = StubRoutines::updateBytesCRC32();
7309 const char *stubName = "updateBytesCRC32";
7310
7311 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7312 stubAddr, stubName, TypePtr::BOTTOM,
7313 crc, src_start, length);
7314 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7315 set_result(result);
7316 return true;
7317 }
7318
7319 /**
7320 * Calculate CRC32 for ByteBuffer.
7321 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7322 */
7323 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7324 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7325 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7326 // no receiver since it is static method
7327 Node* crc = argument(0); // type: int
7328 Node* src = argument(1); // type: long
7329 Node* offset = argument(3); // type: int
7330 Node* length = argument(4); // type: int
7331
7332 src = ConvL2X(src); // adjust Java long to machine word
7333 Node* base = _gvn.transform(new CastX2PNode(src));
7334 offset = ConvI2X(offset);
7335
7336 // 'src_start' points to src array + scaled offset
7337 Node* src_start = basic_plus_adr(top(), base, offset);
7338
7339 // Call the stub.
7340 address stubAddr = StubRoutines::updateBytesCRC32();
7341 const char *stubName = "updateBytesCRC32";
7342
7343 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7344 stubAddr, stubName, TypePtr::BOTTOM,
7345 crc, src_start, length);
7346 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7347 set_result(result);
7348 return true;
7349 }
7350
7351 //------------------------------get_table_from_crc32c_class-----------------------
7352 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7353 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7354 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7355
7356 return table;
7357 }
7358
7359 //------------------------------inline_updateBytesCRC32C-----------------------
7360 //
7361 // Calculate CRC32C for byte[] array.
7362 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7363 //
7364 bool LibraryCallKit::inline_updateBytesCRC32C() {
7365 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7366 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7367 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7368 // no receiver since it is a static method
7369 Node* crc = argument(0); // type: int
7370 Node* src = argument(1); // type: oop
7371 Node* offset = argument(2); // type: int
7372 Node* end = argument(3); // type: int
7373
7374 Node* length = _gvn.transform(new SubINode(end, offset));
7375
7376 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7377 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7378 // failed array check
7379 return false;
7380 }
7381
7382 // Figure out the size and type of the elements we will be copying.
7383 BasicType src_elem = src_type->elem()->array_element_basic_type();
7384 if (src_elem != T_BYTE) {
7385 return false;
7386 }
7387
7388 // 'src_start' points to src array + scaled offset
7389 src = must_be_not_null(src, true);
7390 Node* src_start = array_element_address(src, offset, src_elem);
7391
7392 // static final int[] byteTable in class CRC32C
7393 Node* table = get_table_from_crc32c_class(callee()->holder());
7394 table = must_be_not_null(table, true);
7395 Node* table_start = array_element_address(table, intcon(0), T_INT);
7396
7397 // We assume that range check is done by caller.
7398 // TODO: generate range check (offset+length < src.length) in debug VM.
7399
7400 // Call the stub.
7401 address stubAddr = StubRoutines::updateBytesCRC32C();
7402 const char *stubName = "updateBytesCRC32C";
7403
7404 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7405 stubAddr, stubName, TypePtr::BOTTOM,
7406 crc, src_start, length, table_start);
7407 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7408 set_result(result);
7409 return true;
7410 }
7411
7412 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7413 //
7414 // Calculate CRC32C for DirectByteBuffer.
7415 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7416 //
7417 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7418 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7419 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7420 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7421 // no receiver since it is a static method
7422 Node* crc = argument(0); // type: int
7423 Node* src = argument(1); // type: long
7424 Node* offset = argument(3); // type: int
7425 Node* end = argument(4); // type: int
7426
7427 Node* length = _gvn.transform(new SubINode(end, offset));
7428
7429 src = ConvL2X(src); // adjust Java long to machine word
7430 Node* base = _gvn.transform(new CastX2PNode(src));
7431 offset = ConvI2X(offset);
7432
7433 // 'src_start' points to src array + scaled offset
7434 Node* src_start = basic_plus_adr(top(), base, offset);
7435
7436 // static final int[] byteTable in class CRC32C
7437 Node* table = get_table_from_crc32c_class(callee()->holder());
7438 table = must_be_not_null(table, true);
7439 Node* table_start = array_element_address(table, intcon(0), T_INT);
7440
7441 // Call the stub.
7442 address stubAddr = StubRoutines::updateBytesCRC32C();
7443 const char *stubName = "updateBytesCRC32C";
7444
7445 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7446 stubAddr, stubName, TypePtr::BOTTOM,
7447 crc, src_start, length, table_start);
7448 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7449 set_result(result);
7450 return true;
7451 }
7452
7453 //------------------------------inline_updateBytesAdler32----------------------
7454 //
7455 // Calculate Adler32 checksum for byte[] array.
7456 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7457 //
7458 bool LibraryCallKit::inline_updateBytesAdler32() {
7459 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7460 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7461 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7462 // no receiver since it is static method
7463 Node* crc = argument(0); // type: int
7464 Node* src = argument(1); // type: oop
7465 Node* offset = argument(2); // type: int
7466 Node* length = argument(3); // type: int
7467
7468 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7469 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7470 // failed array check
7471 return false;
7472 }
7473
7474 // Figure out the size and type of the elements we will be copying.
7475 BasicType src_elem = src_type->elem()->array_element_basic_type();
7476 if (src_elem != T_BYTE) {
7477 return false;
7478 }
7479
7480 // 'src_start' points to src array + scaled offset
7481 Node* src_start = array_element_address(src, offset, src_elem);
7482
7483 // We assume that range check is done by caller.
7484 // TODO: generate range check (offset+length < src.length) in debug VM.
7485
7486 // Call the stub.
7487 address stubAddr = StubRoutines::updateBytesAdler32();
7488 const char *stubName = "updateBytesAdler32";
7489
7490 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7491 stubAddr, stubName, TypePtr::BOTTOM,
7492 crc, src_start, length);
7493 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7494 set_result(result);
7495 return true;
7496 }
7497
7498 //------------------------------inline_updateByteBufferAdler32---------------
7499 //
7500 // Calculate Adler32 checksum for DirectByteBuffer.
7501 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7502 //
7503 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7504 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7505 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7506 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7507 // no receiver since it is static method
7508 Node* crc = argument(0); // type: int
7509 Node* src = argument(1); // type: long
7510 Node* offset = argument(3); // type: int
7511 Node* length = argument(4); // type: int
7512
7513 src = ConvL2X(src); // adjust Java long to machine word
7514 Node* base = _gvn.transform(new CastX2PNode(src));
7515 offset = ConvI2X(offset);
7516
7517 // 'src_start' points to src array + scaled offset
7518 Node* src_start = basic_plus_adr(top(), base, offset);
7519
7520 // Call the stub.
7521 address stubAddr = StubRoutines::updateBytesAdler32();
7522 const char *stubName = "updateBytesAdler32";
7523
7524 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7525 stubAddr, stubName, TypePtr::BOTTOM,
7526 crc, src_start, length);
7527
7528 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7529 set_result(result);
7530 return true;
7531 }
7532
7533 //----------------------------inline_reference_get----------------------------
7534 // public T java.lang.ref.Reference.get();
7535 bool LibraryCallKit::inline_reference_get() {
7536 const int referent_offset = java_lang_ref_Reference::referent_offset();
7537
7538 // Get the argument:
7539 Node* reference_obj = null_check_receiver();
7540 if (stopped()) return true;
7541
7542 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7543 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7544 decorators, /*is_static*/ false, nullptr);
7545 if (result == nullptr) return false;
7546
7547 // Add memory barrier to prevent commoning reads from this field
7548 // across safepoint since GC can change its value.
7549 insert_mem_bar(Op_MemBarCPUOrder);
7550
7551 set_result(result);
7552 return true;
7553 }
7554
7555 //----------------------------inline_reference_refersTo0----------------------------
7556 // bool java.lang.ref.Reference.refersTo0();
7557 // bool java.lang.ref.PhantomReference.refersTo0();
7558 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7559 // Get arguments:
7560 Node* reference_obj = null_check_receiver();
7561 Node* other_obj = argument(1);
7562 if (stopped()) return true;
7563
7564 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7565 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7566 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7567 decorators, /*is_static*/ false, nullptr);
7568 if (referent == nullptr) return false;
7569
7570 // Add memory barrier to prevent commoning reads from this field
7571 // across safepoint since GC can change its value.
