1 /*
2 * Copyright (c) 1999, 2023, 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 "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dabs:
258 case vmIntrinsics::_fabs:
259 case vmIntrinsics::_iabs:
260 case vmIntrinsics::_labs:
261 case vmIntrinsics::_datan2:
262 case vmIntrinsics::_dsqrt:
263 case vmIntrinsics::_dsqrt_strict:
264 case vmIntrinsics::_dexp:
265 case vmIntrinsics::_dlog:
266 case vmIntrinsics::_dlog10:
267 case vmIntrinsics::_dpow:
268 case vmIntrinsics::_dcopySign:
269 case vmIntrinsics::_fcopySign:
270 case vmIntrinsics::_dsignum:
271 case vmIntrinsics::_roundF:
272 case vmIntrinsics::_roundD:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
297 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
298 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
299 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
300
301 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
308 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
309
310 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
311 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
312
313 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
314
315 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
316 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
317 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
318 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
319
320 case vmIntrinsics::_compressStringC:
321 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
322 case vmIntrinsics::_inflateStringC:
323 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
324
325 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
326 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
327 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
328 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
329 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
330 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
331 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
332 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
333 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
334
335 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
336 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
337 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
338 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
339 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
340 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
341 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
342 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
343 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
344
345 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
346 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
347 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
348 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
349 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
350 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
351 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
352 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
353 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
354
355 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
356 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
357 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
358 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
359 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
360 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
361 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
362 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
363 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
364
365 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
366 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
367 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
368 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
369
370 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
371 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
372 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
373 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
374
375 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
376 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
377 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
378 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
379 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
380 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
381 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
382 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
383 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
384
385 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
386 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
387 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
388 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
389 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
390 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
391 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
392 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
393 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
394
395 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
396 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
397 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
398 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
399 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
400 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
401 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
402 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
403 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
404
405 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
406 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
407 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
408 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
409 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
410 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
411 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
412 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
413 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
414
415 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
417 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
420
421 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
422 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
423 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
424 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
425 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
426 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
427 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
428 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
429 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
441
442 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
457
458 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
459 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
460 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
461 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
462
463 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
465 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
468
469 case vmIntrinsics::_loadFence:
470 case vmIntrinsics::_storeFence:
471 case vmIntrinsics::_storeStoreFence:
472 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
473
474 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
475
476 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
477 case vmIntrinsics::_currentThread: return inline_native_currentThread();
478 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
479
480 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
481 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
482
483 #if INCLUDE_JVMTI
484 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
485 "notifyJvmtiStart", true, false);
486 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
487 "notifyJvmtiEnd", false, true);
488 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
489 "notifyJvmtiMount", false, false);
490 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
491 "notifyJvmtiUnmount", false, false);
492 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_getLength: return inline_native_getLength();
508 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
509 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
510 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
511 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
512 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
513 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
514 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
515
516 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
517 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
518
519 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
520
521 case vmIntrinsics::_isInstance:
522 case vmIntrinsics::_getModifiers:
523 case vmIntrinsics::_isInterface:
524 case vmIntrinsics::_isArray:
525 case vmIntrinsics::_isPrimitive:
526 case vmIntrinsics::_isHidden:
527 case vmIntrinsics::_getSuperclass:
528 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
529
530 case vmIntrinsics::_floatToRawIntBits:
531 case vmIntrinsics::_floatToIntBits:
532 case vmIntrinsics::_intBitsToFloat:
533 case vmIntrinsics::_doubleToRawLongBits:
534 case vmIntrinsics::_doubleToLongBits:
535 case vmIntrinsics::_longBitsToDouble:
536 case vmIntrinsics::_floatToFloat16:
537 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
538
539 case vmIntrinsics::_floatIsFinite:
540 case vmIntrinsics::_floatIsInfinite:
541 case vmIntrinsics::_doubleIsFinite:
542 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
543
544 case vmIntrinsics::_numberOfLeadingZeros_i:
545 case vmIntrinsics::_numberOfLeadingZeros_l:
546 case vmIntrinsics::_numberOfTrailingZeros_i:
547 case vmIntrinsics::_numberOfTrailingZeros_l:
548 case vmIntrinsics::_bitCount_i:
549 case vmIntrinsics::_bitCount_l:
550 case vmIntrinsics::_reverse_i:
551 case vmIntrinsics::_reverse_l:
552 case vmIntrinsics::_reverseBytes_i:
553 case vmIntrinsics::_reverseBytes_l:
554 case vmIntrinsics::_reverseBytes_s:
555 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
556
557 case vmIntrinsics::_compress_i:
558 case vmIntrinsics::_compress_l:
559 case vmIntrinsics::_expand_i:
560 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
561
562 case vmIntrinsics::_compareUnsigned_i:
563 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
564
565 case vmIntrinsics::_divideUnsigned_i:
566 case vmIntrinsics::_divideUnsigned_l:
567 case vmIntrinsics::_remainderUnsigned_i:
568 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
569
570 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
571
572 case vmIntrinsics::_Reference_get: return inline_reference_get();
573 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
574 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
575
576 case vmIntrinsics::_Class_cast: return inline_Class_cast();
577
578 case vmIntrinsics::_aescrypt_encryptBlock:
579 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
580
581 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
582 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
583 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
586 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
587 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
588
589 case vmIntrinsics::_counterMode_AESCrypt:
590 return inline_counterMode_AESCrypt(intrinsic_id());
591
592 case vmIntrinsics::_galoisCounterMode_AESCrypt:
593 return inline_galoisCounterMode_AESCrypt();
594
595 case vmIntrinsics::_md5_implCompress:
596 case vmIntrinsics::_sha_implCompress:
597 case vmIntrinsics::_sha2_implCompress:
598 case vmIntrinsics::_sha5_implCompress:
599 case vmIntrinsics::_sha3_implCompress:
600 return inline_digestBase_implCompress(intrinsic_id());
601
602 case vmIntrinsics::_digestBase_implCompressMB:
603 return inline_digestBase_implCompressMB(predicate);
604
605 case vmIntrinsics::_multiplyToLen:
606 return inline_multiplyToLen();
607
608 case vmIntrinsics::_squareToLen:
609 return inline_squareToLen();
610
611 case vmIntrinsics::_mulAdd:
612 return inline_mulAdd();
613
614 case vmIntrinsics::_montgomeryMultiply:
615 return inline_montgomeryMultiply();
616 case vmIntrinsics::_montgomerySquare:
617 return inline_montgomerySquare();
618
619 case vmIntrinsics::_bigIntegerRightShiftWorker:
620 return inline_bigIntegerShift(true);
621 case vmIntrinsics::_bigIntegerLeftShiftWorker:
622 return inline_bigIntegerShift(false);
623
624 case vmIntrinsics::_vectorizedMismatch:
625 return inline_vectorizedMismatch();
626
627 case vmIntrinsics::_ghash_processBlocks:
628 return inline_ghash_processBlocks();
629 case vmIntrinsics::_chacha20Block:
630 return inline_chacha20Block();
631 case vmIntrinsics::_base64_encodeBlock:
632 return inline_base64_encodeBlock();
633 case vmIntrinsics::_base64_decodeBlock:
634 return inline_base64_decodeBlock();
635 case vmIntrinsics::_poly1305_processBlocks:
636 return inline_poly1305_processBlocks();
637
638 case vmIntrinsics::_encodeISOArray:
639 case vmIntrinsics::_encodeByteISOArray:
640 return inline_encodeISOArray(false);
641 case vmIntrinsics::_encodeAsciiArray:
642 return inline_encodeISOArray(true);
643
644 case vmIntrinsics::_updateCRC32:
645 return inline_updateCRC32();
646 case vmIntrinsics::_updateBytesCRC32:
647 return inline_updateBytesCRC32();
648 case vmIntrinsics::_updateByteBufferCRC32:
649 return inline_updateByteBufferCRC32();
650
651 case vmIntrinsics::_updateBytesCRC32C:
652 return inline_updateBytesCRC32C();
653 case vmIntrinsics::_updateDirectByteBufferCRC32C:
654 return inline_updateDirectByteBufferCRC32C();
655
656 case vmIntrinsics::_updateBytesAdler32:
657 return inline_updateBytesAdler32();
658 case vmIntrinsics::_updateByteBufferAdler32:
659 return inline_updateByteBufferAdler32();
660
661 case vmIntrinsics::_profileBoolean:
662 return inline_profileBoolean();
663 case vmIntrinsics::_isCompileConstant:
664 return inline_isCompileConstant();
665
666 case vmIntrinsics::_countPositives:
667 return inline_countPositives();
668
669 case vmIntrinsics::_fmaD:
670 case vmIntrinsics::_fmaF:
671 return inline_fma(intrinsic_id());
672
673 case vmIntrinsics::_isDigit:
674 case vmIntrinsics::_isLowerCase:
675 case vmIntrinsics::_isUpperCase:
676 case vmIntrinsics::_isWhitespace:
677 return inline_character_compare(intrinsic_id());
678
679 case vmIntrinsics::_min:
680 case vmIntrinsics::_max:
681 case vmIntrinsics::_min_strict:
682 case vmIntrinsics::_max_strict:
683 return inline_min_max(intrinsic_id());
684
685 case vmIntrinsics::_maxF:
686 case vmIntrinsics::_minF:
687 case vmIntrinsics::_maxD:
688 case vmIntrinsics::_minD:
689 case vmIntrinsics::_maxF_strict:
690 case vmIntrinsics::_minF_strict:
691 case vmIntrinsics::_maxD_strict:
692 case vmIntrinsics::_minD_strict:
693 return inline_fp_min_max(intrinsic_id());
694
695 case vmIntrinsics::_VectorUnaryOp:
696 return inline_vector_nary_operation(1);
697 case vmIntrinsics::_VectorBinaryOp:
698 return inline_vector_nary_operation(2);
699 case vmIntrinsics::_VectorTernaryOp:
700 return inline_vector_nary_operation(3);
701 case vmIntrinsics::_VectorFromBitsCoerced:
702 return inline_vector_frombits_coerced();
703 case vmIntrinsics::_VectorShuffleIota:
704 return inline_vector_shuffle_iota();
705 case vmIntrinsics::_VectorMaskOp:
706 return inline_vector_mask_operation();
707 case vmIntrinsics::_VectorShuffleToVector:
708 return inline_vector_shuffle_to_vector();
709 case vmIntrinsics::_VectorLoadOp:
710 return inline_vector_mem_operation(/*is_store=*/false);
711 case vmIntrinsics::_VectorLoadMaskedOp:
712 return inline_vector_mem_masked_operation(/*is_store*/false);
713 case vmIntrinsics::_VectorStoreOp:
714 return inline_vector_mem_operation(/*is_store=*/true);
715 case vmIntrinsics::_VectorStoreMaskedOp:
716 return inline_vector_mem_masked_operation(/*is_store=*/true);
717 case vmIntrinsics::_VectorGatherOp:
718 return inline_vector_gather_scatter(/*is_scatter*/ false);
719 case vmIntrinsics::_VectorScatterOp:
720 return inline_vector_gather_scatter(/*is_scatter*/ true);
721 case vmIntrinsics::_VectorReductionCoerced:
722 return inline_vector_reduction();
723 case vmIntrinsics::_VectorTest:
724 return inline_vector_test();
725 case vmIntrinsics::_VectorBlend:
726 return inline_vector_blend();
727 case vmIntrinsics::_VectorRearrange:
728 return inline_vector_rearrange();
729 case vmIntrinsics::_VectorCompare:
730 return inline_vector_compare();
731 case vmIntrinsics::_VectorBroadcastInt:
732 return inline_vector_broadcast_int();
733 case vmIntrinsics::_VectorConvert:
734 return inline_vector_convert();
735 case vmIntrinsics::_VectorInsert:
736 return inline_vector_insert();
737 case vmIntrinsics::_VectorExtract:
738 return inline_vector_extract();
739 case vmIntrinsics::_VectorCompressExpand:
740 return inline_vector_compress_expand();
741 case vmIntrinsics::_IndexVector:
742 return inline_index_vector();
743 case vmIntrinsics::_IndexPartiallyInUpperRange:
744 return inline_index_partially_in_upper_range();
745
746 case vmIntrinsics::_getObjectSize:
747 return inline_getObjectSize();
748
749 case vmIntrinsics::_blackhole:
750 return inline_blackhole();
751
752 case vmIntrinsics::_shipilev_magic_timestamp:
753 return inline_timestamp(false);
754 case vmIntrinsics::_shipilev_magic_timestamp_serial:
755 return inline_timestamp(true);
756 case vmIntrinsics::_shipilev_magic_sizeOf:
757 return inline_sizeOf();
758 case vmIntrinsics::_shipilev_magic_addressOf:
759 return inline_addressOf();
760
761 default:
762 // If you get here, it may be that someone has added a new intrinsic
763 // to the list in vmIntrinsics.hpp without implementing it here.
764 #ifndef PRODUCT
765 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
766 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
767 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
768 }
769 #endif
770 return false;
771 }
772 }
773
774 Node* LibraryCallKit::try_to_predicate(int predicate) {
775 if (!jvms()->has_method()) {
776 // Root JVMState has a null method.
777 assert(map()->memory()->Opcode() == Op_Parm, "");
778 // Insert the memory aliasing node
779 set_all_memory(reset_memory());
780 }
781 assert(merged_memory(), "");
782
783 switch (intrinsic_id()) {
784 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
785 return inline_cipherBlockChaining_AESCrypt_predicate(false);
786 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
787 return inline_cipherBlockChaining_AESCrypt_predicate(true);
788 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
789 return inline_electronicCodeBook_AESCrypt_predicate(false);
790 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
791 return inline_electronicCodeBook_AESCrypt_predicate(true);
792 case vmIntrinsics::_counterMode_AESCrypt:
793 return inline_counterMode_AESCrypt_predicate();
794 case vmIntrinsics::_digestBase_implCompressMB:
795 return inline_digestBase_implCompressMB_predicate(predicate);
796 case vmIntrinsics::_galoisCounterMode_AESCrypt:
797 return inline_galoisCounterMode_AESCrypt_predicate();
798
799 default:
800 // If you get here, it may be that someone has added a new intrinsic
801 // to the list in vmIntrinsics.hpp without implementing it here.
802 #ifndef PRODUCT
803 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
804 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
805 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
806 }
807 #endif
808 Node* slow_ctl = control();
809 set_control(top()); // No fast path intrinsic
810 return slow_ctl;
811 }
812 }
813
814 //------------------------------set_result-------------------------------
815 // Helper function for finishing intrinsics.
816 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
817 record_for_igvn(region);
818 set_control(_gvn.transform(region));
819 set_result( _gvn.transform(value));
820 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
821 }
822
823 //------------------------------generate_guard---------------------------
824 // Helper function for generating guarded fast-slow graph structures.
825 // The given 'test', if true, guards a slow path. If the test fails
826 // then a fast path can be taken. (We generally hope it fails.)
827 // In all cases, GraphKit::control() is updated to the fast path.
828 // The returned value represents the control for the slow path.
829 // The return value is never 'top'; it is either a valid control
830 // or null if it is obvious that the slow path can never be taken.
831 // Also, if region and the slow control are not null, the slow edge
832 // is appended to the region.
833 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
834 if (stopped()) {
835 // Already short circuited.
836 return nullptr;
837 }
838
839 // Build an if node and its projections.
840 // If test is true we take the slow path, which we assume is uncommon.
841 if (_gvn.type(test) == TypeInt::ZERO) {
842 // The slow branch is never taken. No need to build this guard.
843 return nullptr;
844 }
845
846 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
847
848 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
849 if (if_slow == top()) {
850 // The slow branch is never taken. No need to build this guard.
851 return nullptr;
852 }
853
854 if (region != nullptr)
855 region->add_req(if_slow);
856
857 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
858 set_control(if_fast);
859
860 return if_slow;
861 }
862
863 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
864 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
865 }
866 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
867 return generate_guard(test, region, PROB_FAIR);
868 }
869
870 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
871 Node* *pos_index) {
872 if (stopped())
873 return nullptr; // already stopped
874 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
875 return nullptr; // index is already adequately typed
876 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
877 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
878 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
879 if (is_neg != nullptr && pos_index != nullptr) {
880 // Emulate effect of Parse::adjust_map_after_if.
881 Node* ccast = new CastIINode(index, TypeInt::POS);
882 ccast->set_req(0, control());
883 (*pos_index) = _gvn.transform(ccast);
884 }
885 return is_neg;
886 }
887
888 // Make sure that 'position' is a valid limit index, in [0..length].
889 // There are two equivalent plans for checking this:
890 // A. (offset + copyLength) unsigned<= arrayLength
891 // B. offset <= (arrayLength - copyLength)
892 // We require that all of the values above, except for the sum and
893 // difference, are already known to be non-negative.
894 // Plan A is robust in the face of overflow, if offset and copyLength
895 // are both hugely positive.