7572 insert_mem_bar(Op_MemBarCPUOrder);
7573
7574 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7575 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7576 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7577
7578 RegionNode* region = new RegionNode(3);
7579 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7580
7581 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7582 region->init_req(1, if_true);
7583 phi->init_req(1, intcon(1));
7584
7585 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7586 region->init_req(2, if_false);
7587 phi->init_req(2, intcon(0));
7588
7589 set_control(_gvn.transform(region));
7590 record_for_igvn(region);
7591 set_result(_gvn.transform(phi));
7592 return true;
7593 }
7594
7595 //----------------------------inline_reference_clear0----------------------------
7596 // void java.lang.ref.Reference.clear0();
7597 // void java.lang.ref.PhantomReference.clear0();
7598 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7599 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7600
7601 // Get arguments
7602 Node* reference_obj = null_check_receiver();
7603 if (stopped()) return true;
7604
7605 // Common access parameters
7606 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7607 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7608 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7609 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7610 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7611
7612 Node* referent = access_load_at(reference_obj,
7613 referent_field_addr,
7614 referent_field_addr_type,
7615 val_type,
7616 T_OBJECT,
7617 decorators);
7618
7619 IdealKit ideal(this);
7620 #define __ ideal.
7621 __ if_then(referent, BoolTest::ne, null());
7622 sync_kit(ideal);
7623 access_store_at(reference_obj,
7624 referent_field_addr,
7625 referent_field_addr_type,
7626 null(),
7627 val_type,
7628 T_OBJECT,
7629 decorators);
7630 __ sync_kit(this);
7631 __ end_if();
7632 final_sync(ideal);
7633 #undef __
7634
7635 return true;
7636 }
7637
7638 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7639 DecoratorSet decorators, bool is_static,
7640 ciInstanceKlass* fromKls) {
7641 if (fromKls == nullptr) {
7642 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7643 assert(tinst != nullptr, "obj is null");
7644 assert(tinst->is_loaded(), "obj is not loaded");
7645 fromKls = tinst->instance_klass();
7646 } else {
7647 assert(is_static, "only for static field access");
7648 }
7649 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7650 ciSymbol::make(fieldTypeString),
7651 is_static);
7652
7653 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7654 if (field == nullptr) return (Node *) nullptr;
7655
7656 if (is_static) {
7657 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7658 fromObj = makecon(tip);
7659 }
7660
7661 // Next code copied from Parse::do_get_xxx():
7662
7663 // Compute address and memory type.
7664 int offset = field->offset_in_bytes();
7665 bool is_vol = field->is_volatile();
7666 ciType* field_klass = field->type();
7667 assert(field_klass->is_loaded(), "should be loaded");
7668 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7669 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7670 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7671 "slice of address and input slice don't match");
7672 BasicType bt = field->layout_type();
7673
7674 // Build the resultant type of the load
7675 const Type *type;
7676 if (bt == T_OBJECT) {
7677 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7678 } else {
7679 type = Type::get_const_basic_type(bt);
7680 }
7681
7682 if (is_vol) {
7683 decorators |= MO_SEQ_CST;
7684 }
7685
7686 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7687 }
7688
7689 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7690 bool is_exact /* true */, bool is_static /* false */,
7691 ciInstanceKlass * fromKls /* nullptr */) {
7692 if (fromKls == nullptr) {
7693 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7694 assert(tinst != nullptr, "obj is null");
7695 assert(tinst->is_loaded(), "obj is not loaded");
7696 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7697 fromKls = tinst->instance_klass();
7698 }
7699 else {
7700 assert(is_static, "only for static field access");
7701 }
7702 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7703 ciSymbol::make(fieldTypeString),
7704 is_static);
7705
7706 assert(field != nullptr, "undefined field");
7707 assert(!field->is_volatile(), "not defined for volatile fields");
7708
7709 if (is_static) {
7710 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7711 fromObj = makecon(tip);
7712 }
7713
7714 // Next code copied from Parse::do_get_xxx():
7715
7716 // Compute address and memory type.
7717 int offset = field->offset_in_bytes();
7718 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7719
7720 return adr;
7721 }
7722
7723 //------------------------------inline_aescrypt_Block-----------------------
7724 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7725 address stubAddr = nullptr;
7726 const char *stubName;
7727 assert(UseAES, "need AES instruction support");
7728
7729 switch(id) {
7730 case vmIntrinsics::_aescrypt_encryptBlock:
7731 stubAddr = StubRoutines::aescrypt_encryptBlock();
7732 stubName = "aescrypt_encryptBlock";
7733 break;
7734 case vmIntrinsics::_aescrypt_decryptBlock:
7735 stubAddr = StubRoutines::aescrypt_decryptBlock();
7736 stubName = "aescrypt_decryptBlock";
7737 break;
7738 default:
7739 break;
7740 }
7741 if (stubAddr == nullptr) return false;
7742
7743 Node* aescrypt_object = argument(0);
7744 Node* src = argument(1);
7745 Node* src_offset = argument(2);
7746 Node* dest = argument(3);
7747 Node* dest_offset = argument(4);
7748
7749 src = must_be_not_null(src, true);
7750 dest = must_be_not_null(dest, true);
7751
7752 // (1) src and dest are arrays.
7753 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7754 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7755 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7756 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7757
7758 // for the quick and dirty code we will skip all the checks.
7759 // we are just trying to get the call to be generated.
7760 Node* src_start = src;
7761 Node* dest_start = dest;
7762 if (src_offset != nullptr || dest_offset != nullptr) {
7763 assert(src_offset != nullptr && dest_offset != nullptr, "");
7764 src_start = array_element_address(src, src_offset, T_BYTE);
7765 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7766 }
7767
7768 // now need to get the start of its expanded key array
7769 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7770 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7771 if (k_start == nullptr) return false;
7772
7773 // Call the stub.
7774 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7775 stubAddr, stubName, TypePtr::BOTTOM,
7776 src_start, dest_start, k_start);
7777
7778 return true;
7779 }
7780
7781 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7782 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7783 address stubAddr = nullptr;
7784 const char *stubName = nullptr;
7785
7786 assert(UseAES, "need AES instruction support");
7787
7788 switch(id) {
7789 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7790 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7791 stubName = "cipherBlockChaining_encryptAESCrypt";
7792 break;
7793 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7794 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7795 stubName = "cipherBlockChaining_decryptAESCrypt";
7796 break;
7797 default:
7798 break;
7799 }
7800 if (stubAddr == nullptr) return false;
7801
7802 Node* cipherBlockChaining_object = argument(0);
7803 Node* src = argument(1);
7804 Node* src_offset = argument(2);
7805 Node* len = argument(3);
7806 Node* dest = argument(4);
7807 Node* dest_offset = argument(5);
7808
7809 src = must_be_not_null(src, false);
7810 dest = must_be_not_null(dest, false);
7811
7812 // (1) src and dest are arrays.
7813 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7814 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7815 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7816 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7817
7818 // checks are the responsibility of the caller
7819 Node* src_start = src;
7820 Node* dest_start = dest;
7821 if (src_offset != nullptr || dest_offset != nullptr) {
7822 assert(src_offset != nullptr && dest_offset != nullptr, "");
7823 src_start = array_element_address(src, src_offset, T_BYTE);
7824 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7825 }
7826
7827 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7828 // (because of the predicated logic executed earlier).
7829 // so we cast it here safely.
7830 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7831
7832 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7833 if (embeddedCipherObj == nullptr) return false;
7834
7835 // cast it to what we know it will be at runtime
7836 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7837 assert(tinst != nullptr, "CBC obj is null");
7838 assert(tinst->is_loaded(), "CBC obj is not loaded");
7839 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7840 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7841
7842 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7843 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7844 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7845 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7846 aescrypt_object = _gvn.transform(aescrypt_object);
7847
7848 // we need to get the start of the aescrypt_object's expanded key array
7849 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7850 if (k_start == nullptr) return false;
7851
7852 // similarly, get the start address of the r vector
7853 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7854 if (objRvec == nullptr) return false;
7855 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7856
7857 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7858 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7859 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7860 stubAddr, stubName, TypePtr::BOTTOM,
7861 src_start, dest_start, k_start, r_start, len);
7862
7863 // return cipher length (int)
7864 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7865 set_result(retvalue);
7866 return true;
7867 }
7868
7869 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7870 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7871 address stubAddr = nullptr;
7872 const char *stubName = nullptr;
7873
7874 assert(UseAES, "need AES instruction support");
7875
7876 switch (id) {
7877 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7878 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7879 stubName = "electronicCodeBook_encryptAESCrypt";
7880 break;
7881 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7882 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7883 stubName = "electronicCodeBook_decryptAESCrypt";
7884 break;
7885 default:
7886 break;
7887 }
7888
7889 if (stubAddr == nullptr) return false;
7890
7891 Node* electronicCodeBook_object = argument(0);
7892 Node* src = argument(1);
7893 Node* src_offset = argument(2);
7894 Node* len = argument(3);
7895 Node* dest = argument(4);
7896 Node* dest_offset = argument(5);
7897
7898 // (1) src and dest are arrays.
7899 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7900 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7901 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7902 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7903
7904 // checks are the responsibility of the caller
7905 Node* src_start = src;
7906 Node* dest_start = dest;
7907 if (src_offset != nullptr || dest_offset != nullptr) {
7908 assert(src_offset != nullptr && dest_offset != nullptr, "");
7909 src_start = array_element_address(src, src_offset, T_BYTE);
7910 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7911 }
7912
7913 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7914 // (because of the predicated logic executed earlier).
7915 // so we cast it here safely.
7916 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7917
7918 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7919 if (embeddedCipherObj == nullptr) return false;
7920
7921 // cast it to what we know it will be at runtime
7922 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7923 assert(tinst != nullptr, "ECB obj is null");
7924 assert(tinst->is_loaded(), "ECB obj is not loaded");
7925 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7926 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7927
7928 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7929 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7930 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7931 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7932 aescrypt_object = _gvn.transform(aescrypt_object);
7933
7934 // we need to get the start of the aescrypt_object's expanded key array
7935 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7936 if (k_start == nullptr) return false;
7937
7938 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7939 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7940 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7941 stubAddr, stubName, TypePtr::BOTTOM,
7942 src_start, dest_start, k_start, len);
7943
7944 // return cipher length (int)
7945 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7946 set_result(retvalue);
7947 return true;
7948 }
7949
7950 //------------------------------inline_counterMode_AESCrypt-----------------------
7951 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7952 assert(UseAES, "need AES instruction support");
7953 if (!UseAESCTRIntrinsics) return false;
7954
7955 address stubAddr = nullptr;
7956 const char *stubName = nullptr;
7957 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7958 stubAddr = StubRoutines::counterMode_AESCrypt();
7959 stubName = "counterMode_AESCrypt";
7960 }
7961 if (stubAddr == nullptr) return false;
7962
7963 Node* counterMode_object = argument(0);
7964 Node* src = argument(1);
7965 Node* src_offset = argument(2);
7966 Node* len = argument(3);
7967 Node* dest = argument(4);
7968 Node* dest_offset = argument(5);
7969
7970 // (1) src and dest are arrays.
7971 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7972 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7973 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7974 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7975
7976 // checks are the responsibility of the caller
7977 Node* src_start = src;
7978 Node* dest_start = dest;
7979 if (src_offset != nullptr || dest_offset != nullptr) {
7980 assert(src_offset != nullptr && dest_offset != nullptr, "");
7981 src_start = array_element_address(src, src_offset, T_BYTE);
7982 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7983 }
7984
7985 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7986 // (because of the predicated logic executed earlier).
7987 // so we cast it here safely.
7988 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7989 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7990 if (embeddedCipherObj == nullptr) return false;
7991 // cast it to what we know it will be at runtime
7992 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7993 assert(tinst != nullptr, "CTR obj is null");
7994 assert(tinst->is_loaded(), "CTR obj is not loaded");
7995 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7996 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7997 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7998 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7999 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8000 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8001 aescrypt_object = _gvn.transform(aescrypt_object);
8002 // we need to get the start of the aescrypt_object's expanded key array
8003 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8004 if (k_start == nullptr) return false;
8005 // similarly, get the start address of the r vector
8006 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8007 if (obj_counter == nullptr) return false;
8008 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8009
8010 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8011 if (saved_encCounter == nullptr) return false;
8012 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8013 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8014
8015 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8016 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8017 OptoRuntime::counterMode_aescrypt_Type(),
8018 stubAddr, stubName, TypePtr::BOTTOM,
8019 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8020
8021 // return cipher length (int)
8022 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8023 set_result(retvalue);
8024 return true;
8025 }
8026
8027 //------------------------------get_key_start_from_aescrypt_object-----------------------
8028 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
8029 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8030 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8031 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8032 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8033 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
8034 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
8035 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
8036 if (objSessionK == nullptr) {
8037 return (Node *) nullptr;
8038 }
8039 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
8040 #else
8041 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
8042 #endif // PPC64
8043 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
8044 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8045
8046 // now have the array, need to get the start address of the K array
8047 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8048 return k_start;
8049 }
8050
8051 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8052 // Return node representing slow path of predicate check.
8053 // the pseudo code we want to emulate with this predicate is:
8054 // for encryption:
8055 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8056 // for decryption:
8057 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8058 // note cipher==plain is more conservative than the original java code but that's OK
8059 //
8060 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8061 // The receiver was checked for null already.
8062 Node* objCBC = argument(0);
8063
8064 Node* src = argument(1);
8065 Node* dest = argument(4);
8066
8067 // Load embeddedCipher field of CipherBlockChaining object.
8068 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8069
8070 // get AESCrypt klass for instanceOf check
8071 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8072 // will have same classloader as CipherBlockChaining object
8073 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8074 assert(tinst != nullptr, "CBCobj is null");
8075 assert(tinst->is_loaded(), "CBCobj is not loaded");
8076
8077 // we want to do an instanceof comparison against the AESCrypt class
8078 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8079 if (!klass_AESCrypt->is_loaded()) {
8080 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8081 Node* ctrl = control();
8082 set_control(top()); // no regular fast path
8083 return ctrl;
8084 }
8085
8086 src = must_be_not_null(src, true);
8087 dest = must_be_not_null(dest, true);
8088
8089 // Resolve oops to stable for CmpP below.
8090 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8091
8092 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8093 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8094 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8095
8096 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8097
8098 // for encryption, we are done
8099 if (!decrypting)
8100 return instof_false; // even if it is null
8101
8102 // for decryption, we need to add a further check to avoid
8103 // taking the intrinsic path when cipher and plain are the same
8104 // see the original java code for why.
8105 RegionNode* region = new RegionNode(3);
8106 region->init_req(1, instof_false);
8107
8108 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8109 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8110 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8111 region->init_req(2, src_dest_conjoint);
8112
8113 record_for_igvn(region);
8114 return _gvn.transform(region);
8115 }
8116
8117 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8118 // Return node representing slow path of predicate check.
8119 // the pseudo code we want to emulate with this predicate is:
8120 // for encryption:
8121 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8122 // for decryption:
8123 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8124 // note cipher==plain is more conservative than the original java code but that's OK
8125 //
8126 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8127 // The receiver was checked for null already.
8128 Node* objECB = argument(0);
8129
8130 // Load embeddedCipher field of ElectronicCodeBook object.
8131 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8132
8133 // get AESCrypt klass for instanceOf check
8134 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8135 // will have same classloader as ElectronicCodeBook object
8136 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8137 assert(tinst != nullptr, "ECBobj is null");
8138 assert(tinst->is_loaded(), "ECBobj is not loaded");
8139
8140 // we want to do an instanceof comparison against the AESCrypt class
8141 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8142 if (!klass_AESCrypt->is_loaded()) {
8143 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8144 Node* ctrl = control();
8145 set_control(top()); // no regular fast path
8146 return ctrl;
8147 }
8148 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8149
8150 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8151 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8152 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8153
8154 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8155
8156 // for encryption, we are done
8157 if (!decrypting)
8158 return instof_false; // even if it is null
8159
8160 // for decryption, we need to add a further check to avoid
8161 // taking the intrinsic path when cipher and plain are the same
8162 // see the original java code for why.
8163 RegionNode* region = new RegionNode(3);
8164 region->init_req(1, instof_false);
8165 Node* src = argument(1);
8166 Node* dest = argument(4);
8167 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8168 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8169 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8170 region->init_req(2, src_dest_conjoint);
8171
8172 record_for_igvn(region);
8173 return _gvn.transform(region);
8174 }
8175
8176 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8177 // Return node representing slow path of predicate check.
8178 // the pseudo code we want to emulate with this predicate is:
8179 // for encryption:
8180 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8181 // for decryption:
8182 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8183 // note cipher==plain is more conservative than the original java code but that's OK
8184 //
8185
8186 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8187 // The receiver was checked for null already.
8188 Node* objCTR = argument(0);
8189
8190 // Load embeddedCipher field of CipherBlockChaining object.