896 //
897 // Plan B is less direct and intuitive, but it does not overflow at
898 // all, since the difference of two non-negatives is always
899 // representable. Whenever Java methods must perform the equivalent
900 // check they generally use Plan B instead of Plan A.
901 // For the moment we use Plan A.
902 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
903 Node* subseq_length,
904 Node* array_length,
905 RegionNode* region) {
906 if (stopped())
907 return nullptr; // already stopped
908 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
909 if (zero_offset && subseq_length->eqv_uncast(array_length))
910 return nullptr; // common case of whole-array copy
911 Node* last = subseq_length;
912 if (!zero_offset) // last += offset
913 last = _gvn.transform(new AddINode(last, offset));
914 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
915 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
916 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
917 return is_over;
918 }
919
920 // Emit range checks for the given String.value byte array
921 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
922 if (stopped()) {
923 return; // already stopped
924 }
925 RegionNode* bailout = new RegionNode(1);
926 record_for_igvn(bailout);
927 if (char_count) {
928 // Convert char count to byte count
929 count = _gvn.transform(new LShiftINode(count, intcon(1)));
930 }
931
932 // Offset and count must not be negative
933 generate_negative_guard(offset, bailout);
934 generate_negative_guard(count, bailout);
935 // Offset + count must not exceed length of array
936 generate_limit_guard(offset, count, load_array_length(array), bailout);
937
938 if (bailout->req() > 1) {
939 PreserveJVMState pjvms(this);
940 set_control(_gvn.transform(bailout));
941 uncommon_trap(Deoptimization::Reason_intrinsic,
942 Deoptimization::Action_maybe_recompile);
943 }
944 }
945
946 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
947 bool is_immutable) {
948 ciKlass* thread_klass = env()->Thread_klass();
949 const Type* thread_type
950 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
951
952 Node* thread = _gvn.transform(new ThreadLocalNode());
953 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
954 tls_output = thread;
955
956 Node* thread_obj_handle
957 = (is_immutable
958 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
959 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
960 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
961 thread_obj_handle = _gvn.transform(thread_obj_handle);
962
963 DecoratorSet decorators = IN_NATIVE;
964 if (is_immutable) {
965 decorators |= C2_IMMUTABLE_MEMORY;
966 }
967 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
968 }
969
970 //--------------------------generate_current_thread--------------------
971 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
972 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
973 /*is_immutable*/false);
974 }
975
976 //--------------------------generate_virtual_thread--------------------
977 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
978 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
979 !C->method()->changes_current_thread());
980 }
981
982 //------------------------------make_string_method_node------------------------
983 // Helper method for String intrinsic functions. This version is called with
984 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
985 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
986 // containing the lengths of str1 and str2.
987 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
988 Node* result = nullptr;
989 switch (opcode) {
990 case Op_StrIndexOf:
991 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
992 str1_start, cnt1, str2_start, cnt2, ae);
993 break;
994 case Op_StrComp:
995 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
996 str1_start, cnt1, str2_start, cnt2, ae);
997 break;
998 case Op_StrEquals:
999 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1000 // Use the constant length if there is one because optimized match rule may exist.
1001 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1002 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1003 break;
1004 default:
1005 ShouldNotReachHere();
1006 return nullptr;
1007 }
1008
1009 // All these intrinsics have checks.
1010 C->set_has_split_ifs(true); // Has chance for split-if optimization
1011 clear_upper_avx();
1012
1013 return _gvn.transform(result);
1014 }
1015
1016 //------------------------------inline_string_compareTo------------------------
1017 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1018 Node* arg1 = argument(0);
1019 Node* arg2 = argument(1);
1020
1021 arg1 = must_be_not_null(arg1, true);
1022 arg2 = must_be_not_null(arg2, true);
1023
1024 // Get start addr and length of first argument
1025 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1026 Node* arg1_cnt = load_array_length(arg1);
1027
1028 // Get start addr and length of second argument
1029 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1030 Node* arg2_cnt = load_array_length(arg2);
1031
1032 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1033 set_result(result);
1034 return true;
1035 }
1036
1037 //------------------------------inline_string_equals------------------------
1038 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1039 Node* arg1 = argument(0);
1040 Node* arg2 = argument(1);
1041
1042 // paths (plus control) merge
1043 RegionNode* region = new RegionNode(3);
1044 Node* phi = new PhiNode(region, TypeInt::BOOL);
1045
1046 if (!stopped()) {
1047
1048 arg1 = must_be_not_null(arg1, true);
1049 arg2 = must_be_not_null(arg2, true);
1050
1051 // Get start addr and length of first argument
1052 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1053 Node* arg1_cnt = load_array_length(arg1);
1054
1055 // Get start addr and length of second argument
1056 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1057 Node* arg2_cnt = load_array_length(arg2);
1058
1059 // Check for arg1_cnt != arg2_cnt
1060 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1061 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1062 Node* if_ne = generate_slow_guard(bol, nullptr);
1063 if (if_ne != nullptr) {
1064 phi->init_req(2, intcon(0));
1065 region->init_req(2, if_ne);
1066 }
1067
1068 // Check for count == 0 is done by assembler code for StrEquals.
1069
1070 if (!stopped()) {
1071 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1072 phi->init_req(1, equals);
1073 region->init_req(1, control());
1074 }
1075 }
1076
1077 // post merge
1078 set_control(_gvn.transform(region));
1079 record_for_igvn(region);
1080
1081 set_result(_gvn.transform(phi));
1082 return true;
1083 }
1084
1085 //------------------------------inline_array_equals----------------------------
1086 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1087 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1088 Node* arg1 = argument(0);
1089 Node* arg2 = argument(1);
1090
1091 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1092 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1093 clear_upper_avx();
1094
1095 return true;
1096 }
1097
1098
1099 //------------------------------inline_countPositives------------------------------
1100 bool LibraryCallKit::inline_countPositives() {
1101 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1102 return false;
1103 }
1104
1105 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1106 // no receiver since it is static method
1107 Node* ba = argument(0);
1108 Node* offset = argument(1);
1109 Node* len = argument(2);
1110
1111 ba = must_be_not_null(ba, true);
1112
1113 // Range checks
1114 generate_string_range_check(ba, offset, len, false);
1115 if (stopped()) {
1116 return true;
1117 }
1118 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1119 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1120 set_result(_gvn.transform(result));
1121 return true;
1122 }
1123
1124 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1125 Node* index = argument(0);
1126 Node* length = bt == T_INT ? argument(1) : argument(2);
1127 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1128 return false;
1129 }
1130
1131 // check that length is positive
1132 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1133 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1134
1135 {
1136 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1137 uncommon_trap(Deoptimization::Reason_intrinsic,
1138 Deoptimization::Action_make_not_entrant);
1139 }
1140
1141 if (stopped()) {
1142 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1143 return true;
1144 }
1145
1146 // length is now known positive, add a cast node to make this explicit
1147 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1148 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1149 casted_length = _gvn.transform(casted_length);
1150 replace_in_map(length, casted_length);
1151 length = casted_length;
1152
1153 // Use an unsigned comparison for the range check itself
1154 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1155 BoolTest::mask btest = BoolTest::lt;
1156 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1157 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1158 _gvn.set_type(rc, rc->Value(&_gvn));
1159 if (!rc_bool->is_Con()) {
1160 record_for_igvn(rc);
1161 }
1162 set_control(_gvn.transform(new IfTrueNode(rc)));
1163 {
1164 PreserveJVMState pjvms(this);
1165 set_control(_gvn.transform(new IfFalseNode(rc)));
1166 uncommon_trap(Deoptimization::Reason_range_check,
1167 Deoptimization::Action_make_not_entrant);
1168 }
1169
1170 if (stopped()) {
1171 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1172 return true;
1173 }
1174
1175 // index is now known to be >= 0 and < length, cast it
1176 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1177 result = _gvn.transform(result);
1178 set_result(result);
1179 replace_in_map(index, result);
1180 return true;
1181 }
1182
1183 //------------------------------inline_string_indexOf------------------------
1184 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1185 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1186 return false;
1187 }
1188 Node* src = argument(0);
1189 Node* tgt = argument(1);
1190
1191 // Make the merge point
1192 RegionNode* result_rgn = new RegionNode(4);
1193 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1194
1195 src = must_be_not_null(src, true);
1196 tgt = must_be_not_null(tgt, true);
1197
1198 // Get start addr and length of source string
1199 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1200 Node* src_count = load_array_length(src);
1201
1202 // Get start addr and length of substring
1203 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1204 Node* tgt_count = load_array_length(tgt);
1205
1206 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1207 // Divide src size by 2 if String is UTF16 encoded
1208 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1209 }
1210 if (ae == StrIntrinsicNode::UU) {
1211 // Divide substring size by 2 if String is UTF16 encoded
1212 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1213 }
1214
1215 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1216 if (result != nullptr) {
1217 result_phi->init_req(3, result);
1218 result_rgn->init_req(3, control());
1219 }
1220 set_control(_gvn.transform(result_rgn));
1221 record_for_igvn(result_rgn);
1222 set_result(_gvn.transform(result_phi));
1223
1224 return true;
1225 }
1226
1227 //-----------------------------inline_string_indexOf-----------------------
1228 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1229 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1230 return false;
1231 }
1232 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1233 return false;
1234 }
1235 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1236 Node* src = argument(0); // byte[]
1237 Node* src_count = argument(1); // char count
1238 Node* tgt = argument(2); // byte[]
1239 Node* tgt_count = argument(3); // char count
1240 Node* from_index = argument(4); // char index
1241
1242 src = must_be_not_null(src, true);
1243 tgt = must_be_not_null(tgt, true);
1244
1245 // Multiply byte array index by 2 if String is UTF16 encoded
1246 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1247 src_count = _gvn.transform(new SubINode(src_count, from_index));
1248 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1249 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1250
1251 // Range checks
1252 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1253 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1254 if (stopped()) {
1255 return true;
1256 }
1257
1258 RegionNode* region = new RegionNode(5);
1259 Node* phi = new PhiNode(region, TypeInt::INT);
1260
1261 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1262 if (result != nullptr) {
1263 // The result is index relative to from_index if substring was found, -1 otherwise.
1264 // Generate code which will fold into cmove.
1265 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1266 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1267
1268 Node* if_lt = generate_slow_guard(bol, nullptr);
1269 if (if_lt != nullptr) {
1270 // result == -1
1271 phi->init_req(3, result);
1272 region->init_req(3, if_lt);
1273 }
1274 if (!stopped()) {
1275 result = _gvn.transform(new AddINode(result, from_index));
1276 phi->init_req(4, result);
1277 region->init_req(4, control());
1278 }
1279 }
1280
1281 set_control(_gvn.transform(region));
1282 record_for_igvn(region);
1283 set_result(_gvn.transform(phi));
1284 clear_upper_avx();
1285
1286 return true;
1287 }
1288
1289 // Create StrIndexOfNode with fast path checks
1290 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1291 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1292 // Check for substr count > string count
1293 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1294 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1295 Node* if_gt = generate_slow_guard(bol, nullptr);
1296 if (if_gt != nullptr) {
1297 phi->init_req(1, intcon(-1));
1298 region->init_req(1, if_gt);
1299 }
1300 if (!stopped()) {
1301 // Check for substr count == 0
1302 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1303 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1304 Node* if_zero = generate_slow_guard(bol, nullptr);
1305 if (if_zero != nullptr) {
1306 phi->init_req(2, intcon(0));
1307 region->init_req(2, if_zero);
1308 }
1309 }
1310 if (!stopped()) {
1311 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1312 }
1313 return nullptr;
1314 }
1315
1316 //-----------------------------inline_string_indexOfChar-----------------------
1317 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1318 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1319 return false;
1320 }
1321 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1322 return false;
1323 }
1324 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1325 Node* src = argument(0); // byte[]
1326 Node* int_ch = argument(1);
1327 Node* from_index = argument(2);
1328 Node* max = argument(3);
1329
1330 src = must_be_not_null(src, true);
1331
1332 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1333 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1334 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1335
1336 // Range checks
1337 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1338
1339 // Check for int_ch >= 0
1340 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1341 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1342 {
1343 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1344 uncommon_trap(Deoptimization::Reason_intrinsic,
1345 Deoptimization::Action_maybe_recompile);
1346 }
1347 if (stopped()) {
1348 return true;
1349 }
1350
1351 RegionNode* region = new RegionNode(3);
1352 Node* phi = new PhiNode(region, TypeInt::INT);
1353
1354 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1355 C->set_has_split_ifs(true); // Has chance for split-if optimization
1356 _gvn.transform(result);
1357
1358 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1359 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1360
1361 Node* if_lt = generate_slow_guard(bol, nullptr);
1362 if (if_lt != nullptr) {
1363 // result == -1
1364 phi->init_req(2, result);
1365 region->init_req(2, if_lt);
1366 }
1367 if (!stopped()) {
1368 result = _gvn.transform(new AddINode(result, from_index));
1369 phi->init_req(1, result);
1370 region->init_req(1, control());
1371 }
1372 set_control(_gvn.transform(region));
1373 record_for_igvn(region);
1374 set_result(_gvn.transform(phi));
1375
1376 return true;
1377 }
1378 //---------------------------inline_string_copy---------------------
1379 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1380 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1381 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1382 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1383 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1384 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1385 bool LibraryCallKit::inline_string_copy(bool compress) {
1386 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1387 return false;
1388 }
1389 int nargs = 5; // 2 oops, 3 ints
1390 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1391
1392 Node* src = argument(0);
1393 Node* src_offset = argument(1);
1394 Node* dst = argument(2);
1395 Node* dst_offset = argument(3);
1396 Node* length = argument(4);
1397
1398 // Check for allocation before we add nodes that would confuse
1399 // tightly_coupled_allocation()
1400 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1401
1402 // Figure out the size and type of the elements we will be copying.
1403 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1404 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1405 if (src_type == nullptr || dst_type == nullptr) {
1406 return false;
1407 }
1408 BasicType src_elem = src_type->elem()->array_element_basic_type();
1409 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1410 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1411 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1412 "Unsupported array types for inline_string_copy");
1413
1414 src = must_be_not_null(src, true);
1415 dst = must_be_not_null(dst, true);
1416
1417 // Convert char[] offsets to byte[] offsets
1418 bool convert_src = (compress && src_elem == T_BYTE);
1419 bool convert_dst = (!compress && dst_elem == T_BYTE);
1420 if (convert_src) {
1421 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1422 } else if (convert_dst) {
1423 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1424 }
1425
1426 // Range checks
1427 generate_string_range_check(src, src_offset, length, convert_src);
1428 generate_string_range_check(dst, dst_offset, length, convert_dst);
1429 if (stopped()) {
1430 return true;
1431 }
1432
1433 Node* src_start = array_element_address(src, src_offset, src_elem);
1434 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1435 // 'src_start' points to src array + scaled offset
1436 // 'dst_start' points to dst array + scaled offset
1437 Node* count = nullptr;
1438 if (compress) {
1439 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1440 } else {
1441 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1442 }
1443
1444 if (alloc != nullptr) {
1445 if (alloc->maybe_set_complete(&_gvn)) {
1446 // "You break it, you buy it."
1447 InitializeNode* init = alloc->initialization();
1448 assert(init->is_complete(), "we just did this");
1449 init->set_complete_with_arraycopy();
1450 assert(dst->is_CheckCastPP(), "sanity");
1451 assert(dst->in(0)->in(0) == init, "dest pinned");
1452 }
1453 // Do not let stores that initialize this object be reordered with
1454 // a subsequent store that would make this object accessible by
1455 // other threads.
1456 // Record what AllocateNode this StoreStore protects so that
1457 // escape analysis can go from the MemBarStoreStoreNode to the
1458 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1459 // based on the escape status of the AllocateNode.
1460 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1461 }
1462 if (compress) {
1463 set_result(_gvn.transform(count));
1464 }
1465 clear_upper_avx();
1466
1467 return true;
1468 }
1469
1470 #ifdef _LP64
1471 #define XTOP ,top() /*additional argument*/
1472 #else //_LP64
1473 #define XTOP /*no additional argument*/
1474 #endif //_LP64
1475
1476 //------------------------inline_string_toBytesU--------------------------
1477 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1478 bool LibraryCallKit::inline_string_toBytesU() {
1479 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1480 return false;
1481 }
1482 // Get the arguments.
1483 Node* value = argument(0);
1484 Node* offset = argument(1);
1485 Node* length = argument(2);
1486
1487 Node* newcopy = nullptr;
1488
1489 // Set the original stack and the reexecute bit for the interpreter to reexecute
1490 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1491 { PreserveReexecuteState preexecs(this);
1492 jvms()->set_should_reexecute(true);
1493
1494 // Check if a null path was taken unconditionally.
1495 value = null_check(value);
1496
1497 RegionNode* bailout = new RegionNode(1);
1498 record_for_igvn(bailout);
1499
1500 // Range checks
1501 generate_negative_guard(offset, bailout);
1502 generate_negative_guard(length, bailout);
1503 generate_limit_guard(offset, length, load_array_length(value), bailout);
1504 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1505 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1506
1507 if (bailout->req() > 1) {
1508 PreserveJVMState pjvms(this);
1509 set_control(_gvn.transform(bailout));
1510 uncommon_trap(Deoptimization::Reason_intrinsic,
1511 Deoptimization::Action_maybe_recompile);
1512 }
1513 if (stopped()) {
1514 return true;
1515 }
1516
1517 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1518 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1519 newcopy = new_array(klass_node, size, 0); // no arguments to push
1520 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1521 guarantee(alloc != nullptr, "created above");
1522
1523 // Calculate starting addresses.
1524 Node* src_start = array_element_address(value, offset, T_CHAR);
1525 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1526
1527 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1528 const TypeInt* toffset = gvn().type(offset)->is_int();
1529 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1530
1531 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1532 const char* copyfunc_name = "arraycopy";
1533 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1534 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1535 OptoRuntime::fast_arraycopy_Type(),
1536 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1537 src_start, dst_start, ConvI2X(length) XTOP);
1538 // Do not let reads from the cloned object float above the arraycopy.
1539 if (alloc->maybe_set_complete(&_gvn)) {
1540 // "You break it, you buy it."
1541 InitializeNode* init = alloc->initialization();
1542 assert(init->is_complete(), "we just did this");
1543 init->set_complete_with_arraycopy();
1544 assert(newcopy->is_CheckCastPP(), "sanity");
1545 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1546 }
1547 // Do not let stores that initialize this object be reordered with
1548 // a subsequent store that would make this object accessible by
1549 // other threads.
1550 // Record what AllocateNode this StoreStore protects so that
1551 // escape analysis can go from the MemBarStoreStoreNode to the
1552 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1553 // based on the escape status of the AllocateNode.
1554 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1555 } // original reexecute is set back here
1556
1557 C->set_has_split_ifs(true); // Has chance for split-if optimization
1558 if (!stopped()) {
1559 set_result(newcopy);
1560 }
1561 clear_upper_avx();
1562
1563 return true;
1564 }
1565
1566 //------------------------inline_string_getCharsU--------------------------
1567 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1568 bool LibraryCallKit::inline_string_getCharsU() {
1569 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1570 return false;
1571 }
1572
1573 // Get the arguments.
1574 Node* src = argument(0);
1575 Node* src_begin = argument(1);
1576 Node* src_end = argument(2); // exclusive offset (i < src_end)
1577 Node* dst = argument(3);
1578 Node* dst_begin = argument(4);
1579
1580 // Check for allocation before we add nodes that would confuse
1581 // tightly_coupled_allocation()
1582 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1583
1584 // Check if a null path was taken unconditionally.
1585 src = null_check(src);
1586 dst = null_check(dst);
1587 if (stopped()) {
1588 return true;
1589 }
1590
1591 // Get length and convert char[] offset to byte[] offset
1592 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1593 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1594
1595 // Range checks
1596 generate_string_range_check(src, src_begin, length, true);
1597 generate_string_range_check(dst, dst_begin, length, false);
1598 if (stopped()) {
1599 return true;
1600 }
1601
1602 if (!stopped()) {
1603 // Calculate starting addresses.
1604 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1605 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1606
1607 // Check if array addresses are aligned to HeapWordSize
1608 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1609 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1610 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1611 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1612
1613 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1614 const char* copyfunc_name = "arraycopy";
1615 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1616 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1617 OptoRuntime::fast_arraycopy_Type(),
1618 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1619 src_start, dst_start, ConvI2X(length) XTOP);
1620 // Do not let reads from the cloned object float above the arraycopy.
1621 if (alloc != nullptr) {
1622 if (alloc->maybe_set_complete(&_gvn)) {
1623 // "You break it, you buy it."
1624 InitializeNode* init = alloc->initialization();
1625 assert(init->is_complete(), "we just did this");
1626 init->set_complete_with_arraycopy();
1627 assert(dst->is_CheckCastPP(), "sanity");
1628 assert(dst->in(0)->in(0) == init, "dest pinned");
1629 }
1630 // Do not let stores that initialize this object be reordered with
1631 // a subsequent store that would make this object accessible by
1632 // other threads.
1633 // Record what AllocateNode this StoreStore protects so that
1634 // escape analysis can go from the MemBarStoreStoreNode to the
1635 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1636 // based on the escape status of the AllocateNode.
1637 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1638 } else {
1639 insert_mem_bar(Op_MemBarCPUOrder);
1640 }
1641 }
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 return true;
1645 }
1646
1647 //----------------------inline_string_char_access----------------------------
1648 // Store/Load char to/from byte[] array.
1649 // static void StringUTF16.putChar(byte[] val, int index, int c)
1650 // static char StringUTF16.getChar(byte[] val, int index)
1651 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1652 Node* value = argument(0);
1653 Node* index = argument(1);
1654 Node* ch = is_store ? argument(2) : nullptr;
1655
1656 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1657 // correctly requires matched array shapes.
1658 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1659 "sanity: byte[] and char[] bases agree");
1660 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1661 "sanity: byte[] and char[] scales agree");
1662
1663 // Bail when getChar over constants is requested: constant folding would
1664 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1665 // Java method would constant fold nicely instead.
1666 if (!is_store && value->is_Con() && index->is_Con()) {
1667 return false;
1668 }
1669
1670 // Save state and restore on bailout
1671 uint old_sp = sp();
1672 SafePointNode* old_map = clone_map();
1673
1674 value = must_be_not_null(value, true);
1675
1676 Node* adr = array_element_address(value, index, T_CHAR);
1677 if (adr->is_top()) {
1678 set_map(old_map);
1679 set_sp(old_sp);
1680 return false;
1681 }
1682 destruct_map_clone(old_map);
1683 if (is_store) {
1684 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1685 } else {
1686 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);
1687 set_result(ch);
1688 }
1689 return true;
1690 }
1691
1692 //--------------------------round_double_node--------------------------------
1693 // Round a double node if necessary.
1694 Node* LibraryCallKit::round_double_node(Node* n) {
1695 if (Matcher::strict_fp_requires_explicit_rounding) {
1696 #ifdef IA32
1697 if (UseSSE < 2) {
1698 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1699 }
1700 #else
1701 Unimplemented();
1702 #endif // IA32
1703 }
1704 return n;
1705 }
1706
1707 //------------------------------inline_math-----------------------------------
1708 // public static double Math.abs(double)
1709 // public static double Math.sqrt(double)
1710 // public static double Math.log(double)
1711 // public static double Math.log10(double)
1712 // public static double Math.round(double)
1713 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1714 Node* arg = round_double_node(argument(0));
1715 Node* n = nullptr;
1716 switch (id) {
1717 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1718 case vmIntrinsics::_dsqrt:
1719 case vmIntrinsics::_dsqrt_strict:
1720 n = new SqrtDNode(C, control(), arg); break;
1721 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1722 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1723 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1724 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1725 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1726 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1727 default: fatal_unexpected_iid(id); break;
1728 }
1729 set_result(_gvn.transform(n));
1730 return true;
1731 }
1732
1733 //------------------------------inline_math-----------------------------------
1734 // public static float Math.abs(float)
1735 // public static int Math.abs(int)
1736 // public static long Math.abs(long)
1737 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1738 Node* arg = argument(0);
1739 Node* n = nullptr;
1740 switch (id) {
1741 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1742 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1743 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1744 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1745 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1746 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1747 default: fatal_unexpected_iid(id); break;
1748 }
1749 set_result(_gvn.transform(n));
1750 return true;
1751 }
1752
1753 //------------------------------runtime_math-----------------------------
1754 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1755 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1756 "must be (DD)D or (D)D type");
1757
1758 // Inputs
1759 Node* a = round_double_node(argument(0));
1760 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1761
1762 const TypePtr* no_memory_effects = nullptr;
1763 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1764 no_memory_effects,
1765 a, top(), b, b ? top() : nullptr);
1766 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1767 #ifdef ASSERT
1768 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1769 assert(value_top == top(), "second value must be top");
1770 #endif
1771
1772 set_result(value);
1773 return true;
1774 }
1775
1776 //------------------------------inline_math_pow-----------------------------
1777 bool LibraryCallKit::inline_math_pow() {
1778 Node* exp = round_double_node(argument(2));
1779 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1780 if (d != nullptr) {
1781 if (d->getd() == 2.0) {
1782 // Special case: pow(x, 2.0) => x * x
1783 Node* base = round_double_node(argument(0));
1784 set_result(_gvn.transform(new MulDNode(base, base)));
1785 return true;
1786 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1787 // Special case: pow(x, 0.5) => sqrt(x)
1788 Node* base = round_double_node(argument(0));
1789 Node* zero = _gvn.zerocon(T_DOUBLE);
1790
1791 RegionNode* region = new RegionNode(3);
1792 Node* phi = new PhiNode(region, Type::DOUBLE);
1793
1794 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1795 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1796 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1797 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1798 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1799
1800 Node* if_pow = generate_slow_guard(test, nullptr);
1801 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1802 phi->init_req(1, value_sqrt);
1803 region->init_req(1, control());
1804
1805 if (if_pow != nullptr) {
1806 set_control(if_pow);
1807 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1808 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1809 const TypePtr* no_memory_effects = nullptr;
1810 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1811 no_memory_effects, base, top(), exp, top());
1812 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1813 #ifdef ASSERT
1814 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1815 assert(value_top == top(), "second value must be top");
1816 #endif
1817 phi->init_req(2, value_pow);
1818 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1819 }
1820
1821 C->set_has_split_ifs(true); // Has chance for split-if optimization
1822 set_control(_gvn.transform(region));
1823 record_for_igvn(region);
1824 set_result(_gvn.transform(phi));
1825
1826 return true;
1827 }
1828 }
1829
1830 return StubRoutines::dpow() != nullptr ?
1831 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1832 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1833 }
1834
1835 //------------------------------inline_math_native-----------------------------
1836 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1837 switch (id) {
1838 case vmIntrinsics::_dsin:
1839 return StubRoutines::dsin() != nullptr ?
1840 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1841 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1842 case vmIntrinsics::_dcos:
1843 return StubRoutines::dcos() != nullptr ?
1844 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1845 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1846 case vmIntrinsics::_dtan:
1847 return StubRoutines::dtan() != nullptr ?
1848 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1849 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1850 case vmIntrinsics::_dexp:
1851 return StubRoutines::dexp() != nullptr ?
1852 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1853 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1854 case vmIntrinsics::_dlog:
1855 return StubRoutines::dlog() != nullptr ?
1856 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1857 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1858 case vmIntrinsics::_dlog10:
1859 return StubRoutines::dlog10() != nullptr ?
1860 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1861 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1862
1863 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1864 case vmIntrinsics::_ceil:
1865 case vmIntrinsics::_floor:
1866 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1867
1868 case vmIntrinsics::_dsqrt:
1869 case vmIntrinsics::_dsqrt_strict:
1870 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1871 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1872 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1873 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1874 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1875
1876 case vmIntrinsics::_dpow: return inline_math_pow();
1877 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1878 case vmIntrinsics::_fcopySign: return inline_math(id);
1879 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1880 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1881 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1882
1883 // These intrinsics are not yet correctly implemented
1884 case vmIntrinsics::_datan2:
1885 return false;
1886
1887 default:
1888 fatal_unexpected_iid(id);
1889 return false;
1890 }
1891 }
1892
1893 //----------------------------inline_notify-----------------------------------*
1894 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1895 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1896 address func;
1897 if (id == vmIntrinsics::_notify) {
1898 func = OptoRuntime::monitor_notify_Java();
1899 } else {
1900 func = OptoRuntime::monitor_notifyAll_Java();
1901 }
1902 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1903 make_slow_call_ex(call, env()->Throwable_klass(), false);
1904 return true;
1905 }
1906
1907
1908 //----------------------------inline_min_max-----------------------------------
1909 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1910 set_result(generate_min_max(id, argument(0), argument(1)));
1911 return true;
1912 }
1913
1914 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1915 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1916 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1917 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1918 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1919
1920 {
1921 PreserveJVMState pjvms(this);
1922 PreserveReexecuteState preexecs(this);
1923 jvms()->set_should_reexecute(true);
1924
1925 set_control(slow_path);
1926 set_i_o(i_o());
1927
1928 uncommon_trap(Deoptimization::Reason_intrinsic,
1929 Deoptimization::Action_none);
1930 }
1931
1932 set_control(fast_path);
1933 set_result(math);
1934 }
1935
1936 template <typename OverflowOp>
1937 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1938 typedef typename OverflowOp::MathOp MathOp;
1939
1940 MathOp* mathOp = new MathOp(arg1, arg2);
1941 Node* operation = _gvn.transform( mathOp );
1942 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1943 inline_math_mathExact(operation, ofcheck);
1944 return true;
1945 }
1946
1947 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1948 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1949 }
1950
1951 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1952 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1953 }
1954
1955 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1956 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1957 }
1958
1959 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1960 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1961 }
1962
1963 bool LibraryCallKit::inline_math_negateExactI() {
1964 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1965 }
1966
1967 bool LibraryCallKit::inline_math_negateExactL() {
1968 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1969 }
1970
1971 bool LibraryCallKit::inline_math_multiplyExactI() {
1972 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1973 }
1974
1975 bool LibraryCallKit::inline_math_multiplyExactL() {
1976 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1977 }
1978
1979 bool LibraryCallKit::inline_math_multiplyHigh() {
1980 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1981 return true;
1982 }
1983
1984 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1985 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1986 return true;
1987 }
1988
1989 Node*
1990 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1991 Node* result_val = nullptr;
1992 switch (id) {
1993 case vmIntrinsics::_min:
1994 case vmIntrinsics::_min_strict:
1995 result_val = _gvn.transform(new MinINode(x0, y0));
1996 break;
1997 case vmIntrinsics::_max:
1998 case vmIntrinsics::_max_strict:
1999 result_val = _gvn.transform(new MaxINode(x0, y0));
2000 break;
2001 default:
2002 fatal_unexpected_iid(id);
2003 break;
2004 }
2005 return result_val;
2006 }
2007
2008 inline int
2009 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2010 const TypePtr* base_type = TypePtr::NULL_PTR;
2011 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2012 if (base_type == nullptr) {
2013 // Unknown type.
2014 return Type::AnyPtr;
2015 } else if (base_type == TypePtr::NULL_PTR) {
2016 // Since this is a null+long form, we have to switch to a rawptr.
2017 base = _gvn.transform(new CastX2PNode(offset));
2018 offset = MakeConX(0);
2019 return Type::RawPtr;
2020 } else if (base_type->base() == Type::RawPtr) {
2021 return Type::RawPtr;
2022 } else if (base_type->isa_oopptr()) {
2023 // Base is never null => always a heap address.
2024 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2025 return Type::OopPtr;
2026 }
2027 // Offset is small => always a heap address.
2028 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2029 if (offset_type != nullptr &&
2030 base_type->offset() == 0 && // (should always be?)
2031 offset_type->_lo >= 0 &&
2032 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2033 return Type::OopPtr;
2034 } else if (type == T_OBJECT) {
2035 // off heap access to an oop doesn't make any sense. Has to be on
2036 // heap.
2037 return Type::OopPtr;
2038 }
2039 // Otherwise, it might either be oop+off or null+addr.
2040 return Type::AnyPtr;
2041 } else {
2042 // No information:
2043 return Type::AnyPtr;
2044 }
2045 }
2046
2047 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2048 Node* uncasted_base = base;
2049 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2050 if (kind == Type::RawPtr) {
2051 return basic_plus_adr(top(), uncasted_base, offset);
2052 } else if (kind == Type::AnyPtr) {
2053 assert(base == uncasted_base, "unexpected base change");
2054 if (can_cast) {
2055 if (!_gvn.type(base)->speculative_maybe_null() &&
2056 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2057 // According to profiling, this access is always on
2058 // heap. Casting the base to not null and thus avoiding membars
2059 // around the access should allow better optimizations
2060 Node* null_ctl = top();
2061 base = null_check_oop(base, &null_ctl, true, true, true);
2062 assert(null_ctl->is_top(), "no null control here");
2063 return basic_plus_adr(base, offset);
2064 } else if (_gvn.type(base)->speculative_always_null() &&
2065 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2066 // According to profiling, this access is always off
2067 // heap.
2068 base = null_assert(base);
2069 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2070 offset = MakeConX(0);
2071 return basic_plus_adr(top(), raw_base, offset);
2072 }
2073 }
2074 // We don't know if it's an on heap or off heap access. Fall back
2075 // to raw memory access.