8191 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8192
8193 // get AESCrypt klass for instanceOf check
8194 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8195 // will have same classloader as CipherBlockChaining object
8196 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8197 assert(tinst != nullptr, "CTRobj is null");
8198 assert(tinst->is_loaded(), "CTRobj is not loaded");
8199
8200 // we want to do an instanceof comparison against the AESCrypt class
8201 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8202 if (!klass_AESCrypt->is_loaded()) {
8203 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8204 Node* ctrl = control();
8205 set_control(top()); // no regular fast path
8206 return ctrl;
8207 }
8208
8209 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8210 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8211 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8212 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8213 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8214
8215 return instof_false; // even if it is null
8216 }
8217
8218 //------------------------------inline_ghash_processBlocks
8219 bool LibraryCallKit::inline_ghash_processBlocks() {
8220 address stubAddr;
8221 const char *stubName;
8222 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8223
8224 stubAddr = StubRoutines::ghash_processBlocks();
8225 stubName = "ghash_processBlocks";
8226
8227 Node* data = argument(0);
8228 Node* offset = argument(1);
8229 Node* len = argument(2);
8230 Node* state = argument(3);
8231 Node* subkeyH = argument(4);
8232
8233 state = must_be_not_null(state, true);
8234 subkeyH = must_be_not_null(subkeyH, true);
8235 data = must_be_not_null(data, true);
8236
8237 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8238 assert(state_start, "state is null");
8239 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8240 assert(subkeyH_start, "subkeyH is null");
8241 Node* data_start = array_element_address(data, offset, T_BYTE);
8242 assert(data_start, "data is null");
8243
8244 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8245 OptoRuntime::ghash_processBlocks_Type(),
8246 stubAddr, stubName, TypePtr::BOTTOM,
8247 state_start, subkeyH_start, data_start, len);
8248 return true;
8249 }
8250
8251 //------------------------------inline_chacha20Block
8252 bool LibraryCallKit::inline_chacha20Block() {
8253 address stubAddr;
8254 const char *stubName;
8255 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8256
8257 stubAddr = StubRoutines::chacha20Block();
8258 stubName = "chacha20Block";
8259
8260 Node* state = argument(0);
8261 Node* result = argument(1);
8262
8263 state = must_be_not_null(state, true);
8264 result = must_be_not_null(result, true);
8265
8266 Node* state_start = array_element_address(state, intcon(0), T_INT);
8267 assert(state_start, "state is null");
8268 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8269 assert(result_start, "result is null");
8270
8271 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8272 OptoRuntime::chacha20Block_Type(),
8273 stubAddr, stubName, TypePtr::BOTTOM,
8274 state_start, result_start);
8275 // return key stream length (int)
8276 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8277 set_result(retvalue);
8278 return true;
8279 }
8280
8281 //------------------------------inline_kyberNtt
8282 bool LibraryCallKit::inline_kyberNtt() {
8283 address stubAddr;
8284 const char *stubName;
8285 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8286 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8287
8288 stubAddr = StubRoutines::kyberNtt();
8289 stubName = "kyberNtt";
8290 if (!stubAddr) return false;
8291
8292 Node* coeffs = argument(0);
8293 Node* ntt_zetas = argument(1);
8294
8295 coeffs = must_be_not_null(coeffs, true);
8296 ntt_zetas = must_be_not_null(ntt_zetas, true);
8297
8298 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8299 assert(coeffs_start, "coeffs is null");
8300 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8301 assert(ntt_zetas_start, "ntt_zetas is null");
8302 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8303 OptoRuntime::kyberNtt_Type(),
8304 stubAddr, stubName, TypePtr::BOTTOM,
8305 coeffs_start, ntt_zetas_start);
8306 // return an int
8307 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8308 set_result(retvalue);
8309 return true;
8310 }
8311
8312 //------------------------------inline_kyberInverseNtt
8313 bool LibraryCallKit::inline_kyberInverseNtt() {
8314 address stubAddr;
8315 const char *stubName;
8316 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8317 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8318
8319 stubAddr = StubRoutines::kyberInverseNtt();
8320 stubName = "kyberInverseNtt";
8321 if (!stubAddr) return false;
8322
8323 Node* coeffs = argument(0);
8324 Node* zetas = argument(1);
8325
8326 coeffs = must_be_not_null(coeffs, true);
8327 zetas = must_be_not_null(zetas, true);
8328
8329 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8330 assert(coeffs_start, "coeffs is null");
8331 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8332 assert(zetas_start, "inverseNtt_zetas is null");
8333 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8334 OptoRuntime::kyberInverseNtt_Type(),
8335 stubAddr, stubName, TypePtr::BOTTOM,
8336 coeffs_start, zetas_start);
8337
8338 // return an int
8339 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8340 set_result(retvalue);
8341 return true;
8342 }
8343
8344 //------------------------------inline_kyberNttMult
8345 bool LibraryCallKit::inline_kyberNttMult() {
8346 address stubAddr;
8347 const char *stubName;
8348 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8349 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8350
8351 stubAddr = StubRoutines::kyberNttMult();
8352 stubName = "kyberNttMult";
8353 if (!stubAddr) return false;
8354
8355 Node* result = argument(0);
8356 Node* ntta = argument(1);
8357 Node* nttb = argument(2);
8358 Node* zetas = argument(3);
8359
8360 result = must_be_not_null(result, true);
8361 ntta = must_be_not_null(ntta, true);
8362 nttb = must_be_not_null(nttb, true);
8363 zetas = must_be_not_null(zetas, true);
8364 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8365 assert(result_start, "result is null");
8366 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8367 assert(ntta_start, "ntta is null");
8368 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8369 assert(nttb_start, "nttb is null");
8370 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8371 assert(zetas_start, "nttMult_zetas is null");
8372 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8373 OptoRuntime::kyberNttMult_Type(),
8374 stubAddr, stubName, TypePtr::BOTTOM,
8375 result_start, ntta_start, nttb_start,
8376 zetas_start);
8377
8378 // return an int
8379 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8380 set_result(retvalue);
8381
8382 return true;
8383 }
8384
8385 //------------------------------inline_kyberAddPoly_2
8386 bool LibraryCallKit::inline_kyberAddPoly_2() {
8387 address stubAddr;
8388 const char *stubName;
8389 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8390 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8391
8392 stubAddr = StubRoutines::kyberAddPoly_2();
8393 stubName = "kyberAddPoly_2";
8394 if (!stubAddr) return false;
8395
8396 Node* result = argument(0);
8397 Node* a = argument(1);
8398 Node* b = argument(2);
8399
8400 result = must_be_not_null(result, true);
8401 a = must_be_not_null(a, true);
8402 b = must_be_not_null(b, true);
8403
8404 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8405 assert(result_start, "result is null");
8406 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8407 assert(a_start, "a is null");
8408 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8409 assert(b_start, "b is null");
8410 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8411 OptoRuntime::kyberAddPoly_2_Type(),
8412 stubAddr, stubName, TypePtr::BOTTOM,
8413 result_start, a_start, b_start);
8414 // return an int
8415 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8416 set_result(retvalue);
8417 return true;
8418 }
8419
8420 //------------------------------inline_kyberAddPoly_3
8421 bool LibraryCallKit::inline_kyberAddPoly_3() {
8422 address stubAddr;
8423 const char *stubName;
8424 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8425 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8426
8427 stubAddr = StubRoutines::kyberAddPoly_3();
8428 stubName = "kyberAddPoly_3";
8429 if (!stubAddr) return false;
8430
8431 Node* result = argument(0);
8432 Node* a = argument(1);
8433 Node* b = argument(2);
8434 Node* c = argument(3);
8435
8436 result = must_be_not_null(result, true);
8437 a = must_be_not_null(a, true);
8438 b = must_be_not_null(b, true);
8439 c = must_be_not_null(c, true);
8440
8441 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8442 assert(result_start, "result is null");
8443 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8444 assert(a_start, "a is null");
8445 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8446 assert(b_start, "b is null");
8447 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8448 assert(c_start, "c is null");
8449 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8450 OptoRuntime::kyberAddPoly_3_Type(),
8451 stubAddr, stubName, TypePtr::BOTTOM,
8452 result_start, a_start, b_start, c_start);
8453 // return an int
8454 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8455 set_result(retvalue);
8456 return true;
8457 }
8458
8459 //------------------------------inline_kyber12To16
8460 bool LibraryCallKit::inline_kyber12To16() {
8461 address stubAddr;
8462 const char *stubName;
8463 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8464 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8465
8466 stubAddr = StubRoutines::kyber12To16();
8467 stubName = "kyber12To16";
8468 if (!stubAddr) return false;
8469
8470 Node* condensed = argument(0);
8471 Node* condensedOffs = argument(1);
8472 Node* parsed = argument(2);
8473 Node* parsedLength = argument(3);
8474
8475 condensed = must_be_not_null(condensed, true);
8476 parsed = must_be_not_null(parsed, true);
8477
8478 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8479 assert(condensed_start, "condensed is null");
8480 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8481 assert(parsed_start, "parsed is null");
8482 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8483 OptoRuntime::kyber12To16_Type(),
8484 stubAddr, stubName, TypePtr::BOTTOM,
8485 condensed_start, condensedOffs, parsed_start, parsedLength);
8486 // return an int
8487 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8488 set_result(retvalue);
8489 return true;
8490
8491 }
8492
8493 //------------------------------inline_kyberBarrettReduce
8494 bool LibraryCallKit::inline_kyberBarrettReduce() {
8495 address stubAddr;
8496 const char *stubName;
8497 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8498 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8499
8500 stubAddr = StubRoutines::kyberBarrettReduce();
8501 stubName = "kyberBarrettReduce";
8502 if (!stubAddr) return false;
8503
8504 Node* coeffs = argument(0);
8505
8506 coeffs = must_be_not_null(coeffs, true);
8507
8508 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8509 assert(coeffs_start, "coeffs is null");
8510 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8511 OptoRuntime::kyberBarrettReduce_Type(),
8512 stubAddr, stubName, TypePtr::BOTTOM,
8513 coeffs_start);
8514 // return an int
8515 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8516 set_result(retvalue);
8517 return true;
8518 }
8519
8520 //------------------------------inline_dilithiumAlmostNtt
8521 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8522 address stubAddr;
8523 const char *stubName;