2076 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2077 return basic_plus_adr(top(), raw, offset);
2078 } else {
2079 assert(base == uncasted_base, "unexpected base change");
2080 // We know it's an on heap access so base can't be null
2081 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2082 base = must_be_not_null(base, true);
2083 }
2084 return basic_plus_adr(base, offset);
2085 }
2086 }
2087
2088 //--------------------------inline_number_methods-----------------------------
2089 // inline int Integer.numberOfLeadingZeros(int)
2090 // inline int Long.numberOfLeadingZeros(long)
2091 //
2092 // inline int Integer.numberOfTrailingZeros(int)
2093 // inline int Long.numberOfTrailingZeros(long)
2094 //
2095 // inline int Integer.bitCount(int)
2096 // inline int Long.bitCount(long)
2097 //
2098 // inline char Character.reverseBytes(char)
2099 // inline short Short.reverseBytes(short)
2100 // inline int Integer.reverseBytes(int)
2101 // inline long Long.reverseBytes(long)
2102 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2103 Node* arg = argument(0);
2104 Node* n = nullptr;
2105 switch (id) {
2106 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2107 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2108 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2109 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2110 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2111 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2112 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2113 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2114 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2115 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2116 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2117 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2118 default: fatal_unexpected_iid(id); break;
2119 }
2120 set_result(_gvn.transform(n));
2121 return true;
2122 }
2123
2124 //--------------------------inline_bitshuffle_methods-----------------------------
2125 // inline int Integer.compress(int, int)
2126 // inline int Integer.expand(int, int)
2127 // inline long Long.compress(long, long)
2128 // inline long Long.expand(long, long)
2129 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2130 Node* n = nullptr;
2131 switch (id) {
2132 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2133 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2134 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2135 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2136 default: fatal_unexpected_iid(id); break;
2137 }
2138 set_result(_gvn.transform(n));
2139 return true;
2140 }
2141
2142 //--------------------------inline_number_methods-----------------------------
2143 // inline int Integer.compareUnsigned(int, int)
2144 // inline int Long.compareUnsigned(long, long)
2145 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2146 Node* arg1 = argument(0);
2147 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2148 Node* n = nullptr;
2149 switch (id) {
2150 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2151 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2152 default: fatal_unexpected_iid(id); break;
2153 }
2154 set_result(_gvn.transform(n));
2155 return true;
2156 }
2157
2158 //--------------------------inline_unsigned_divmod_methods-----------------------------
2159 // inline int Integer.divideUnsigned(int, int)
2160 // inline int Integer.remainderUnsigned(int, int)
2161 // inline long Long.divideUnsigned(long, long)
2162 // inline long Long.remainderUnsigned(long, long)
2163 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2164 Node* n = nullptr;
2165 switch (id) {
2166 case vmIntrinsics::_divideUnsigned_i: {
2167 zero_check_int(argument(1));
2168 // Compile-time detect of null-exception
2169 if (stopped()) {
2170 return true; // keep the graph constructed so far
2171 }
2172 n = new UDivINode(control(), argument(0), argument(1));
2173 break;
2174 }
2175 case vmIntrinsics::_divideUnsigned_l: {
2176 zero_check_long(argument(2));
2177 // Compile-time detect of null-exception
2178 if (stopped()) {
2179 return true; // keep the graph constructed so far
2180 }
2181 n = new UDivLNode(control(), argument(0), argument(2));
2182 break;
2183 }
2184 case vmIntrinsics::_remainderUnsigned_i: {
2185 zero_check_int(argument(1));
2186 // Compile-time detect of null-exception
2187 if (stopped()) {
2188 return true; // keep the graph constructed so far
2189 }
2190 n = new UModINode(control(), argument(0), argument(1));
2191 break;
2192 }
2193 case vmIntrinsics::_remainderUnsigned_l: {
2194 zero_check_long(argument(2));
2195 // Compile-time detect of null-exception
2196 if (stopped()) {
2197 return true; // keep the graph constructed so far
2198 }
2199 n = new UModLNode(control(), argument(0), argument(2));
2200 break;
2201 }
2202 default: fatal_unexpected_iid(id); break;
2203 }
2204 set_result(_gvn.transform(n));
2205 return true;
2206 }
2207
2208 //----------------------------inline_unsafe_access----------------------------
2209
2210 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2211 // Attempt to infer a sharper value type from the offset and base type.
2212 ciKlass* sharpened_klass = nullptr;
2213
2214 // See if it is an instance field, with an object type.
2215 if (alias_type->field() != nullptr) {
2216 if (alias_type->field()->type()->is_klass()) {
2217 sharpened_klass = alias_type->field()->type()->as_klass();
2218 }
2219 }
2220
2221 const TypeOopPtr* result = nullptr;
2222 // See if it is a narrow oop array.
2223 if (adr_type->isa_aryptr()) {
2224 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2225 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2226 if (elem_type != nullptr && elem_type->is_loaded()) {
2227 // Sharpen the value type.
2228 result = elem_type;
2229 }
2230 }
2231 }
2232
2233 // The sharpened class might be unloaded if there is no class loader
2234 // contraint in place.
2235 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2236 // Sharpen the value type.
2237 result = TypeOopPtr::make_from_klass(sharpened_klass);
2238 }
2239 if (result != nullptr) {
2240 #ifndef PRODUCT
2241 if (C->print_intrinsics() || C->print_inlining()) {
2242 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2243 tty->print(" sharpened value: "); result->dump(); tty->cr();
2244 }
2245 #endif
2246 }
2247 return result;
2248 }
2249
2250 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2251 switch (kind) {
2252 case Relaxed:
2253 return MO_UNORDERED;
2254 case Opaque:
2255 return MO_RELAXED;
2256 case Acquire:
2257 return MO_ACQUIRE;
2258 case Release:
2259 return MO_RELEASE;
2260 case Volatile:
2261 return MO_SEQ_CST;
2262 default:
2263 ShouldNotReachHere();
2264 return 0;
2265 }
2266 }
2267
2268 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2269 if (callee()->is_static()) return false; // caller must have the capability!
2270 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2271 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2272 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2273 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2274
2275 if (is_reference_type(type)) {
2276 decorators |= ON_UNKNOWN_OOP_REF;
2277 }
2278
2279 if (unaligned) {
2280 decorators |= C2_UNALIGNED;
2281 }
2282
2283 #ifndef PRODUCT
2284 {
2285 ResourceMark rm;
2286 // Check the signatures.
2287 ciSignature* sig = callee()->signature();
2288 #ifdef ASSERT
2289 if (!is_store) {
2290 // Object getReference(Object base, int/long offset), etc.
2291 BasicType rtype = sig->return_type()->basic_type();
2292 assert(rtype == type, "getter must return the expected value");
2293 assert(sig->count() == 2, "oop getter has 2 arguments");
2294 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2295 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2296 } else {
2297 // void putReference(Object base, int/long offset, Object x), etc.
2298 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2299 assert(sig->count() == 3, "oop putter has 3 arguments");
2300 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2301 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2302 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2303 assert(vtype == type, "putter must accept the expected value");
2304 }
2305 #endif // ASSERT
2306 }
2307 #endif //PRODUCT
2308
2309 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2310
2311 Node* receiver = argument(0); // type: oop
2312
2313 // Build address expression.
2314 Node* heap_base_oop = top();
2315
2316 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2317 Node* base = argument(1); // type: oop
2318 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2319 Node* offset = argument(2); // type: long
2320 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2321 // to be plain byte offsets, which are also the same as those accepted
2322 // by oopDesc::field_addr.
2323 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2324 "fieldOffset must be byte-scaled");
2325 // 32-bit machines ignore the high half!
2326 offset = ConvL2X(offset);
2327
2328 // Save state and restore on bailout
2329 uint old_sp = sp();
2330 SafePointNode* old_map = clone_map();
2331
2332 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2333
2334 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2335 if (type != T_OBJECT) {
2336 decorators |= IN_NATIVE; // off-heap primitive access
2337 } else {
2338 set_map(old_map);
2339 set_sp(old_sp);
2340 return false; // off-heap oop accesses are not supported
2341 }
2342 } else {
2343 heap_base_oop = base; // on-heap or mixed access
2344 }
2345
2346 // Can base be null? Otherwise, always on-heap access.
2347 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2348
2349 if (!can_access_non_heap) {
2350 decorators |= IN_HEAP;
2351 }
2352
2353 Node* val = is_store ? argument(4) : nullptr;
2354
2355 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2356 if (adr_type == TypePtr::NULL_PTR) {
2357 set_map(old_map);
2358 set_sp(old_sp);
2359 return false; // off-heap access with zero address
2360 }
2361
2362 // Try to categorize the address.
2363 Compile::AliasType* alias_type = C->alias_type(adr_type);
2364 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2365
2366 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2367 alias_type->adr_type() == TypeAryPtr::RANGE) {
2368 set_map(old_map);
2369 set_sp(old_sp);
2370 return false; // not supported
2371 }
2372
2373 bool mismatched = false;
2374 BasicType bt = alias_type->basic_type();
2375 if (bt != T_ILLEGAL) {
2376 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2377 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2378 // Alias type doesn't differentiate between byte[] and boolean[]).
2379 // Use address type to get the element type.
2380 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2381 }
2382 if (is_reference_type(bt, true)) {
2383 // accessing an array field with getReference is not a mismatch
2384 bt = T_OBJECT;
2385 }
2386 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2387 // Don't intrinsify mismatched object accesses
2388 set_map(old_map);
2389 set_sp(old_sp);
2390 return false;
2391 }
2392 mismatched = (bt != type);
2393 } else if (alias_type->adr_type()->isa_oopptr()) {
2394 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2395 }
2396
2397 destruct_map_clone(old_map);
2398 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2399
2400 if (mismatched) {
2401 decorators |= C2_MISMATCHED;
2402 }
2403
2404 // First guess at the value type.
2405 const Type *value_type = Type::get_const_basic_type(type);
2406
2407 // Figure out the memory ordering.
2408 decorators |= mo_decorator_for_access_kind(kind);
2409
2410 if (!is_store && type == T_OBJECT) {
2411 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2412 if (tjp != nullptr) {
2413 value_type = tjp;
2414 }
2415 }
2416
2417 receiver = null_check(receiver);
2418 if (stopped()) {
2419 return true;
2420 }
2421 // Heap pointers get a null-check from the interpreter,
2422 // as a courtesy. However, this is not guaranteed by Unsafe,
2423 // and it is not possible to fully distinguish unintended nulls
2424 // from intended ones in this API.
2425
2426 if (!is_store) {
2427 Node* p = nullptr;
2428 // Try to constant fold a load from a constant field
2429 ciField* field = alias_type->field();
2430 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2431 // final or stable field
2432 p = make_constant_from_field(field, heap_base_oop);
2433 }
2434
2435 if (p == nullptr) { // Could not constant fold the load
2436 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2437 // Normalize the value returned by getBoolean in the following cases
2438 if (type == T_BOOLEAN &&
2439 (mismatched ||
2440 heap_base_oop == top() || // - heap_base_oop is null or
2441 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2442 // and the unsafe access is made to large offset
2443 // (i.e., larger than the maximum offset necessary for any
2444 // field access)
2445 ) {
2446 IdealKit ideal = IdealKit(this);
2447 #define __ ideal.
2448 IdealVariable normalized_result(ideal);
2449 __ declarations_done();
2450 __ set(normalized_result, p);
2451 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2452 __ set(normalized_result, ideal.ConI(1));
2453 ideal.end_if();
2454 final_sync(ideal);
2455 p = __ value(normalized_result);
2456 #undef __
2457 }
2458 }
2459 if (type == T_ADDRESS) {
2460 p = gvn().transform(new CastP2XNode(nullptr, p));
2461 p = ConvX2UL(p);
2462 }
2463 // The load node has the control of the preceding MemBarCPUOrder. All
2464 // following nodes will have the control of the MemBarCPUOrder inserted at
2465 // the end of this method. So, pushing the load onto the stack at a later
2466 // point is fine.
2467 set_result(p);
2468 } else {
2469 if (bt == T_ADDRESS) {
2470 // Repackage the long as a pointer.
2471 val = ConvL2X(val);
2472 val = gvn().transform(new CastX2PNode(val));
2473 }
2474 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2475 }
2476
2477 return true;
2478 }
2479
2480 //----------------------------inline_unsafe_load_store----------------------------
2481 // This method serves a couple of different customers (depending on LoadStoreKind):
2482 //
2483 // LS_cmp_swap:
2484 //
2485 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2486 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2487 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2488 //
2489 // LS_cmp_swap_weak:
2490 //
2491 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2492 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2493 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2494 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2495 //
2496 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2497 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2498 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2499 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2500 //
2501 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2502 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2503 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2504 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2505 //
2506 // LS_cmp_exchange:
2507 //
2508 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2509 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2510 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2511 //
2512 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2513 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2514 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2515 //
2516 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2517 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2518 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2519 //
2520 // LS_get_add:
2521 //
2522 // int getAndAddInt( Object o, long offset, int delta)
2523 // long getAndAddLong(Object o, long offset, long delta)
2524 //
2525 // LS_get_set:
2526 //
2527 // int getAndSet(Object o, long offset, int newValue)
2528 // long getAndSet(Object o, long offset, long newValue)
2529 // Object getAndSet(Object o, long offset, Object newValue)
2530 //
2531 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2532 // This basic scheme here is the same as inline_unsafe_access, but
2533 // differs in enough details that combining them would make the code
2534 // overly confusing. (This is a true fact! I originally combined
2535 // them, but even I was confused by it!) As much code/comments as
2536 // possible are retained from inline_unsafe_access though to make
2537 // the correspondences clearer. - dl
2538
2539 if (callee()->is_static()) return false; // caller must have the capability!
2540
2541 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2542 decorators |= mo_decorator_for_access_kind(access_kind);
2543
2544 #ifndef PRODUCT
2545 BasicType rtype;
2546 {
2547 ResourceMark rm;
2548 // Check the signatures.
2549 ciSignature* sig = callee()->signature();
2550 rtype = sig->return_type()->basic_type();
2551 switch(kind) {
2552 case LS_get_add:
2553 case LS_get_set: {
2554 // Check the signatures.
2555 #ifdef ASSERT
2556 assert(rtype == type, "get and set must return the expected type");
2557 assert(sig->count() == 3, "get and set has 3 arguments");
2558 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2559 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2560 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2561 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2562 #endif // ASSERT
2563 break;
2564 }
2565 case LS_cmp_swap:
2566 case LS_cmp_swap_weak: {
2567 // Check the signatures.
2568 #ifdef ASSERT
2569 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2570 assert(sig->count() == 4, "CAS has 4 arguments");
2571 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2572 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2573 #endif // ASSERT
2574 break;
2575 }
2576 case LS_cmp_exchange: {
2577 // Check the signatures.
2578 #ifdef ASSERT
2579 assert(rtype == type, "CAS must return the expected type");
2580 assert(sig->count() == 4, "CAS has 4 arguments");
2581 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2582 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2583 #endif // ASSERT
2584 break;
2585 }
2586 default:
2587 ShouldNotReachHere();
2588 }
2589 }
2590 #endif //PRODUCT
2591
2592 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2593
2594 // Get arguments:
2595 Node* receiver = nullptr;
2596 Node* base = nullptr;
2597 Node* offset = nullptr;
2598 Node* oldval = nullptr;
2599 Node* newval = nullptr;
2600 switch(kind) {
2601 case LS_cmp_swap:
2602 case LS_cmp_swap_weak:
2603 case LS_cmp_exchange: {
2604 const bool two_slot_type = type2size[type] == 2;
2605 receiver = argument(0); // type: oop
2606 base = argument(1); // type: oop
2607 offset = argument(2); // type: long
2608 oldval = argument(4); // type: oop, int, or long
2609 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2610 break;
2611 }
2612 case LS_get_add:
2613 case LS_get_set: {
2614 receiver = argument(0); // type: oop
2615 base = argument(1); // type: oop
2616 offset = argument(2); // type: long
2617 oldval = nullptr;
2618 newval = argument(4); // type: oop, int, or long
2619 break;
2620 }
2621 default:
2622 ShouldNotReachHere();
2623 }
2624
2625 // Build field offset expression.
2626 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2627 // to be plain byte offsets, which are also the same as those accepted
2628 // by oopDesc::field_addr.
2629 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2630 // 32-bit machines ignore the high half of long offsets
2631 offset = ConvL2X(offset);
2632 // Save state and restore on bailout
2633 uint old_sp = sp();
2634 SafePointNode* old_map = clone_map();
2635 Node* adr = make_unsafe_address(base, offset,type, false);
2636 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2637
2638 Compile::AliasType* alias_type = C->alias_type(adr_type);
2639 BasicType bt = alias_type->basic_type();
2640 if (bt != T_ILLEGAL &&
2641 (is_reference_type(bt) != (type == T_OBJECT))) {
2642 // Don't intrinsify mismatched object accesses.
2643 set_map(old_map);
2644 set_sp(old_sp);
2645 return false;
2646 }
2647
2648 destruct_map_clone(old_map);
2649
2650 // For CAS, unlike inline_unsafe_access, there seems no point in
2651 // trying to refine types. Just use the coarse types here.
2652 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2653 const Type *value_type = Type::get_const_basic_type(type);
2654
2655 switch (kind) {
2656 case LS_get_set:
2657 case LS_cmp_exchange: {
2658 if (type == T_OBJECT) {
2659 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2660 if (tjp != nullptr) {
2661 value_type = tjp;
2662 }
2663 }
2664 break;
2665 }
2666 case LS_cmp_swap:
2667 case LS_cmp_swap_weak:
2668 case LS_get_add:
2669 break;
2670 default:
2671 ShouldNotReachHere();
2672 }
2673
2674 // Null check receiver.
2675 receiver = null_check(receiver);
2676 if (stopped()) {
2677 return true;
2678 }
2679
2680 int alias_idx = C->get_alias_index(adr_type);
2681
2682 if (is_reference_type(type)) {
2683 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2684
2685 // Transformation of a value which could be null pointer (CastPP #null)
2686 // could be delayed during Parse (for example, in adjust_map_after_if()).
2687 // Execute transformation here to avoid barrier generation in such case.
2688 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2689 newval = _gvn.makecon(TypePtr::NULL_PTR);
2690
2691 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2692 // Refine the value to a null constant, when it is known to be null
2693 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2694 }
2695 }
2696
2697 Node* result = nullptr;
2698 switch (kind) {
2699 case LS_cmp_exchange: {
2700 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2701 oldval, newval, value_type, type, decorators);
2702 break;
2703 }
2704 case LS_cmp_swap_weak:
2705 decorators |= C2_WEAK_CMPXCHG;
2706 case LS_cmp_swap: {
2707 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2708 oldval, newval, value_type, type, decorators);
2709 break;
2710 }
2711 case LS_get_set: {
2712 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2713 newval, value_type, type, decorators);
2714 break;
2715 }
2716 case LS_get_add: {
2717 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2718 newval, value_type, type, decorators);
2719 break;
2720 }
2721 default:
2722 ShouldNotReachHere();
2723 }
2724
2725 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2726 set_result(result);
2727 return true;
2728 }
2729
2730 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2731 // Regardless of form, don't allow previous ld/st to move down,
2732 // then issue acquire, release, or volatile mem_bar.
2733 insert_mem_bar(Op_MemBarCPUOrder);
2734 switch(id) {
2735 case vmIntrinsics::_loadFence:
2736 insert_mem_bar(Op_LoadFence);
2737 return true;
2738 case vmIntrinsics::_storeFence:
2739 insert_mem_bar(Op_StoreFence);
2740 return true;
2741 case vmIntrinsics::_storeStoreFence:
2742 insert_mem_bar(Op_StoreStoreFence);
2743 return true;
2744 case vmIntrinsics::_fullFence:
2745 insert_mem_bar(Op_MemBarVolatile);
2746 return true;
2747 default:
2748 fatal_unexpected_iid(id);
2749 return false;
2750 }
2751 }
2752
2753 bool LibraryCallKit::inline_onspinwait() {
2754 insert_mem_bar(Op_OnSpinWait);
2755 return true;
2756 }
2757
2758 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2759 if (!kls->is_Con()) {
2760 return true;
2761 }
2762 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2763 if (klsptr == nullptr) {
2764 return true;
2765 }
2766 ciInstanceKlass* ik = klsptr->instance_klass();
2767 // don't need a guard for a klass that is already initialized
2768 return !ik->is_initialized();
2769 }
2770
2771 //----------------------------inline_unsafe_writeback0-------------------------
2772 // public native void Unsafe.writeback0(long address)
2773 bool LibraryCallKit::inline_unsafe_writeback0() {
2774 if (!Matcher::has_match_rule(Op_CacheWB)) {
2775 return false;
2776 }
2777 #ifndef PRODUCT
2778 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2779 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2780 ciSignature* sig = callee()->signature();
2781 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2782 #endif
2783 null_check_receiver(); // null-check, then ignore
2784 Node *addr = argument(1);
2785 addr = new CastX2PNode(addr);
2786 addr = _gvn.transform(addr);
2787 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2788 flush = _gvn.transform(flush);
2789 set_memory(flush, TypeRawPtr::BOTTOM);
2790 return true;
2791 }
2792
2793 //----------------------------inline_unsafe_writeback0-------------------------
2794 // public native void Unsafe.writeback0(long address)
2795 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2796 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2797 return false;
2798 }
2799 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2800 return false;
2801 }
2802 #ifndef PRODUCT
2803 assert(Matcher::has_match_rule(Op_CacheWB),
2804 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2805 : "found match rule for CacheWBPostSync but not CacheWB"));
2806
2807 #endif
2808 null_check_receiver(); // null-check, then ignore
2809 Node *sync;
2810 if (is_pre) {
2811 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2812 } else {
2813 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2814 }
2815 sync = _gvn.transform(sync);
2816 set_memory(sync, TypeRawPtr::BOTTOM);
2817 return true;
2818 }
2819
2820 //----------------------------inline_unsafe_allocate---------------------------
2821 // public native Object Unsafe.allocateInstance(Class<?> cls);
2822 bool LibraryCallKit::inline_unsafe_allocate() {
2823
2824 #if INCLUDE_JVMTI
2825 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2826 return false;
2827 }
2828 #endif //INCLUDE_JVMTI
2829
2830 if (callee()->is_static()) return false; // caller must have the capability!
2831
2832 null_check_receiver(); // null-check, then ignore
2833 Node* cls = null_check(argument(1));
2834 if (stopped()) return true;
2835
2836 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2837 kls = null_check(kls);
2838 if (stopped()) return true; // argument was like int.class
2839
2840 #if INCLUDE_JVMTI
2841 // Don't try to access new allocated obj in the intrinsic.
2842 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2843 // Deoptimize and allocate in interpreter instead.
2844 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2845 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2846 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2847 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2848 {
2849 BuildCutout unless(this, tst, PROB_MAX);
2850 uncommon_trap(Deoptimization::Reason_intrinsic,
2851 Deoptimization::Action_make_not_entrant);
2852 }
2853 if (stopped()) {
2854 return true;
2855 }
2856 #endif //INCLUDE_JVMTI
2857
2858 Node* test = nullptr;
2859 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2860 // Note: The argument might still be an illegal value like
2861 // Serializable.class or Object[].class. The runtime will handle it.
2862 // But we must make an explicit check for initialization.
2863 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2864 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2865 // can generate code to load it as unsigned byte.
2866 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2867 Node* bits = intcon(InstanceKlass::fully_initialized);
2868 test = _gvn.transform(new SubINode(inst, bits));
2869 // The 'test' is non-zero if we need to take a slow path.
2870 }
2871
2872 Node* obj = new_instance(kls, test);
2873 set_result(obj);
2874 return true;
2875 }
2876
2877 //------------------------inline_native_time_funcs--------------
2878 // inline code for System.currentTimeMillis() and System.nanoTime()
2879 // these have the same type and signature
2880 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2881 const TypeFunc* tf = OptoRuntime::void_long_Type();
2882 const TypePtr* no_memory_effects = nullptr;
2883 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2884 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2885 #ifdef ASSERT
2886 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2887 assert(value_top == top(), "second value must be top");
2888 #endif
2889 set_result(value);
2890 return true;
2891 }
2892
2893
2894 #if INCLUDE_JVMTI
2895
2896 // When notifications are disabled then just update the VTMS transition bit and return.
2897 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2898 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2899 if (!DoJVMTIVirtualThreadTransitions) {
2900 return true;
2901 }
2902 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2903 IdealKit ideal(this);
2904
2905 Node* ONE = ideal.ConI(1);
2906 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2907 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2908 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2909
2910 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2911 sync_kit(ideal);
2912 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2913 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2914 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2915 ideal.sync_kit(this);
2916 } ideal.else_(); {
2917 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2918 Node* thread = ideal.thread();
2919 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2920 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2921 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2922
2923 sync_kit(ideal);
2924 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2925 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2926
2927 ideal.sync_kit(this);
2928 } ideal.end_if();
2929 final_sync(ideal);
2930
2931 return true;
2932 }
2933
2934 // Always update the temporary VTMS transition bit.
2935 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2936 if (!DoJVMTIVirtualThreadTransitions) {
2937 return true;
2938 }
2939 IdealKit ideal(this);
2940
2941 {
2942 // unconditionally update the temporary VTMS transition bit in current JavaThread
2943 Node* thread = ideal.thread();
2944 Node* hide = _gvn.transform(argument(1)); // hide argument for temporary VTMS transition notification
2945 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2946 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2947
2948 sync_kit(ideal);
2949 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2950 ideal.sync_kit(this);
2951 }
2952 final_sync(ideal);
2953
2954 return true;
2955 }
2956
2957 #endif // INCLUDE_JVMTI
2958
2959 #ifdef JFR_HAVE_INTRINSICS
2960
2961 /**
2962 * if oop->klass != null
2963 * // normal class
2964 * epoch = _epoch_state ? 2 : 1
2965 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2966 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2967 * }
2968 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2969 * else
2970 * // primitive class
2971 * if oop->array_klass != null
2972 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2973 * else
2974 * id = LAST_TYPE_ID + 1 // void class path
2975 * if (!signaled)
2976 * signaled = true
2977 */
2978 bool LibraryCallKit::inline_native_classID() {
2979 Node* cls = argument(0);
2980
2981 IdealKit ideal(this);
2982 #define __ ideal.
2983 IdealVariable result(ideal); __ declarations_done();
2984 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
2985 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2986 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2987
2988
2989 __ if_then(kls, BoolTest::ne, null()); {
2990 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2991 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2992
2993 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
2994 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2995 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2996 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2997 mask = _gvn.transform(new OrLNode(mask, epoch));
2998 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2999
3000 float unlikely = PROB_UNLIKELY(0.999);
3001 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3002 sync_kit(ideal);
3003 make_runtime_call(RC_LEAF,
3004 OptoRuntime::class_id_load_barrier_Type(),
3005 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3006 "class id load barrier",
3007 TypePtr::BOTTOM,
3008 kls);
3009 ideal.sync_kit(this);
3010 } __ end_if();
3011
3012 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3013 } __ else_(); {
3014 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3015 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3016 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3017 __ if_then(array_kls, BoolTest::ne, null()); {
3018 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3019 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3020 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3021 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3022 } __ else_(); {
3023 // void class case
3024 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3025 } __ end_if();
3026
3027 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3028 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3029 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3030 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3031 } __ end_if();
3032 } __ end_if();
3033
3034 final_sync(ideal);
3035 set_result(ideal.value(result));
3036 #undef __
3037 return true;
3038 }
3039
3040 //------------------------inline_native_jvm_commit------------------
3041 bool LibraryCallKit::inline_native_jvm_commit() {
3042 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3043
3044 // Save input memory and i_o state.
3045 Node* input_memory_state = reset_memory();
3046 set_all_memory(input_memory_state);
3047 Node* input_io_state = i_o();
3048
3049 // TLS.
3050 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3051 // Jfr java buffer.
3052 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3053 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3054 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3055
3056 // Load the current value of the notified field in the JfrThreadLocal.
3057 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3058 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3059
3060 // Test for notification.
3061 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3062 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3063 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3064
3065 // True branch, is notified.
3066 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3067 set_control(is_notified);
3068
3069 // Reset notified state.
3070 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3071
3072 // 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.
3073 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3074 // Convert the machine-word to a long.
3075 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3076
3077 // False branch, not notified.
3078 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3079 set_control(not_notified);
3080 set_all_memory(input_memory_state);
3081
3082 // Arg is the next position as a long.
3083 Node* arg = argument(0);
3084 // Convert long to machine-word.
3085 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3086
3087 // Store the next_position to the underlying jfr java buffer.
3088 Node* commit_memory;
3089 #ifdef _LP64
3090 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3091 #else
3092 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3093 #endif
3094
3095 // 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.
3096 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3097 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3098 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3099
3100 // And flags with lease constant.
3101 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3102
3103 // Branch on lease to conditionalize returning the leased java buffer.
3104 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3105 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3106 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3107
3108 // False branch, not a lease.
3109 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3110
3111 // True branch, is lease.
3112 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3113 set_control(is_lease);
3114
3115 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3116 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3117 OptoRuntime::void_void_Type(),
3118 StubRoutines::jfr_return_lease(),
3119 "return_lease", TypePtr::BOTTOM);
3120 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3121
3122 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3123 record_for_igvn(lease_compare_rgn);
3124 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3125 record_for_igvn(lease_compare_mem);
3126 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3127 record_for_igvn(lease_compare_io);
3128 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3129 record_for_igvn(lease_result_value);
3130
3131 // Update control and phi nodes.
3132 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3133 lease_compare_rgn->init_req(_false_path, not_lease);
3134
3135 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3136 lease_compare_mem->init_req(_false_path, commit_memory);
3137
3138 lease_compare_io->init_req(_true_path, i_o());
3139 lease_compare_io->init_req(_false_path, input_io_state);
3140
3141 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3142 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3143
3144 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3145 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3146 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3147 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3148
3149 // Update control and phi nodes.
3150 result_rgn->init_req(_true_path, is_notified);
3151 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3152
3153 result_mem->init_req(_true_path, notified_reset_memory);
3154 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3155
3156 result_io->init_req(_true_path, input_io_state);
3157 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3158
3159 result_value->init_req(_true_path, current_pos);
3160 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3161
3162 // Set output state.
3163 set_control(_gvn.transform(result_rgn));
3164 set_all_memory(_gvn.transform(result_mem));
3165 set_i_o(_gvn.transform(result_io));
3166 set_result(result_rgn, result_value);
3167 return true;
3168 }
3169
3170 /*
3171 * The intrinsic is a model of this pseudo-code:
3172 *
3173 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3174 * jobject h_event_writer = tl->java_event_writer();
3175 * if (h_event_writer == nullptr) {
3176 * return nullptr;
3177 * }
3178 * oop threadObj = Thread::threadObj();
3179 * oop vthread = java_lang_Thread::vthread(threadObj);
3180 * traceid tid;
3181 * bool excluded;
3182 * if (vthread != threadObj) { // i.e. current thread is virtual
3183 * tid = java_lang_Thread::tid(vthread);
3184 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3185 * excluded = vthread_epoch_raw & excluded_mask;
3186 * if (!excluded) {
3187 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3188 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3189 * if (vthread_epoch != current_epoch) {
3190 * write_checkpoint();
3191 * }
3192 * }
3193 * } else {
3194 * tid = java_lang_Thread::tid(threadObj);
3195 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3196 * excluded = thread_epoch_raw & excluded_mask;
3197 * }
3198 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3199 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3200 * if (tid_in_event_writer != tid) {
3201 * setField(event_writer, "threadID", tid);
3202 * setField(event_writer, "excluded", excluded);
3203 * }
3204 * return event_writer
3205 */
3206 bool LibraryCallKit::inline_native_getEventWriter() {
3207 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3208
3209 // Save input memory and i_o state.
3210 Node* input_memory_state = reset_memory();
3211 set_all_memory(input_memory_state);
3212 Node* input_io_state = i_o();
3213
3214 Node* excluded_mask = _gvn.intcon(32768);
3215 Node* epoch_mask = _gvn.intcon(32767);
3216
3217 // TLS
3218 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3219
3220 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3221 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3222
3223 // Load the eventwriter jobject handle.
3224 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3225
3226 // Null check the jobject handle.
3227 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3228 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3229 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3230
3231 // False path, jobj is null.
3232 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3233
3234 // True path, jobj is not null.
3235 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3236
3237 set_control(jobj_is_not_null);
3238
3239 // Load the threadObj for the CarrierThread.
3240 Node* threadObj = generate_current_thread(tls_ptr);
3241
3242 // Load the vthread.
3243 Node* vthread = generate_virtual_thread(tls_ptr);
3244
3245 // If vthread != threadObj, this is a virtual thread.
3246 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3247 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3248 IfNode* iff_vthread_not_equal_threadObj =
3249 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3250
3251 // False branch, fallback to threadObj.
3252 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3253 set_control(vthread_equal_threadObj);
3254
3255 // Load the tid field from the vthread object.
3256 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3257
3258 // Load the raw epoch value from the threadObj.
3259 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3260 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3261 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3262
3263 // Mask off the excluded information from the epoch.
3264 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3265
3266 // True branch, this is a virtual thread.
3267 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3268 set_control(vthread_not_equal_threadObj);
3269
3270 // Load the tid field from the vthread object.
3271 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3272
3273 // Load the raw epoch value from the vthread.
3274 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3275 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3276 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3277
3278 // Mask off the excluded information from the epoch.
3279 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3280
3281 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3282 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3283 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3284 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3285
3286 // False branch, vthread is excluded, no need to write epoch info.