8524 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8525 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8526
8527 stubAddr = StubRoutines::dilithiumAlmostNtt();
8528 stubName = "dilithiumAlmostNtt";
8529 if (!stubAddr) return false;
8530
8531 Node* coeffs = argument(0);
8532 Node* ntt_zetas = argument(1);
8533
8534 coeffs = must_be_not_null(coeffs, true);
8535 ntt_zetas = must_be_not_null(ntt_zetas, true);
8536
8537 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8538 assert(coeffs_start, "coeffs is null");
8539 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8540 assert(ntt_zetas_start, "ntt_zetas is null");
8541 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8542 OptoRuntime::dilithiumAlmostNtt_Type(),
8543 stubAddr, stubName, TypePtr::BOTTOM,
8544 coeffs_start, ntt_zetas_start);
8545 // return an int
8546 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8547 set_result(retvalue);
8548 return true;
8549 }
8550
8551 //------------------------------inline_dilithiumAlmostInverseNtt
8552 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8553 address stubAddr;
8554 const char *stubName;
8555 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8556 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8557
8558 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8559 stubName = "dilithiumAlmostInverseNtt";
8560 if (!stubAddr) return false;
8561
8562 Node* coeffs = argument(0);
8563 Node* zetas = argument(1);
8564
8565 coeffs = must_be_not_null(coeffs, true);
8566 zetas = must_be_not_null(zetas, true);
8567
8568 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8569 assert(coeffs_start, "coeffs is null");
8570 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8571 assert(zetas_start, "inverseNtt_zetas is null");
8572 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8573 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8574 stubAddr, stubName, TypePtr::BOTTOM,
8575 coeffs_start, zetas_start);
8576 // return an int
8577 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8578 set_result(retvalue);
8579 return true;
8580 }
8581
8582 //------------------------------inline_dilithiumNttMult
8583 bool LibraryCallKit::inline_dilithiumNttMult() {
8584 address stubAddr;
8585 const char *stubName;
8586 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8587 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8588
8589 stubAddr = StubRoutines::dilithiumNttMult();
8590 stubName = "dilithiumNttMult";
8591 if (!stubAddr) return false;
8592
8593 Node* result = argument(0);
8594 Node* ntta = argument(1);
8595 Node* nttb = argument(2);
8596 Node* zetas = argument(3);
8597
8598 result = must_be_not_null(result, true);
8599 ntta = must_be_not_null(ntta, true);
8600 nttb = must_be_not_null(nttb, true);
8601 zetas = must_be_not_null(zetas, true);
8602
8603 Node* result_start = array_element_address(result, intcon(0), T_INT);
8604 assert(result_start, "result is null");
8605 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8606 assert(ntta_start, "ntta is null");
8607 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8608 assert(nttb_start, "nttb is null");
8609 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8610 OptoRuntime::dilithiumNttMult_Type(),
8611 stubAddr, stubName, TypePtr::BOTTOM,
8612 result_start, ntta_start, nttb_start);
8613
8614 // return an int
8615 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8616 set_result(retvalue);
8617
8618 return true;
8619 }
8620
8621 //------------------------------inline_dilithiumMontMulByConstant
8622 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8623 address stubAddr;
8624 const char *stubName;
8625 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8626 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8627
8628 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8629 stubName = "dilithiumMontMulByConstant";
8630 if (!stubAddr) return false;
8631
8632 Node* coeffs = argument(0);
8633 Node* constant = argument(1);
8634
8635 coeffs = must_be_not_null(coeffs, true);
8636
8637 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8638 assert(coeffs_start, "coeffs is null");
8639 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8640 OptoRuntime::dilithiumMontMulByConstant_Type(),
8641 stubAddr, stubName, TypePtr::BOTTOM,
8642 coeffs_start, constant);
8643
8644 // return an int
8645 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8646 set_result(retvalue);
8647 return true;
8648 }
8649
8650
8651 //------------------------------inline_dilithiumDecomposePoly
8652 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8653 address stubAddr;
8654 const char *stubName;
8655 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8656 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8657
8658 stubAddr = StubRoutines::dilithiumDecomposePoly();
8659 stubName = "dilithiumDecomposePoly";
8660 if (!stubAddr) return false;
8661
8662 Node* input = argument(0);
8663 Node* lowPart = argument(1);
8664 Node* highPart = argument(2);
8665 Node* twoGamma2 = argument(3);
8666 Node* multiplier = argument(4);
8667
8668 input = must_be_not_null(input, true);
8669 lowPart = must_be_not_null(lowPart, true);
8670 highPart = must_be_not_null(highPart, true);
8671
8672 Node* input_start = array_element_address(input, intcon(0), T_INT);
8673 assert(input_start, "input is null");
8674 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8675 assert(lowPart_start, "lowPart is null");
8676 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8677 assert(highPart_start, "highPart is null");
8678
8679 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8680 OptoRuntime::dilithiumDecomposePoly_Type(),
8681 stubAddr, stubName, TypePtr::BOTTOM,
8682 input_start, lowPart_start, highPart_start,
8683 twoGamma2, multiplier);
8684
8685 // return an int
8686 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8687 set_result(retvalue);
8688 return true;
8689 }
8690
8691 bool LibraryCallKit::inline_base64_encodeBlock() {
8692 address stubAddr;
8693 const char *stubName;
8694 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8695 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8696 stubAddr = StubRoutines::base64_encodeBlock();
8697 stubName = "encodeBlock";
8698
8699 if (!stubAddr) return false;
8700 Node* base64obj = argument(0);
8701 Node* src = argument(1);
8702 Node* offset = argument(2);
8703 Node* len = argument(3);
8704 Node* dest = argument(4);
8705 Node* dp = argument(5);
8706 Node* isURL = argument(6);
8707
8708 src = must_be_not_null(src, true);
8709 dest = must_be_not_null(dest, true);
8710
8711 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8712 assert(src_start, "source array is null");
8713 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8714 assert(dest_start, "destination array is null");
8715
8716 Node* base64 = make_runtime_call(RC_LEAF,
8717 OptoRuntime::base64_encodeBlock_Type(),
8718 stubAddr, stubName, TypePtr::BOTTOM,
8719 src_start, offset, len, dest_start, dp, isURL);
8720 return true;
8721 }
8722
8723 bool LibraryCallKit::inline_base64_decodeBlock() {
8724 address stubAddr;
8725 const char *stubName;
8726 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8727 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8728 stubAddr = StubRoutines::base64_decodeBlock();
8729 stubName = "decodeBlock";
8730
8731 if (!stubAddr) return false;
8732 Node* base64obj = argument(0);
8733 Node* src = argument(1);
8734 Node* src_offset = argument(2);
8735 Node* len = argument(3);
8736 Node* dest = argument(4);
8737 Node* dest_offset = argument(5);
8738 Node* isURL = argument(6);
8739 Node* isMIME = argument(7);
8740
8741 src = must_be_not_null(src, true);
8742 dest = must_be_not_null(dest, true);
8743
8744 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8745 assert(src_start, "source array is null");
8746 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8747 assert(dest_start, "destination array is null");
8748
8749 Node* call = make_runtime_call(RC_LEAF,
8750 OptoRuntime::base64_decodeBlock_Type(),
8751 stubAddr, stubName, TypePtr::BOTTOM,
8752 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8753 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8754 set_result(result);
8755 return true;
8756 }
8757
8758 bool LibraryCallKit::inline_poly1305_processBlocks() {
8759 address stubAddr;
8760 const char *stubName;
8761 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8762 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8763 stubAddr = StubRoutines::poly1305_processBlocks();
8764 stubName = "poly1305_processBlocks";
8765
8766 if (!stubAddr) return false;
8767 null_check_receiver(); // null-check receiver
8768 if (stopped()) return true;
8769
8770 Node* input = argument(1);
8771 Node* input_offset = argument(2);
8772 Node* len = argument(3);
8773 Node* alimbs = argument(4);
8774 Node* rlimbs = argument(5);
8775
8776 input = must_be_not_null(input, true);
8777 alimbs = must_be_not_null(alimbs, true);
8778 rlimbs = must_be_not_null(rlimbs, true);
8779
8780 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8781 assert(input_start, "input array is null");
8782 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8783 assert(acc_start, "acc array is null");
8784 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8785 assert(r_start, "r array is null");
8786
8787 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8788 OptoRuntime::poly1305_processBlocks_Type(),
8789 stubAddr, stubName, TypePtr::BOTTOM,
8790 input_start, len, acc_start, r_start);
8791 return true;
8792 }
8793
8794 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8795 address stubAddr;
8796 const char *stubName;
8797 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8798 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8799 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8800 stubName = "intpoly_montgomeryMult_P256";
8801
8802 if (!stubAddr) return false;
8803 null_check_receiver(); // null-check receiver
8804 if (stopped()) return true;
8805
8806 Node* a = argument(1);
8807 Node* b = argument(2);
8808 Node* r = argument(3);
8809
8810 a = must_be_not_null(a, true);
8811 b = must_be_not_null(b, true);
8812 r = must_be_not_null(r, true);
8813
8814 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8815 assert(a_start, "a array is null");
8816 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8817 assert(b_start, "b array is null");
8818 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8819 assert(r_start, "r array is null");
8820
8821 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8822 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8823 stubAddr, stubName, TypePtr::BOTTOM,
8824 a_start, b_start, r_start);
8825 return true;
8826 }
8827
8828 bool LibraryCallKit::inline_intpoly_assign() {
8829 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8830 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8831 const char *stubName = "intpoly_assign";
8832 address stubAddr = StubRoutines::intpoly_assign();
8833 if (!stubAddr) return false;
8834
8835 Node* set = argument(0);
8836 Node* a = argument(1);
8837 Node* b = argument(2);
8838 Node* arr_length = load_array_length(a);
8839
8840 a = must_be_not_null(a, true);
8841 b = must_be_not_null(b, true);
8842
8843 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8844 assert(a_start, "a array is null");
8845 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8846 assert(b_start, "b array is null");
8847
8848 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8849 OptoRuntime::intpoly_assign_Type(),
8850 stubAddr, stubName, TypePtr::BOTTOM,
8851 set, a_start, b_start, arr_length);
8852 return true;
8853 }
8854
8855 //------------------------------inline_digestBase_implCompress-----------------------
8856 //
8857 // Calculate MD5 for single-block byte[] array.