3287 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3288
3289 // True branch, vthread is included, update epoch info.
3290 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3291 set_control(included);
3292
3293 // Get epoch value.
3294 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3295
3296 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3297 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3298 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3299
3300 // Compare the epoch in the vthread to the current epoch generation.
3301 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3302 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3303 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3304
3305 // False path, epoch is equal, checkpoint information is valid.
3306 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3307
3308 // True path, epoch is not equal, write a checkpoint for the vthread.
3309 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3310
3311 set_control(epoch_is_not_equal);
3312
3313 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3314 // The call also updates the native thread local thread id and the vthread with the current epoch.
3315 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3316 OptoRuntime::jfr_write_checkpoint_Type(),
3317 StubRoutines::jfr_write_checkpoint(),
3318 "write_checkpoint", TypePtr::BOTTOM);
3319 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3320
3321 // vthread epoch != current epoch
3322 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3323 record_for_igvn(epoch_compare_rgn);
3324 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3325 record_for_igvn(epoch_compare_mem);
3326 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3327 record_for_igvn(epoch_compare_io);
3328
3329 // Update control and phi nodes.
3330 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3331 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3332 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3333 epoch_compare_mem->init_req(_false_path, input_memory_state);
3334 epoch_compare_io->init_req(_true_path, i_o());
3335 epoch_compare_io->init_req(_false_path, input_io_state);
3336
3337 // excluded != true
3338 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3339 record_for_igvn(exclude_compare_rgn);
3340 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3341 record_for_igvn(exclude_compare_mem);
3342 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3343 record_for_igvn(exclude_compare_io);
3344
3345 // Update control and phi nodes.
3346 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3347 exclude_compare_rgn->init_req(_false_path, excluded);
3348 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3349 exclude_compare_mem->init_req(_false_path, input_memory_state);
3350 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3351 exclude_compare_io->init_req(_false_path, input_io_state);
3352
3353 // vthread != threadObj
3354 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3355 record_for_igvn(vthread_compare_rgn);
3356 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3357 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3358 record_for_igvn(vthread_compare_io);
3359 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3360 record_for_igvn(tid);
3361 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3362 record_for_igvn(exclusion);
3363
3364 // Update control and phi nodes.
3365 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3366 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3367 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3368 vthread_compare_mem->init_req(_false_path, input_memory_state);
3369 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3370 vthread_compare_io->init_req(_false_path, input_io_state);
3371 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3372 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3373 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3374 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3375
3376 // Update branch state.
3377 set_control(_gvn.transform(vthread_compare_rgn));
3378 set_all_memory(_gvn.transform(vthread_compare_mem));
3379 set_i_o(_gvn.transform(vthread_compare_io));
3380
3381 // Load the event writer oop by dereferencing the jobject handle.
3382 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3383 assert(klass_EventWriter->is_loaded(), "invariant");
3384 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3385 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3386 const TypeOopPtr* const xtype = aklass->as_instance_type();
3387 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3388 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3389
3390 // Load the current thread id from the event writer object.
3391 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3392 // Get the field offset to, conditionally, store an updated tid value later.
3393 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3394 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3395 // Get the field offset to, conditionally, store an updated exclusion value later.
3396 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3397 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3398
3399 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3400 record_for_igvn(event_writer_tid_compare_rgn);
3401 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3402 record_for_igvn(event_writer_tid_compare_mem);
3403 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3404 record_for_igvn(event_writer_tid_compare_io);
3405
3406 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3407 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3408 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3409 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3410
3411 // False path, tids are the same.
3412 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3413
3414 // True path, tid is not equal, need to update the tid in the event writer.
3415 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3416 record_for_igvn(tid_is_not_equal);
3417
3418 // Store the exclusion state to the event writer.
3419 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3420
3421 // Store the tid to the event writer.
3422 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3423
3424 // Update control and phi nodes.
3425 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3426 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3427 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3428 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3429 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3430 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3431
3432 // Result of top level CFG, Memory, IO and Value.
3433 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3434 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3435 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3436 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3437
3438 // Result control.
3439 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3440 result_rgn->init_req(_false_path, jobj_is_null);
3441
3442 // Result memory.
3443 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3444 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3445
3446 // Result IO.
3447 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3448 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3449
3450 // Result value.
3451 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3452 result_value->init_req(_false_path, null()); // return null
3453
3454 // Set output state.
3455 set_control(_gvn.transform(result_rgn));
3456 set_all_memory(_gvn.transform(result_mem));
3457 set_i_o(_gvn.transform(result_io));
3458 set_result(result_rgn, result_value);
3459 return true;
3460 }
3461
3462 /*
3463 * The intrinsic is a model of this pseudo-code:
3464 *
3465 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3466 * if (carrierThread != thread) { // is virtual thread
3467 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3468 * bool excluded = vthread_epoch_raw & excluded_mask;
3469 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3470 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3471 * if (!excluded) {
3472 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3473 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3474 * }
3475 * Atomic::release_store(&tl->_vthread, true);
3476 * return;
3477 * }
3478 * Atomic::release_store(&tl->_vthread, false);
3479 */
3480 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3481 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3482
3483 Node* input_memory_state = reset_memory();
3484 set_all_memory(input_memory_state);
3485
3486 Node* excluded_mask = _gvn.intcon(32768);
3487 Node* epoch_mask = _gvn.intcon(32767);
3488
3489 Node* const carrierThread = generate_current_thread(jt);
3490 // If thread != carrierThread, this is a virtual thread.
3491 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3492 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3493 IfNode* iff_thread_not_equal_carrierThread =
3494 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3495
3496 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3497
3498 // False branch, is carrierThread.
3499 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3500 // Store release
3501 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3502
3503 set_all_memory(input_memory_state);
3504
3505 // True branch, is virtual thread.
3506 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3507 set_control(thread_not_equal_carrierThread);
3508
3509 // Load the raw epoch value from the vthread.
3510 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3511 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3512 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3513
3514 // Mask off the excluded information from the epoch.
3515 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3516
3517 // Load the tid field from the thread.
3518 Node* tid = load_field_from_object(thread, "tid", "J");
3519
3520 // Store the vthread tid to the jfr thread local.
3521 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3522 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3523
3524 // Branch is_excluded to conditionalize updating the epoch .
3525 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3526 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3527 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3528
3529 // True branch, vthread is excluded, no need to write epoch info.
3530 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3531 set_control(excluded);
3532 Node* vthread_is_excluded = _gvn.intcon(1);
3533
3534 // False branch, vthread is included, update epoch info.
3535 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3536 set_control(included);
3537 Node* vthread_is_included = _gvn.intcon(0);
3538
3539 // Get epoch value.
3540 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3541
3542 // Store the vthread epoch to the jfr thread local.
3543 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3544 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3545
3546 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3547 record_for_igvn(excluded_rgn);
3548 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3549 record_for_igvn(excluded_mem);
3550 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3551 record_for_igvn(exclusion);
3552
3553 // Merge the excluded control and memory.
3554 excluded_rgn->init_req(_true_path, excluded);
3555 excluded_rgn->init_req(_false_path, included);
3556 excluded_mem->init_req(_true_path, tid_memory);
3557 excluded_mem->init_req(_false_path, included_memory);
3558 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3559 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3560
3561 // Set intermediate state.
3562 set_control(_gvn.transform(excluded_rgn));
3563 set_all_memory(excluded_mem);
3564
3565 // Store the vthread exclusion state to the jfr thread local.
3566 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3567 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3568
3569 // Store release
3570 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3571
3572 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3573 record_for_igvn(thread_compare_rgn);
3574 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3575 record_for_igvn(thread_compare_mem);
3576 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3577 record_for_igvn(vthread);
3578
3579 // Merge the thread_compare control and memory.
3580 thread_compare_rgn->init_req(_true_path, control());
3581 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3582 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3583 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3584
3585 // Set output state.
3586 set_control(_gvn.transform(thread_compare_rgn));
3587 set_all_memory(_gvn.transform(thread_compare_mem));
3588 }
3589
3590 #endif // JFR_HAVE_INTRINSICS
3591
3592 //------------------------inline_native_currentCarrierThread------------------
3593 bool LibraryCallKit::inline_native_currentCarrierThread() {
3594 Node* junk = nullptr;
3595 set_result(generate_current_thread(junk));
3596 return true;
3597 }
3598
3599 //------------------------inline_native_currentThread------------------
3600 bool LibraryCallKit::inline_native_currentThread() {
3601 Node* junk = nullptr;
3602 set_result(generate_virtual_thread(junk));
3603 return true;
3604 }
3605
3606 //------------------------inline_native_setVthread------------------
3607 bool LibraryCallKit::inline_native_setCurrentThread() {
3608 assert(C->method()->changes_current_thread(),
3609 "method changes current Thread but is not annotated ChangesCurrentThread");
3610 Node* arr = argument(1);
3611 Node* thread = _gvn.transform(new ThreadLocalNode());
3612 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3613 Node* thread_obj_handle
3614 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3615 thread_obj_handle = _gvn.transform(thread_obj_handle);
3616 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3617 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3618 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3619 return true;
3620 }
3621
3622 const Type* LibraryCallKit::scopedValueCache_type() {
3623 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3624 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3625 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3626
3627 // Because we create the scopedValue cache lazily we have to make the
3628 // type of the result BotPTR.
3629 bool xk = etype->klass_is_exact();
3630 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3631 return objects_type;
3632 }
3633
3634 Node* LibraryCallKit::scopedValueCache_helper() {
3635 Node* thread = _gvn.transform(new ThreadLocalNode());
3636 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3637 // We cannot use immutable_memory() because we might flip onto a
3638 // different carrier thread, at which point we'll need to use that
3639 // carrier thread's cache.
3640 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3641 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3642 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3643 }
3644
3645 //------------------------inline_native_scopedValueCache------------------
3646 bool LibraryCallKit::inline_native_scopedValueCache() {
3647 Node* cache_obj_handle = scopedValueCache_helper();
3648 const Type* objects_type = scopedValueCache_type();
3649 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3650
3651 return true;
3652 }
3653
3654 //------------------------inline_native_setScopedValueCache------------------
3655 bool LibraryCallKit::inline_native_setScopedValueCache() {
3656 Node* arr = argument(0);
3657 Node* cache_obj_handle = scopedValueCache_helper();
3658 const Type* objects_type = scopedValueCache_type();
3659
3660 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3661 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3662
3663 return true;
3664 }
3665
3666 //---------------------------load_mirror_from_klass----------------------------
3667 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3668 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3669 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3670 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3671 // mirror = ((OopHandle)mirror)->resolve();
3672 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3673 }
3674
3675 //-----------------------load_klass_from_mirror_common-------------------------
3676 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3677 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3678 // and branch to the given path on the region.
3679 // If never_see_null, take an uncommon trap on null, so we can optimistically
3680 // compile for the non-null case.
3681 // If the region is null, force never_see_null = true.
3682 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3683 bool never_see_null,
3684 RegionNode* region,
3685 int null_path,
3686 int offset) {
3687 if (region == nullptr) never_see_null = true;
3688 Node* p = basic_plus_adr(mirror, offset);
3689 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3690 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3691 Node* null_ctl = top();
3692 kls = null_check_oop(kls, &null_ctl, never_see_null);
3693 if (region != nullptr) {
3694 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3695 region->init_req(null_path, null_ctl);
3696 } else {
3697 assert(null_ctl == top(), "no loose ends");
3698 }
3699 return kls;
3700 }
3701
3702 //--------------------(inline_native_Class_query helpers)---------------------
3703 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3704 // Fall through if (mods & mask) == bits, take the guard otherwise.
3705 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3706 // Branch around if the given klass has the given modifier bit set.
3707 // Like generate_guard, adds a new path onto the region.
3708 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3709 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3710 Node* mask = intcon(modifier_mask);
3711 Node* bits = intcon(modifier_bits);
3712 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3713 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3714 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3715 return generate_fair_guard(bol, region);
3716 }
3717 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3718 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3719 }
3720 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3721 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3722 }
3723
3724 //-------------------------inline_native_Class_query-------------------
3725 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3726 const Type* return_type = TypeInt::BOOL;
3727 Node* prim_return_value = top(); // what happens if it's a primitive class?
3728 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3729 bool expect_prim = false; // most of these guys expect to work on refs
3730
3731 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3732
3733 Node* mirror = argument(0);
3734 Node* obj = top();
3735
3736 switch (id) {
3737 case vmIntrinsics::_isInstance:
3738 // nothing is an instance of a primitive type
3739 prim_return_value = intcon(0);
3740 obj = argument(1);
3741 break;
3742 case vmIntrinsics::_getModifiers:
3743 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3744 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3745 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3746 break;
3747 case vmIntrinsics::_isInterface:
3748 prim_return_value = intcon(0);
3749 break;
3750 case vmIntrinsics::_isArray:
3751 prim_return_value = intcon(0);
3752 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3753 break;
3754 case vmIntrinsics::_isPrimitive:
3755 prim_return_value = intcon(1);
3756 expect_prim = true; // obviously
3757 break;
3758 case vmIntrinsics::_isHidden:
3759 prim_return_value = intcon(0);
3760 break;
3761 case vmIntrinsics::_getSuperclass:
3762 prim_return_value = null();
3763 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3764 break;
3765 case vmIntrinsics::_getClassAccessFlags:
3766 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3767 return_type = TypeInt::INT; // not bool! 6297094
3768 break;
3769 default:
3770 fatal_unexpected_iid(id);
3771 break;
3772 }
3773
3774 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3775 if (mirror_con == nullptr) return false; // cannot happen?
3776
3777 #ifndef PRODUCT
3778 if (C->print_intrinsics() || C->print_inlining()) {
3779 ciType* k = mirror_con->java_mirror_type();
3780 if (k) {
3781 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3782 k->print_name();
3783 tty->cr();
3784 }
3785 }
3786 #endif
3787
3788 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3789 RegionNode* region = new RegionNode(PATH_LIMIT);
3790 record_for_igvn(region);
3791 PhiNode* phi = new PhiNode(region, return_type);
3792
3793 // The mirror will never be null of Reflection.getClassAccessFlags, however
3794 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3795 // if it is. See bug 4774291.
3796
3797 // For Reflection.getClassAccessFlags(), the null check occurs in
3798 // the wrong place; see inline_unsafe_access(), above, for a similar
3799 // situation.
3800 mirror = null_check(mirror);
3801 // If mirror or obj is dead, only null-path is taken.
3802 if (stopped()) return true;
3803
3804 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3805
3806 // Now load the mirror's klass metaobject, and null-check it.
3807 // Side-effects region with the control path if the klass is null.
3808 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3809 // If kls is null, we have a primitive mirror.
3810 phi->init_req(_prim_path, prim_return_value);
3811 if (stopped()) { set_result(region, phi); return true; }
3812 bool safe_for_replace = (region->in(_prim_path) == top());
3813
3814 Node* p; // handy temp
3815 Node* null_ctl;
3816
3817 // Now that we have the non-null klass, we can perform the real query.
3818 // For constant classes, the query will constant-fold in LoadNode::Value.
3819 Node* query_value = top();
3820 switch (id) {
3821 case vmIntrinsics::_isInstance:
3822 // nothing is an instance of a primitive type
3823 query_value = gen_instanceof(obj, kls, safe_for_replace);
3824 break;
3825
3826 case vmIntrinsics::_getModifiers:
3827 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3828 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3829 break;
3830
3831 case vmIntrinsics::_isInterface:
3832 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3833 if (generate_interface_guard(kls, region) != nullptr)
3834 // A guard was added. If the guard is taken, it was an interface.
3835 phi->add_req(intcon(1));
3836 // If we fall through, it's a plain class.
3837 query_value = intcon(0);
3838 break;
3839
3840 case vmIntrinsics::_isArray:
3841 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3842 if (generate_array_guard(kls, region) != nullptr)
3843 // A guard was added. If the guard is taken, it was an array.
3844 phi->add_req(intcon(1));
3845 // If we fall through, it's a plain class.
3846 query_value = intcon(0);
3847 break;
3848
3849 case vmIntrinsics::_isPrimitive:
3850 query_value = intcon(0); // "normal" path produces false
3851 break;
3852
3853 case vmIntrinsics::_isHidden:
3854 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3855 if (generate_hidden_class_guard(kls, region) != nullptr)
3856 // A guard was added. If the guard is taken, it was an hidden class.
3857 phi->add_req(intcon(1));
3858 // If we fall through, it's a plain class.
3859 query_value = intcon(0);
3860 break;
3861
3862
3863 case vmIntrinsics::_getSuperclass:
3864 // The rules here are somewhat unfortunate, but we can still do better
3865 // with random logic than with a JNI call.
3866 // Interfaces store null or Object as _super, but must report null.
3867 // Arrays store an intermediate super as _super, but must report Object.
3868 // Other types can report the actual _super.
3869 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3870 if (generate_interface_guard(kls, region) != nullptr)
3871 // A guard was added. If the guard is taken, it was an interface.