8858 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8859 //
8860 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8861 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8862 //
8863 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8864 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8865 //
8866 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8867 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8868 //
8869 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8870 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8871 //
8872 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8873 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8874
8875 Node* digestBase_obj = argument(0);
8876 Node* src = argument(1); // type oop
8877 Node* ofs = argument(2); // type int
8878
8879 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8880 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8881 // failed array check
8882 return false;
8883 }
8884 // Figure out the size and type of the elements we will be copying.
8885 BasicType src_elem = src_type->elem()->array_element_basic_type();
8886 if (src_elem != T_BYTE) {
8887 return false;
8888 }
8889 // 'src_start' points to src array + offset
8890 src = must_be_not_null(src, true);
8891 Node* src_start = array_element_address(src, ofs, src_elem);
8892 Node* state = nullptr;
8893 Node* block_size = nullptr;
8894 address stubAddr;
8895 const char *stubName;
8896
8897 switch(id) {
8898 case vmIntrinsics::_md5_implCompress:
8899 assert(UseMD5Intrinsics, "need MD5 instruction support");
8900 state = get_state_from_digest_object(digestBase_obj, T_INT);
8901 stubAddr = StubRoutines::md5_implCompress();
8902 stubName = "md5_implCompress";
8903 break;
8904 case vmIntrinsics::_sha_implCompress:
8905 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8906 state = get_state_from_digest_object(digestBase_obj, T_INT);
8907 stubAddr = StubRoutines::sha1_implCompress();
8908 stubName = "sha1_implCompress";
8909 break;
8910 case vmIntrinsics::_sha2_implCompress:
8911 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8912 state = get_state_from_digest_object(digestBase_obj, T_INT);
8913 stubAddr = StubRoutines::sha256_implCompress();
8914 stubName = "sha256_implCompress";
8915 break;
8916 case vmIntrinsics::_sha5_implCompress:
8917 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8918 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8919 stubAddr = StubRoutines::sha512_implCompress();
8920 stubName = "sha512_implCompress";
8921 break;
8922 case vmIntrinsics::_sha3_implCompress:
8923 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8924 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8925 stubAddr = StubRoutines::sha3_implCompress();
8926 stubName = "sha3_implCompress";
8927 block_size = get_block_size_from_digest_object(digestBase_obj);
8928 if (block_size == nullptr) return false;
8929 break;
8930 default:
8931 fatal_unexpected_iid(id);
8932 return false;
8933 }
8934 if (state == nullptr) return false;
8935
8936 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8937 if (stubAddr == nullptr) return false;
8938
8939 // Call the stub.
8940 Node* call;
8941 if (block_size == nullptr) {
8942 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8943 stubAddr, stubName, TypePtr::BOTTOM,
8944 src_start, state);
8945 } else {
8946 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8947 stubAddr, stubName, TypePtr::BOTTOM,
8948 src_start, state, block_size);
8949 }
8950
8951 return true;
8952 }
8953
8954 //------------------------------inline_double_keccak
8955 bool LibraryCallKit::inline_double_keccak() {
8956 address stubAddr;
8957 const char *stubName;
8958 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8959 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8960
8961 stubAddr = StubRoutines::double_keccak();
8962 stubName = "double_keccak";
8963 if (!stubAddr) return false;
8964
8965 Node* status0 = argument(0);
8966 Node* status1 = argument(1);
8967
8968 status0 = must_be_not_null(status0, true);
8969 status1 = must_be_not_null(status1, true);
8970
8971 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8972 assert(status0_start, "status0 is null");
8973 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8974 assert(status1_start, "status1 is null");
8975 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8976 OptoRuntime::double_keccak_Type(),
8977 stubAddr, stubName, TypePtr::BOTTOM,
8978 status0_start, status1_start);
8979 // return an int
8980 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8981 set_result(retvalue);
8982 return true;
8983 }
8984
8985
8986 //------------------------------inline_digestBase_implCompressMB-----------------------
8987 //
8988 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8989 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8990 //
8991 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8992 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8993 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8994 assert((uint)predicate < 5, "sanity");
8995 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8996
8997 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8998 Node* src = argument(1); // byte[] array
8999 Node* ofs = argument(2); // type int
9000 Node* limit = argument(3); // type int
9001
9002 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9003 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9004 // failed array check
9005 return false;
9006 }
9007 // Figure out the size and type of the elements we will be copying.
9008 BasicType src_elem = src_type->elem()->array_element_basic_type();
9009 if (src_elem != T_BYTE) {
9010 return false;
9011 }
9012 // 'src_start' points to src array + offset
9013 src = must_be_not_null(src, false);
9014 Node* src_start = array_element_address(src, ofs, src_elem);
9015
9016 const char* klass_digestBase_name = nullptr;
9017 const char* stub_name = nullptr;
9018 address stub_addr = nullptr;
9019 BasicType elem_type = T_INT;
9020
9021 switch (predicate) {
9022 case 0:
9023 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9024 klass_digestBase_name = "sun/security/provider/MD5";
9025 stub_name = "md5_implCompressMB";
9026 stub_addr = StubRoutines::md5_implCompressMB();
9027 }
9028 break;
9029 case 1:
9030 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9031 klass_digestBase_name = "sun/security/provider/SHA";
9032 stub_name = "sha1_implCompressMB";
9033 stub_addr = StubRoutines::sha1_implCompressMB();
9034 }
9035 break;
9036 case 2:
9037 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9038 klass_digestBase_name = "sun/security/provider/SHA2";
9039 stub_name = "sha256_implCompressMB";
9040 stub_addr = StubRoutines::sha256_implCompressMB();
9041 }
9042 break;
9043 case 3:
9044 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9045 klass_digestBase_name = "sun/security/provider/SHA5";
9046 stub_name = "sha512_implCompressMB";
9047 stub_addr = StubRoutines::sha512_implCompressMB();
9048 elem_type = T_LONG;
9049 }
9050 break;
9051 case 4:
9052 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9053 klass_digestBase_name = "sun/security/provider/SHA3";
9054 stub_name = "sha3_implCompressMB";
9055 stub_addr = StubRoutines::sha3_implCompressMB();
9056 elem_type = T_LONG;
9057 }
9058 break;
9059 default:
9060 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9061 }
9062 if (klass_digestBase_name != nullptr) {
9063 assert(stub_addr != nullptr, "Stub is generated");
9064 if (stub_addr == nullptr) return false;
9065
9066 // get DigestBase klass to lookup for SHA klass
9067 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9068 assert(tinst != nullptr, "digestBase_obj is not instance???");
9069 assert(tinst->is_loaded(), "DigestBase is not loaded");
9070
9071 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9072 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9073 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9074 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9075 }
9076 return false;
9077 }
9078
9079 //------------------------------inline_digestBase_implCompressMB-----------------------
9080 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9081 BasicType elem_type, address stubAddr, const char *stubName,
9082 Node* src_start, Node* ofs, Node* limit) {
9083 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9084 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9085 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9086 digest_obj = _gvn.transform(digest_obj);
9087
9088 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9089 if (state == nullptr) return false;
9090
9091 Node* block_size = nullptr;
9092 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9093 block_size = get_block_size_from_digest_object(digest_obj);
9094 if (block_size == nullptr) return false;
9095 }
9096
9097 // Call the stub.
9098 Node* call;
9099 if (block_size == nullptr) {
9100 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9101 OptoRuntime::digestBase_implCompressMB_Type(false),
9102 stubAddr, stubName, TypePtr::BOTTOM,
9103 src_start, state, ofs, limit);
9104 } else {
9105 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9106 OptoRuntime::digestBase_implCompressMB_Type(true),
9107 stubAddr, stubName, TypePtr::BOTTOM,
9108 src_start, state, block_size, ofs, limit);
9109 }
9110
9111 // return ofs (int)
9112 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9113 set_result(result);
9114
9115 return true;
9116 }
9117
9118 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9119 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9120 assert(UseAES, "need AES instruction support");
9121 address stubAddr = nullptr;
9122 const char *stubName = nullptr;
9123 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9124 stubName = "galoisCounterMode_AESCrypt";
9125
9126 if (stubAddr == nullptr) return false;
9127
9128 Node* in = argument(0);
9129 Node* inOfs = argument(1);
9130 Node* len = argument(2);
9131 Node* ct = argument(3);
9132 Node* ctOfs = argument(4);
9133 Node* out = argument(5);
9134 Node* outOfs = argument(6);
9135 Node* gctr_object = argument(7);
9136 Node* ghash_object = argument(8);
9137
9138 // (1) in, ct and out are arrays.
9139 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9140 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9141 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9142 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9143 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9144 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9145
9146 // checks are the responsibility of the caller
9147 Node* in_start = in;
9148 Node* ct_start = ct;
9149 Node* out_start = out;
9150 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9151 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9152 in_start = array_element_address(in, inOfs, T_BYTE);
9153 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9154 out_start = array_element_address(out, outOfs, T_BYTE);
9155 }
9156
9157 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9158 // (because of the predicated logic executed earlier).
9159 // so we cast it here safely.
9160 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9161 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9162 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9163 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9164 Node* state = load_field_from_object(ghash_object, "state", "[J");
9165
9166 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9167 return false;
9168 }
9169 // cast it to what we know it will be at runtime
9170 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9171 assert(tinst != nullptr, "GCTR obj is null");
9172 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9173 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9174 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9175 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9176 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9177 const TypeOopPtr* xtype = aklass->as_instance_type();
9178 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9179 aescrypt_object = _gvn.transform(aescrypt_object);
9180 // we need to get the start of the aescrypt_object's expanded key array
9181 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
9182 if (k_start == nullptr) return false;
9183 // similarly, get the start address of the r vector
9184 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9185 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9186 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9187
9188
9189 // Call the stub, passing params
9190 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9191 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9192 stubAddr, stubName, TypePtr::BOTTOM,
9193 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9194
9195 // return cipher length (int)
9196 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9197 set_result(retvalue);
9198
9199 return true;
9200 }
9201
9202 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9203 // Return node representing slow path of predicate check.
9204 // the pseudo code we want to emulate with this predicate is:
9205 // for encryption:
9206 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9207 // for decryption:
9208 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9209 // note cipher==plain is more conservative than the original java code but that's OK
9210 //
9211
9212 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9213 // The receiver was checked for null already.
9214 Node* objGCTR = argument(7);
9215 // Load embeddedCipher field of GCTR object.
9216 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9217 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9218
9219 // get AESCrypt klass for instanceOf check
9220 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9221 // will have same classloader as CipherBlockChaining object
9222 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9223 assert(tinst != nullptr, "GCTR obj is null");
9224 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9225
9226 // we want to do an instanceof comparison against the AESCrypt class
9227 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9228 if (!klass_AESCrypt->is_loaded()) {
9229 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9230 Node* ctrl = control();
9231 set_control(top()); // no regular fast path
9232 return ctrl;
9233 }
9234
9235 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9236 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9237 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9238 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9239 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9240
9241 return instof_false; // even if it is null
9242 }
9243
9244 //------------------------------get_state_from_digest_object-----------------------
9245 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9246 const char* state_type;
9247 switch (elem_type) {
9248 case T_BYTE: state_type = "[B"; break;
9249 case T_INT: state_type = "[I"; break;
9250 case T_LONG: state_type = "[J"; break;
9251 default: ShouldNotReachHere();
9252 }
9253 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9254 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9255 if (digest_state == nullptr) return (Node *) nullptr;
9256
9257 // now have the array, need to get the start address of the state array
9258 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9259 return state;
9260 }
9261
9262 //------------------------------get_block_size_from_sha3_object----------------------------------
9263 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9264 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9265 assert (block_size != nullptr, "sanity");
9266 return block_size;
9267 }
9268
9269 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9270 // Return node representing slow path of predicate check.
9271 // the pseudo code we want to emulate with this predicate is:
9272 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9273 //
9274 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9275 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9276 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9277 assert((uint)predicate < 5, "sanity");
9278
9279 // The receiver was checked for null already.
9280 Node* digestBaseObj = argument(0);
9281
9282 // get DigestBase klass for instanceOf check
9283 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9284 assert(tinst != nullptr, "digestBaseObj is null");
9285 assert(tinst->is_loaded(), "DigestBase is not loaded");
9286
9287 const char* klass_name = nullptr;
9288 switch (predicate) {
9289 case 0:
9290 if (UseMD5Intrinsics) {
9291 // we want to do an instanceof comparison against the MD5 class
9292 klass_name = "sun/security/provider/MD5";
9293 }
9294 break;
9295 case 1:
9296 if (UseSHA1Intrinsics) {
9297 // we want to do an instanceof comparison against the SHA class
9298 klass_name = "sun/security/provider/SHA";
9299 }
9300 break;
9301 case 2:
9302 if (UseSHA256Intrinsics) {
9303 // we want to do an instanceof comparison against the SHA2 class
9304 klass_name = "sun/security/provider/SHA2";
9305 }
9306 break;
9307 case 3:
9308 if (UseSHA512Intrinsics) {
9309 // we want to do an instanceof comparison against the SHA5 class
9310 klass_name = "sun/security/provider/SHA5";
9311 }
9312 break;
9313 case 4:
9314 if (UseSHA3Intrinsics) {
9315 // we want to do an instanceof comparison against the SHA3 class
9316 klass_name = "sun/security/provider/SHA3";
9317 }
9318 break;
9319 default:
9320 fatal("unknown SHA intrinsic predicate: %d", predicate);
9321 }
9322
9323 ciKlass* klass = nullptr;
9324 if (klass_name != nullptr) {
9325 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9326 }
9327 if ((klass == nullptr) || !klass->is_loaded()) {
9328 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9329 Node* ctrl = control();
9330 set_control(top()); // no intrinsic path
9331 return ctrl;
9332 }
9333 ciInstanceKlass* instklass = klass->as_instance_klass();
9334
9335 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9336 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9337 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9338 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9339
9340 return instof_false; // even if it is null
9341 }
9342
9343 //-------------inline_fma-----------------------------------
9344 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9345 Node *a = nullptr;
9346 Node *b = nullptr;
9347 Node *c = nullptr;
9348 Node* result = nullptr;
9349 switch (id) {
9350 case vmIntrinsics::_fmaD:
9351 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9352 // no receiver since it is static method
9353 a = argument(0);
9354 b = argument(2);
9355 c = argument(4);
9356 result = _gvn.transform(new FmaDNode(a, b, c));
9357 break;
9358 case vmIntrinsics::_fmaF:
9359 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9360 a = argument(0);
9361 b = argument(1);
9362 c = argument(2);
9363 result = _gvn.transform(new FmaFNode(a, b, c));
9364 break;
9365 default:
9366 fatal_unexpected_iid(id); break;
9367 }
9368 set_result(result);
9369 return true;
9370 }
9371
9372 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9373 // argument(0) is receiver
9374 Node* codePoint = argument(1);
9375 Node* n = nullptr;
9376
9377 switch (id) {
9378 case vmIntrinsics::_isDigit :
9379 n = new DigitNode(control(), codePoint);
9380 break;
9381 case vmIntrinsics::_isLowerCase :
9382 n = new LowerCaseNode(control(), codePoint);
9383 break;
9384 case vmIntrinsics::_isUpperCase :
9385 n = new UpperCaseNode(control(), codePoint);
9386 break;
9387 case vmIntrinsics::_isWhitespace :
9388 n = new WhitespaceNode(control(), codePoint);
9389 break;
9390 default:
9391 fatal_unexpected_iid(id);
9392 }
9393
9394 set_result(_gvn.transform(n));
9395 return true;
9396 }
9397
9398 bool LibraryCallKit::inline_profileBoolean() {
9399 Node* counts = argument(1);
9400 const TypeAryPtr* ary = nullptr;
9401 ciArray* aobj = nullptr;
9402 if (counts->is_Con()
9403 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9404 && (aobj = ary->const_oop()->as_array()) != nullptr
9405 && (aobj->length() == 2)) {
9406 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9407 jint false_cnt = aobj->element_value(0).as_int();
9408 jint true_cnt = aobj->element_value(1).as_int();
9409
9410 if (C->log() != nullptr) {
9411 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9412 false_cnt, true_cnt);
9413 }
9414
9415 if (false_cnt + true_cnt == 0) {
9416 // According to profile, never executed.