3872 phi->add_req(null());
3873 if (generate_array_guard(kls, region) != nullptr)
3874 // A guard was added. If the guard is taken, it was an array.
3875 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3876 // If we fall through, it's a plain class. Get its _super.
3877 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3878 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3879 null_ctl = top();
3880 kls = null_check_oop(kls, &null_ctl);
3881 if (null_ctl != top()) {
3882 // If the guard is taken, Object.superClass is null (both klass and mirror).
3883 region->add_req(null_ctl);
3884 phi ->add_req(null());
3885 }
3886 if (!stopped()) {
3887 query_value = load_mirror_from_klass(kls);
3888 }
3889 break;
3890
3891 case vmIntrinsics::_getClassAccessFlags:
3892 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3893 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3894 break;
3895
3896 default:
3897 fatal_unexpected_iid(id);
3898 break;
3899 }
3900
3901 // Fall-through is the normal case of a query to a real class.
3902 phi->init_req(1, query_value);
3903 region->init_req(1, control());
3904
3905 C->set_has_split_ifs(true); // Has chance for split-if optimization
3906 set_result(region, phi);
3907 return true;
3908 }
3909
3910 //-------------------------inline_Class_cast-------------------
3911 bool LibraryCallKit::inline_Class_cast() {
3912 Node* mirror = argument(0); // Class
3913 Node* obj = argument(1);
3914 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3915 if (mirror_con == nullptr) {
3916 return false; // dead path (mirror->is_top()).
3917 }
3918 if (obj == nullptr || obj->is_top()) {
3919 return false; // dead path
3920 }
3921 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3922
3923 // First, see if Class.cast() can be folded statically.
3924 // java_mirror_type() returns non-null for compile-time Class constants.
3925 ciType* tm = mirror_con->java_mirror_type();
3926 if (tm != nullptr && tm->is_klass() &&
3927 tp != nullptr) {
3928 if (!tp->is_loaded()) {
3929 // Don't use intrinsic when class is not loaded.
3930 return false;
3931 } else {
3932 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
3933 if (static_res == Compile::SSC_always_true) {
3934 // isInstance() is true - fold the code.
3935 set_result(obj);
3936 return true;
3937 } else if (static_res == Compile::SSC_always_false) {
3938 // Don't use intrinsic, have to throw ClassCastException.
3939 // If the reference is null, the non-intrinsic bytecode will
3940 // be optimized appropriately.
3941 return false;
3942 }
3943 }
3944 }
3945
3946 // Bailout intrinsic and do normal inlining if exception path is frequent.
3947 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3948 return false;
3949 }
3950
3951 // Generate dynamic checks.
3952 // Class.cast() is java implementation of _checkcast bytecode.
3953 // Do checkcast (Parse::do_checkcast()) optimizations here.
3954
3955 mirror = null_check(mirror);
3956 // If mirror is dead, only null-path is taken.
3957 if (stopped()) {
3958 return true;
3959 }
3960
3961 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
3962 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3963 RegionNode* region = new RegionNode(PATH_LIMIT);
3964 record_for_igvn(region);
3965
3966 // Now load the mirror's klass metaobject, and null-check it.
3967 // If kls is null, we have a primitive mirror and
3968 // nothing is an instance of a primitive type.
3969 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3970
3971 Node* res = top();
3972 if (!stopped()) {
3973 Node* bad_type_ctrl = top();
3974 // Do checkcast optimizations.
3975 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3976 region->init_req(_bad_type_path, bad_type_ctrl);
3977 }
3978 if (region->in(_prim_path) != top() ||
3979 region->in(_bad_type_path) != top()) {
3980 // Let Interpreter throw ClassCastException.
3981 PreserveJVMState pjvms(this);
3982 set_control(_gvn.transform(region));
3983 uncommon_trap(Deoptimization::Reason_intrinsic,
3984 Deoptimization::Action_maybe_recompile);
3985 }
3986 if (!stopped()) {
3987 set_result(res);
3988 }
3989 return true;
3990 }
3991
3992
3993 //--------------------------inline_native_subtype_check------------------------
3994 // This intrinsic takes the JNI calls out of the heart of
3995 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3996 bool LibraryCallKit::inline_native_subtype_check() {
3997 // Pull both arguments off the stack.
3998 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3999 args[0] = argument(0);
4000 args[1] = argument(1);
4001 Node* klasses[2]; // corresponding Klasses: superk, subk
4002 klasses[0] = klasses[1] = top();
4003
4004 enum {
4005 // A full decision tree on {superc is prim, subc is prim}:
4006 _prim_0_path = 1, // {P,N} => false
4007 // {P,P} & superc!=subc => false
4008 _prim_same_path, // {P,P} & superc==subc => true
4009 _prim_1_path, // {N,P} => false
4010 _ref_subtype_path, // {N,N} & subtype check wins => true
4011 _both_ref_path, // {N,N} & subtype check loses => false
4012 PATH_LIMIT
4013 };
4014
4015 RegionNode* region = new RegionNode(PATH_LIMIT);
4016 Node* phi = new PhiNode(region, TypeInt::BOOL);
4017 record_for_igvn(region);
4018
4019 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4020 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4021 int class_klass_offset = java_lang_Class::klass_offset();
4022
4023 // First null-check both mirrors and load each mirror's klass metaobject.
4024 int which_arg;
4025 for (which_arg = 0; which_arg <= 1; which_arg++) {
4026 Node* arg = args[which_arg];
4027 arg = null_check(arg);
4028 if (stopped()) break;
4029 args[which_arg] = arg;
4030
4031 Node* p = basic_plus_adr(arg, class_klass_offset);
4032 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4033 klasses[which_arg] = _gvn.transform(kls);
4034 }
4035
4036 // Having loaded both klasses, test each for null.
4037 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4038 for (which_arg = 0; which_arg <= 1; which_arg++) {
4039 Node* kls = klasses[which_arg];
4040 Node* null_ctl = top();
4041 kls = null_check_oop(kls, &null_ctl, never_see_null);
4042 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4043 region->init_req(prim_path, null_ctl);
4044 if (stopped()) break;
4045 klasses[which_arg] = kls;
4046 }
4047
4048 if (!stopped()) {
4049 // now we have two reference types, in klasses[0..1]
4050 Node* subk = klasses[1]; // the argument to isAssignableFrom
4051 Node* superk = klasses[0]; // the receiver
4052 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4053 // now we have a successful reference subtype check
4054 region->set_req(_ref_subtype_path, control());
4055 }
4056
4057 // If both operands are primitive (both klasses null), then
4058 // we must return true when they are identical primitives.
4059 // It is convenient to test this after the first null klass check.
4060 set_control(region->in(_prim_0_path)); // go back to first null check
4061 if (!stopped()) {
4062 // Since superc is primitive, make a guard for the superc==subc case.
4063 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4064 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4065 generate_guard(bol_eq, region, PROB_FAIR);
4066 if (region->req() == PATH_LIMIT+1) {
4067 // A guard was added. If the added guard is taken, superc==subc.
4068 region->swap_edges(PATH_LIMIT, _prim_same_path);
4069 region->del_req(PATH_LIMIT);
4070 }
4071 region->set_req(_prim_0_path, control()); // Not equal after all.
4072 }
4073
4074 // these are the only paths that produce 'true':
4075 phi->set_req(_prim_same_path, intcon(1));
4076 phi->set_req(_ref_subtype_path, intcon(1));
4077
4078 // pull together the cases:
4079 assert(region->req() == PATH_LIMIT, "sane region");
4080 for (uint i = 1; i < region->req(); i++) {
4081 Node* ctl = region->in(i);
4082 if (ctl == nullptr || ctl == top()) {
4083 region->set_req(i, top());
4084 phi ->set_req(i, top());
4085 } else if (phi->in(i) == nullptr) {
4086 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4087 }
4088 }
4089
4090 set_control(_gvn.transform(region));
4091 set_result(_gvn.transform(phi));
4092 return true;
4093 }
4094
4095 //---------------------generate_array_guard_common------------------------
4096 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4097 bool obj_array, bool not_array) {
4098
4099 if (stopped()) {
4100 return nullptr;
4101 }
4102
4103 // If obj_array/non_array==false/false:
4104 // Branch around if the given klass is in fact an array (either obj or prim).
4105 // If obj_array/non_array==false/true:
4106 // Branch around if the given klass is not an array klass of any kind.
4107 // If obj_array/non_array==true/true:
4108 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4109 // If obj_array/non_array==true/false:
4110 // Branch around if the kls is an oop array (Object[] or subtype)
4111 //
4112 // Like generate_guard, adds a new path onto the region.
4113 jint layout_con = 0;
4114 Node* layout_val = get_layout_helper(kls, layout_con);
4115 if (layout_val == nullptr) {
4116 bool query = (obj_array
4117 ? Klass::layout_helper_is_objArray(layout_con)
4118 : Klass::layout_helper_is_array(layout_con));
4119 if (query == not_array) {
4120 return nullptr; // never a branch
4121 } else { // always a branch
4122 Node* always_branch = control();
4123 if (region != nullptr)
4124 region->add_req(always_branch);
4125 set_control(top());
4126 return always_branch;
4127 }
4128 }
4129 // Now test the correct condition.
4130 jint nval = (obj_array
4131 ? (jint)(Klass::_lh_array_tag_type_value
4132 << Klass::_lh_array_tag_shift)
4133 : Klass::_lh_neutral_value);
4134 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4135 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4136 // invert the test if we are looking for a non-array
4137 if (not_array) btest = BoolTest(btest).negate();
4138 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4139 return generate_fair_guard(bol, region);
4140 }
4141
4142
4143 //-----------------------inline_native_newArray--------------------------
4144 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4145 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4146 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4147 Node* mirror;
4148 Node* count_val;
4149 if (uninitialized) {
4150 null_check_receiver();
4151 mirror = argument(1);
4152 count_val = argument(2);
4153 } else {
4154 mirror = argument(0);
4155 count_val = argument(1);
4156 }
4157
4158 mirror = null_check(mirror);
4159 // If mirror or obj is dead, only null-path is taken.
4160 if (stopped()) return true;
4161
4162 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4163 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4164 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4165 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4166 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4167
4168 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4169 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4170 result_reg, _slow_path);
4171 Node* normal_ctl = control();
4172 Node* no_array_ctl = result_reg->in(_slow_path);
4173
4174 // Generate code for the slow case. We make a call to newArray().
4175 set_control(no_array_ctl);
4176 if (!stopped()) {
4177 // Either the input type is void.class, or else the
4178 // array klass has not yet been cached. Either the
4179 // ensuing call will throw an exception, or else it
4180 // will cache the array klass for next time.
4181 PreserveJVMState pjvms(this);
4182 CallJavaNode* slow_call = nullptr;
4183 if (uninitialized) {
4184 // Generate optimized virtual call (holder class 'Unsafe' is final)
4185 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4186 } else {
4187 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4188 }
4189 Node* slow_result = set_results_for_java_call(slow_call);
4190 // this->control() comes from set_results_for_java_call
4191 result_reg->set_req(_slow_path, control());
4192 result_val->set_req(_slow_path, slow_result);
4193 result_io ->set_req(_slow_path, i_o());
4194 result_mem->set_req(_slow_path, reset_memory());
4195 }
4196
4197 set_control(normal_ctl);
4198 if (!stopped()) {
4199 // Normal case: The array type has been cached in the java.lang.Class.
4200 // The following call works fine even if the array type is polymorphic.
4201 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4202 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4203 result_reg->init_req(_normal_path, control());
4204 result_val->init_req(_normal_path, obj);
4205 result_io ->init_req(_normal_path, i_o());
4206 result_mem->init_req(_normal_path, reset_memory());
4207
4208 if (uninitialized) {
4209 // Mark the allocation so that zeroing is skipped
4210 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4211 alloc->maybe_set_complete(&_gvn);
4212 }
4213 }
4214
4215 // Return the combined state.
4216 set_i_o( _gvn.transform(result_io) );
4217 set_all_memory( _gvn.transform(result_mem));
4218
4219 C->set_has_split_ifs(true); // Has chance for split-if optimization
4220 set_result(result_reg, result_val);
4221 return true;
4222 }
4223
4224 //----------------------inline_native_getLength--------------------------
4225 // public static native int java.lang.reflect.Array.getLength(Object array);
4226 bool LibraryCallKit::inline_native_getLength() {
4227 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4228
4229 Node* array = null_check(argument(0));
4230 // If array is dead, only null-path is taken.
4231 if (stopped()) return true;
4232
4233 // Deoptimize if it is a non-array.
4234 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4235
4236 if (non_array != nullptr) {
4237 PreserveJVMState pjvms(this);
4238 set_control(non_array);
4239 uncommon_trap(Deoptimization::Reason_intrinsic,
4240 Deoptimization::Action_maybe_recompile);
4241 }
4242
4243 // If control is dead, only non-array-path is taken.
4244 if (stopped()) return true;
4245
4246 // The works fine even if the array type is polymorphic.
4247 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4248 Node* result = load_array_length(array);
4249
4250 C->set_has_split_ifs(true); // Has chance for split-if optimization
4251 set_result(result);
4252 return true;
4253 }
4254
4255 //------------------------inline_array_copyOf----------------------------
4256 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4257 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4258 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4259 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4260
4261 // Get the arguments.
4262 Node* original = argument(0);
4263 Node* start = is_copyOfRange? argument(1): intcon(0);
4264 Node* end = is_copyOfRange? argument(2): argument(1);
4265 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4266
4267 Node* newcopy = nullptr;
4268
4269 // Set the original stack and the reexecute bit for the interpreter to reexecute
4270 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4271 { PreserveReexecuteState preexecs(this);
4272 jvms()->set_should_reexecute(true);
4273
4274 array_type_mirror = null_check(array_type_mirror);
4275 original = null_check(original);
4276
4277 // Check if a null path was taken unconditionally.
4278 if (stopped()) return true;
4279
4280 Node* orig_length = load_array_length(original);
4281
4282 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4283 klass_node = null_check(klass_node);
4284
4285 RegionNode* bailout = new RegionNode(1);
4286 record_for_igvn(bailout);
4287
4288 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4289 // Bail out if that is so.
4290 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4291 if (not_objArray != nullptr) {
4292 // Improve the klass node's type from the new optimistic assumption:
4293 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4294 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4295 Node* cast = new CastPPNode(klass_node, akls);
4296 cast->init_req(0, control());
4297 klass_node = _gvn.transform(cast);
4298 }
4299
4300 // Bail out if either start or end is negative.
4301 generate_negative_guard(start, bailout, &start);
4302 generate_negative_guard(end, bailout, &end);
4303
4304 Node* length = end;
4305 if (_gvn.type(start) != TypeInt::ZERO) {
4306 length = _gvn.transform(new SubINode(end, start));
4307 }
4308
4309 // Bail out if length is negative.
4310 // Without this the new_array would throw
4311 // NegativeArraySizeException but IllegalArgumentException is what
4312 // should be thrown
4313 generate_negative_guard(length, bailout, &length);
4314
4315 if (bailout->req() > 1) {
4316 PreserveJVMState pjvms(this);
4317 set_control(_gvn.transform(bailout));
4318 uncommon_trap(Deoptimization::Reason_intrinsic,
4319 Deoptimization::Action_maybe_recompile);
4320 }
4321
4322 if (!stopped()) {
4323 // How many elements will we copy from the original?
4324 // The answer is MinI(orig_length - start, length).
4325 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4326 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4327
4328 // Generate a direct call to the right arraycopy function(s).
4329 // We know the copy is disjoint but we might not know if the
4330 // oop stores need checking.
4331 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4332 // This will fail a store-check if x contains any non-nulls.
4333
4334 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4335 // loads/stores but it is legal only if we're sure the
4336 // Arrays.copyOf would succeed. So we need all input arguments
4337 // to the copyOf to be validated, including that the copy to the
4338 // new array won't trigger an ArrayStoreException. That subtype
4339 // check can be optimized if we know something on the type of
4340 // the input array from type speculation.
4341 if (_gvn.type(klass_node)->singleton()) {
4342 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4343 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4344
4345 int test = C->static_subtype_check(superk, subk);
4346 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4347 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4348 if (t_original->speculative_type() != nullptr) {
4349 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4350 }
4351 }
4352 }
4353
4354 bool validated = false;
4355 // Reason_class_check rather than Reason_intrinsic because we
4356 // want to intrinsify even if this traps.