9417 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9418 Deoptimization::Action_reinterpret);
9419 return true;
9420 }
9421
9422 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9423 // is a number of each value occurrences.
9424 Node* result = argument(0);
9425 if (false_cnt == 0 || true_cnt == 0) {
9426 // According to profile, one value has been never seen.
9427 int expected_val = (false_cnt == 0) ? 1 : 0;
9428
9429 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9430 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9431
9432 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9433 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9434 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9435
9436 { // Slow path: uncommon trap for never seen value and then reexecute
9437 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9438 // the value has been seen at least once.
9439 PreserveJVMState pjvms(this);
9440 PreserveReexecuteState preexecs(this);
9441 jvms()->set_should_reexecute(true);
9442
9443 set_control(slow_path);
9444 set_i_o(i_o());
9445
9446 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9447 Deoptimization::Action_reinterpret);
9448 }
9449 // The guard for never seen value enables sharpening of the result and
9450 // returning a constant. It allows to eliminate branches on the same value
9451 // later on.
9452 set_control(fast_path);
9453 result = intcon(expected_val);
9454 }
9455 // Stop profiling.
9456 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9457 // By replacing method body with profile data (represented as ProfileBooleanNode
9458 // on IR level) we effectively disable profiling.
9459 // It enables full speed execution once optimized code is generated.
9460 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9461 C->record_for_igvn(profile);
9462 set_result(profile);
9463 return true;
9464 } else {
9465 // Continue profiling.
9466 // Profile data isn't available at the moment. So, execute method's bytecode version.
9467 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9468 // is compiled and counters aren't available since corresponding MethodHandle
9469 // isn't a compile-time constant.
9470 return false;
9471 }
9472 }
9473
9474 bool LibraryCallKit::inline_isCompileConstant() {
9475 Node* n = argument(0);
9476 set_result(n->is_Con() ? intcon(1) : intcon(0));
9477 return true;
9478 }
9479
9480 //------------------------------- inline_getObjectSize --------------------------------------
9481 //
9482 // Calculate the runtime size of the object/array.
9483 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9484 //
9485 bool LibraryCallKit::inline_getObjectSize() {
9486 Node* obj = argument(3);
9487 Node* klass_node = load_object_klass(obj);
9488
9489 jint layout_con = Klass::_lh_neutral_value;
9490 Node* layout_val = get_layout_helper(klass_node, layout_con);
9491 int layout_is_con = (layout_val == nullptr);
9492
9493 if (layout_is_con) {
9494 // Layout helper is constant, can figure out things at compile time.
9495
9496 if (Klass::layout_helper_is_instance(layout_con)) {
9497 // Instance case: layout_con contains the size itself.
9498 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9499 set_result(size);
9500 } else {
9501 // Array case: size is round(header + element_size*arraylength).
9502 // Since arraylength is different for every array instance, we have to
9503 // compute the whole thing at runtime.
9504
9505 Node* arr_length = load_array_length(obj);
9506
9507 int round_mask = MinObjAlignmentInBytes - 1;
9508 int hsize = Klass::layout_helper_header_size(layout_con);
9509 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9510
9511 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9512 round_mask = 0; // strength-reduce it if it goes away completely
9513 }
9514 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9515 Node* header_size = intcon(hsize + round_mask);
9516
9517 Node* lengthx = ConvI2X(arr_length);
9518 Node* headerx = ConvI2X(header_size);
9519
9520 Node* abody = lengthx;
9521 if (eshift != 0) {
9522 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9523 }
9524 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9525 if (round_mask != 0) {
9526 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9527 }
9528 size = ConvX2L(size);
9529 set_result(size);
9530 }
9531 } else {
9532 // Layout helper is not constant, need to test for array-ness at runtime.
9533
9534 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9535 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9536 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9537 record_for_igvn(result_reg);
9538
9539 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9540 if (array_ctl != nullptr) {
9541 // Array case: size is round(header + element_size*arraylength).
9542 // Since arraylength is different for every array instance, we have to
9543 // compute the whole thing at runtime.
9544
9545 PreserveJVMState pjvms(this);
9546 set_control(array_ctl);
9547 Node* arr_length = load_array_length(obj);
9548
9549 int round_mask = MinObjAlignmentInBytes - 1;
9550 Node* mask = intcon(round_mask);
9551
9552 Node* hss = intcon(Klass::_lh_header_size_shift);
9553 Node* hsm = intcon(Klass::_lh_header_size_mask);
9554 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9555 header_size = _gvn.transform(new AndINode(header_size, hsm));
9556 header_size = _gvn.transform(new AddINode(header_size, mask));
9557
9558 // There is no need to mask or shift this value.
9559 // The semantics of LShiftINode include an implicit mask to 0x1F.
9560 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9561 Node* elem_shift = layout_val;
9562
9563 Node* lengthx = ConvI2X(arr_length);
9564 Node* headerx = ConvI2X(header_size);
9565
9566 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9567 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9568 if (round_mask != 0) {
9569 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9570 }
9571 size = ConvX2L(size);
9572
9573 result_reg->init_req(_array_path, control());
9574 result_val->init_req(_array_path, size);
9575 }
9576
9577 if (!stopped()) {
9578 // Instance case: the layout helper gives us instance size almost directly,
9579 // but we need to mask out the _lh_instance_slow_path_bit.
9580 Node* size = ConvI2X(layout_val);
9581 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9582 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9583 size = _gvn.transform(new AndXNode(size, mask));
9584 size = ConvX2L(size);
9585
9586 result_reg->init_req(_instance_path, control());
9587 result_val->init_req(_instance_path, size);
9588 }
9589
9590 set_result(result_reg, result_val);
9591 }
9592
9593 return true;
9594 }
9595
9596 //------------------------------- inline_blackhole --------------------------------------
9597 //
9598 // Make sure all arguments to this node are alive.
9599 // This matches methods that were requested to be blackholed through compile commands.
9600 //
9601 bool LibraryCallKit::inline_blackhole() {
9602 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9603 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9604 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9605
9606 // Blackhole node pinches only the control, not memory. This allows
9607 // the blackhole to be pinned in the loop that computes blackholed
9608 // values, but have no other side effects, like breaking the optimizations
9609 // across the blackhole.
9610
9611 Node* bh = _gvn.transform(new BlackholeNode(control()));
9612 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9613
9614 // Bind call arguments as blackhole arguments to keep them alive
9615 uint nargs = callee()->arg_size();
9616 for (uint i = 0; i < nargs; i++) {
9617 bh->add_req(argument(i));
9618 }
9619
9620 return true;
9621 }
9622
9623 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9624 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9625 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9626 return nullptr; // box klass is not Float16
9627 }
9628
9629 // Null check; get notnull casted pointer
9630 Node* null_ctl = top();
9631 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9632 // If not_null_box is dead, only null-path is taken
9633 if (stopped()) {
9634 set_control(null_ctl);
9635 return nullptr;
9636 }
9637 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9638 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9639 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9640 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9641 }
9642
9643 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9644 PreserveReexecuteState preexecs(this);
9645 jvms()->set_should_reexecute(true);
9646
9647 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9648 Node* klass_node = makecon(klass_type);
9649 Node* box = new_instance(klass_node);
9650
9651 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9652 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9653
9654 Node* field_store = _gvn.transform(access_store_at(box,
9655 value_field,
9656 value_adr_type,
9657 value,
9658 TypeInt::SHORT,
9659 T_SHORT,
9660 IN_HEAP));
9661 set_memory(field_store, value_adr_type);
9662 return box;
9663 }
9664
9665 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9666 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9667 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9668 return false;
9669 }
9670
9671 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9672 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9673 return false;
9674 }
9675
9676 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9677 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9678 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9679 ciSymbols::short_signature(),
9680 false);
9681 assert(field != nullptr, "");
9682
9683 // Transformed nodes
9684 Node* fld1 = nullptr;
9685 Node* fld2 = nullptr;
9686 Node* fld3 = nullptr;
9687 switch(num_args) {
9688 case 3:
9689 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9690 if (fld3 == nullptr) {
9691 return false;
9692 }
9693 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9694 // fall-through
9695 case 2:
9696 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9697 if (fld2 == nullptr) {
9698 return false;
9699 }
9700 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9701 // fall-through
9702 case 1:
9703 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9704 if (fld1 == nullptr) {
9705 return false;
9706 }
9707 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9708 break;
9709 default: fatal("Unsupported number of arguments %d", num_args);
9710 }
9711
9712 Node* result = nullptr;
9713 switch (id) {
9714 // Unary operations
9715 case vmIntrinsics::_sqrt_float16:
9716 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9717 break;
9718 // Ternary operations
9719 case vmIntrinsics::_fma_float16:
9720 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9721 break;
9722 default:
9723 fatal_unexpected_iid(id);
9724 break;
9725 }
9726 result = _gvn.transform(new ReinterpretHF2SNode(result));
9727 set_result(box_fp16_value(float16_box_type, field, result));
9728 return true;
9729 }
9730