4357 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4358 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4359
4360 if (not_subtype_ctrl != top()) {
4361 PreserveJVMState pjvms(this);
4362 set_control(not_subtype_ctrl);
4363 uncommon_trap(Deoptimization::Reason_class_check,
4364 Deoptimization::Action_make_not_entrant);
4365 assert(stopped(), "Should be stopped");
4366 }
4367 validated = true;
4368 }
4369
4370 if (!stopped()) {
4371 newcopy = new_array(klass_node, length, 0); // no arguments to push
4372
4373 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4374 load_object_klass(original), klass_node);
4375 if (!is_copyOfRange) {
4376 ac->set_copyof(validated);
4377 } else {
4378 ac->set_copyofrange(validated);
4379 }
4380 Node* n = _gvn.transform(ac);
4381 if (n == ac) {
4382 ac->connect_outputs(this);
4383 } else {
4384 assert(validated, "shouldn't transform if all arguments not validated");
4385 set_all_memory(n);
4386 }
4387 }
4388 }
4389 } // original reexecute is set back here
4390
4391 C->set_has_split_ifs(true); // Has chance for split-if optimization
4392 if (!stopped()) {
4393 set_result(newcopy);
4394 }
4395 return true;
4396 }
4397
4398
4399 //----------------------generate_virtual_guard---------------------------
4400 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4401 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4402 RegionNode* slow_region) {
4403 ciMethod* method = callee();
4404 int vtable_index = method->vtable_index();
4405 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4406 "bad index %d", vtable_index);
4407 // Get the Method* out of the appropriate vtable entry.
4408 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4409 vtable_index*vtableEntry::size_in_bytes() +
4410 in_bytes(vtableEntry::method_offset());
4411 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4412 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4413
4414 // Compare the target method with the expected method (e.g., Object.hashCode).
4415 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4416
4417 Node* native_call = makecon(native_call_addr);
4418 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4419 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4420
4421 return generate_slow_guard(test_native, slow_region);
4422 }
4423
4424 //-----------------------generate_method_call----------------------------
4425 // Use generate_method_call to make a slow-call to the real
4426 // method if the fast path fails. An alternative would be to
4427 // use a stub like OptoRuntime::slow_arraycopy_Java.
4428 // This only works for expanding the current library call,
4429 // not another intrinsic. (E.g., don't use this for making an
4430 // arraycopy call inside of the copyOf intrinsic.)
4431 CallJavaNode*
4432 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4433 // When compiling the intrinsic method itself, do not use this technique.
4434 guarantee(callee() != C->method(), "cannot make slow-call to self");
4435
4436 ciMethod* method = callee();
4437 // ensure the JVMS we have will be correct for this call
4438 guarantee(method_id == method->intrinsic_id(), "must match");
4439
4440 const TypeFunc* tf = TypeFunc::make(method);
4441 if (res_not_null) {
4442 assert(tf->return_type() == T_OBJECT, "");
4443 const TypeTuple* range = tf->range();
4444 const Type** fields = TypeTuple::fields(range->cnt());
4445 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4446 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4447 tf = TypeFunc::make(tf->domain(), new_range);
4448 }
4449 CallJavaNode* slow_call;
4450 if (is_static) {
4451 assert(!is_virtual, "");
4452 slow_call = new CallStaticJavaNode(C, tf,
4453 SharedRuntime::get_resolve_static_call_stub(), method);
4454 } else if (is_virtual) {
4455 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4456 int vtable_index = Method::invalid_vtable_index;
4457 if (UseInlineCaches) {
4458 // Suppress the vtable call
4459 } else {
4460 // hashCode and clone are not a miranda methods,
4461 // so the vtable index is fixed.
4462 // No need to use the linkResolver to get it.
4463 vtable_index = method->vtable_index();
4464 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4465 "bad index %d", vtable_index);
4466 }
4467 slow_call = new CallDynamicJavaNode(tf,
4468 SharedRuntime::get_resolve_virtual_call_stub(),
4469 method, vtable_index);
4470 } else { // neither virtual nor static: opt_virtual
4471 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4472 slow_call = new CallStaticJavaNode(C, tf,
4473 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4474 slow_call->set_optimized_virtual(true);
4475 }
4476 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4477 // To be able to issue a direct call (optimized virtual or virtual)
4478 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4479 // about the method being invoked should be attached to the call site to
4480 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4481 slow_call->set_override_symbolic_info(true);
4482 }
4483 set_arguments_for_java_call(slow_call);
4484 set_edges_for_java_call(slow_call);
4485 return slow_call;
4486 }
4487
4488
4489 /**
4490 * Build special case code for calls to hashCode on an object. This call may
4491 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4492 * slightly different code.
4493 */
4494 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4495 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4496 assert(!(is_virtual && is_static), "either virtual, special, or static");
4497
4498 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4499
4500 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4501 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4502 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4503 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4504 Node* obj = nullptr;
4505 if (!is_static) {
4506 // Check for hashing null object
4507 obj = null_check_receiver();
4508 if (stopped()) return true; // unconditionally null
4509 result_reg->init_req(_null_path, top());
4510 result_val->init_req(_null_path, top());
4511 } else {
4512 // Do a null check, and return zero if null.
4513 // System.identityHashCode(null) == 0
4514 obj = argument(0);
4515 Node* null_ctl = top();
4516 obj = null_check_oop(obj, &null_ctl);
4517 result_reg->init_req(_null_path, null_ctl);
4518 result_val->init_req(_null_path, _gvn.intcon(0));
4519 }
4520
4521 // Unconditionally null? Then return right away.
4522 if (stopped()) {
4523 set_control( result_reg->in(_null_path));
4524 if (!stopped())
4525 set_result(result_val->in(_null_path));
4526 return true;
4527 }
4528
4529 // We only go to the fast case code if we pass a number of guards. The
4530 // paths which do not pass are accumulated in the slow_region.
4531 RegionNode* slow_region = new RegionNode(1);
4532 record_for_igvn(slow_region);
4533
4534 // If this is a virtual call, we generate a funny guard. We pull out
4535 // the vtable entry corresponding to hashCode() from the target object.
4536 // If the target method which we are calling happens to be the native
4537 // Object hashCode() method, we pass the guard. We do not need this
4538 // guard for non-virtual calls -- the caller is known to be the native
4539 // Object hashCode().
4540 if (is_virtual) {
4541 // After null check, get the object's klass.
4542 Node* obj_klass = load_object_klass(obj);
4543 generate_virtual_guard(obj_klass, slow_region);
4544 }
4545
4546 // Get the header out of the object, use LoadMarkNode when available
4547 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4548 // The control of the load must be null. Otherwise, the load can move before
4549 // the null check after castPP removal.
4550 Node* no_ctrl = nullptr;
4551 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4552
4553 // Test the header to see if it is unlocked.
4554 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4555 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4556 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4557 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4558 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4559
4560 generate_slow_guard(test_unlocked, slow_region);
4561
4562 // Get the hash value and check to see that it has been properly assigned.
4563 // We depend on hash_mask being at most 32 bits and avoid the use of
4564 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4565 // vm: see markWord.hpp.
4566 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4567 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4568 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4569 // This hack lets the hash bits live anywhere in the mark object now, as long
4570 // as the shift drops the relevant bits into the low 32 bits. Note that
4571 // Java spec says that HashCode is an int so there's no point in capturing
4572 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4573 hshifted_header = ConvX2I(hshifted_header);
4574 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4575
4576 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4577 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4578 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4579
4580 generate_slow_guard(test_assigned, slow_region);
4581
4582 Node* init_mem = reset_memory();
4583 // fill in the rest of the null path:
4584 result_io ->init_req(_null_path, i_o());
4585 result_mem->init_req(_null_path, init_mem);
4586
4587 result_val->init_req(_fast_path, hash_val);
4588 result_reg->init_req(_fast_path, control());
4589 result_io ->init_req(_fast_path, i_o());
4590 result_mem->init_req(_fast_path, init_mem);
4591
4592 // Generate code for the slow case. We make a call to hashCode().
4593 set_control(_gvn.transform(slow_region));
4594 if (!stopped()) {
4595 // No need for PreserveJVMState, because we're using up the present state.
4596 set_all_memory(init_mem);
4597 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4598 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4599 Node* slow_result = set_results_for_java_call(slow_call);
4600 // this->control() comes from set_results_for_java_call
4601 result_reg->init_req(_slow_path, control());
4602 result_val->init_req(_slow_path, slow_result);
4603 result_io ->set_req(_slow_path, i_o());
4604 result_mem ->set_req(_slow_path, reset_memory());
4605 }
4606
4607 // Return the combined state.
4608 set_i_o( _gvn.transform(result_io) );
4609 set_all_memory( _gvn.transform(result_mem));
4610
4611 set_result(result_reg, result_val);
4612 return true;
4613 }
4614
4615 //---------------------------inline_native_getClass----------------------------
4616 // public final native Class<?> java.lang.Object.getClass();
4617 //
4618 // Build special case code for calls to getClass on an object.
4619 bool LibraryCallKit::inline_native_getClass() {
4620 Node* obj = null_check_receiver();
4621 if (stopped()) return true;
4622 set_result(load_mirror_from_klass(load_object_klass(obj)));
4623 return true;
4624 }
4625
4626 //-----------------inline_native_Reflection_getCallerClass---------------------
4627 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4628 //
4629 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4630 //
4631 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4632 // in that it must skip particular security frames and checks for
4633 // caller sensitive methods.
4634 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4635 #ifndef PRODUCT
4636 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4637 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4638 }
4639 #endif
4640
4641 if (!jvms()->has_method()) {
4642 #ifndef PRODUCT
4643 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4644 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4645 }
4646 #endif
4647 return false;
4648 }
4649
4650 // Walk back up the JVM state to find the caller at the required
4651 // depth.
4652 JVMState* caller_jvms = jvms();
4653
4654 // Cf. JVM_GetCallerClass
4655 // NOTE: Start the loop at depth 1 because the current JVM state does
4656 // not include the Reflection.getCallerClass() frame.
4657 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4658 ciMethod* m = caller_jvms->method();
4659 switch (n) {
4660 case 0:
4661 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4662 break;
4663 case 1:
4664 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4665 if (!m->caller_sensitive()) {
4666 #ifndef PRODUCT
4667 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4668 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4669 }
4670 #endif
4671 return false; // bail-out; let JVM_GetCallerClass do the work
4672 }
4673 break;
4674 default:
4675 if (!m->is_ignored_by_security_stack_walk()) {
4676 // We have reached the desired frame; return the holder class.
4677 // Acquire method holder as java.lang.Class and push as constant.
4678 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4679 ciInstance* caller_mirror = caller_klass->java_mirror();
4680 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4681
4682 #ifndef PRODUCT
4683 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4684 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());
4685 tty->print_cr(" JVM state at this point:");
4686 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4687 ciMethod* m = jvms()->of_depth(i)->method();
4688 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4689 }
4690 }
4691 #endif
4692 return true;
4693 }
4694 break;
4695 }
4696 }
4697
4698 #ifndef PRODUCT
4699 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4700 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4701 tty->print_cr(" JVM state at this point:");
4702 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4703 ciMethod* m = jvms()->of_depth(i)->method();
4704 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4705 }
4706 }
4707 #endif
4708
4709 return false; // bail-out; let JVM_GetCallerClass do the work
4710 }
4711
4712 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4713 Node* arg = argument(0);
4714 Node* result = nullptr;
4715
4716 switch (id) {
4717 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4718 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4719 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4720 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4721 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4722 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4723
4724 case vmIntrinsics::_doubleToLongBits: {
4725 // two paths (plus control) merge in a wood
4726 RegionNode *r = new RegionNode(3);
4727 Node *phi = new PhiNode(r, TypeLong::LONG);
4728
4729 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4730 // Build the boolean node
4731 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4732
4733 // Branch either way.
4734 // NaN case is less traveled, which makes all the difference.
4735 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4736 Node *opt_isnan = _gvn.transform(ifisnan);
4737 assert( opt_isnan->is_If(), "Expect an IfNode");
4738 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4739 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4740
4741 set_control(iftrue);
4742
4743 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4744 Node *slow_result = longcon(nan_bits); // return NaN
4745 phi->init_req(1, _gvn.transform( slow_result ));
4746 r->init_req(1, iftrue);
4747
4748 // Else fall through
4749 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4750 set_control(iffalse);
4751
4752 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4753 r->init_req(2, iffalse);
4754
4755 // Post merge
4756 set_control(_gvn.transform(r));
4757 record_for_igvn(r);
4758
4759 C->set_has_split_ifs(true); // Has chance for split-if optimization
4760 result = phi;
4761 assert(result->bottom_type()->isa_long(), "must be");
4762 break;
4763 }
4764
4765 case vmIntrinsics::_floatToIntBits: {
4766 // two paths (plus control) merge in a wood
4767 RegionNode *r = new RegionNode(3);
4768 Node *phi = new PhiNode(r, TypeInt::INT);
4769
4770 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4771 // Build the boolean node
4772 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4773
4774 // Branch either way.
4775 // NaN case is less traveled, which makes all the difference.
4776 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4777 Node *opt_isnan = _gvn.transform(ifisnan);
4778 assert( opt_isnan->is_If(), "Expect an IfNode");
4779 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4780 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4781
4782 set_control(iftrue);
4783
4784 static const jint nan_bits = 0x7fc00000;
4785 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4786 phi->init_req(1, _gvn.transform( slow_result ));
4787 r->init_req(1, iftrue);
4788
4789 // Else fall through
4790 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4791 set_control(iffalse);
4792
4793 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4794 r->init_req(2, iffalse);
4795
4796 // Post merge
4797 set_control(_gvn.transform(r));
4798 record_for_igvn(r);
4799
4800 C->set_has_split_ifs(true); // Has chance for split-if optimization
4801 result = phi;
4802 assert(result->bottom_type()->isa_int(), "must be");
4803 break;
4804 }
4805
4806 default:
4807 fatal_unexpected_iid(id);
4808 break;
4809 }
4810 set_result(_gvn.transform(result));
4811 return true;
4812 }
4813
4814 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4815 Node* arg = argument(0);
4816 Node* result = nullptr;
4817
4818 switch (id) {
4819 case vmIntrinsics::_floatIsInfinite:
4820 result = new IsInfiniteFNode(arg);
4821 break;
4822 case vmIntrinsics::_floatIsFinite:
4823 result = new IsFiniteFNode(arg);
4824 break;
4825 case vmIntrinsics::_doubleIsInfinite:
4826 result = new IsInfiniteDNode(arg);
4827 break;
4828 case vmIntrinsics::_doubleIsFinite:
4829 result = new IsFiniteDNode(arg);
4830 break;
4831 default:
4832 fatal_unexpected_iid(id);
4833 break;
4834 }
4835 set_result(_gvn.transform(result));
4836 return true;
4837 }
4838
4839 //----------------------inline_unsafe_copyMemory-------------------------
4840 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4841
4842 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4843 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4844 const Type* base_t = gvn.type(base);
4845
4846 bool in_native = (base_t == TypePtr::NULL_PTR);
4847 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4848 bool is_mixed = !in_heap && !in_native;
4849
4850 if (is_mixed) {
4851 return true; // mixed accesses can touch both on-heap and off-heap memory
4852 }
4853 if (in_heap) {
4854 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
4855 if (!is_prim_array) {
4856 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4857 // there's not enough type information available to determine proper memory slice for it.
4858 return true;
4859 }
4860 }
4861 return false;
4862 }
4863
4864 bool LibraryCallKit::inline_unsafe_copyMemory() {
4865 if (callee()->is_static()) return false; // caller must have the capability!
4866 null_check_receiver(); // null-check receiver
4867 if (stopped()) return true;
4868
4869 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4870
4871 Node* src_base = argument(1); // type: oop
4872 Node* src_off = ConvL2X(argument(2)); // type: long
4873 Node* dst_base = argument(4); // type: oop
4874 Node* dst_off = ConvL2X(argument(5)); // type: long
4875 Node* size = ConvL2X(argument(7)); // type: long
4876
4877 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4878 "fieldOffset must be byte-scaled");
4879
4880 Node* src_addr = make_unsafe_address(src_base, src_off);
4881 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4882
4883 Node* thread = _gvn.transform(new ThreadLocalNode());
4884 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4885 BasicType doing_unsafe_access_bt = T_BYTE;
4886 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4887
4888 // update volatile field
4889 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4890
4891 int flags = RC_LEAF | RC_NO_FP;
4892
4893 const TypePtr* dst_type = TypePtr::BOTTOM;
4894
4895 // Adjust memory effects of the runtime call based on input values.
4896 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4897 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4898 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4899
4900 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4901 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4902 flags |= RC_NARROW_MEM; // narrow in memory
4903 }
4904 }
4905
4906 // Call it. Note that the length argument is not scaled.
4907 make_runtime_call(flags,
4908 OptoRuntime::fast_arraycopy_Type(),
4909 StubRoutines::unsafe_arraycopy(),
4910 "unsafe_arraycopy",
4911 dst_type,
4912 src_addr, dst_addr, size XTOP);
4913
4914 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4915
4916 return true;
4917 }
4918
4919 #undef XTOP
4920
4921 //------------------------clone_coping-----------------------------------
4922 // Helper function for inline_native_clone.
4923 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4924 assert(obj_size != nullptr, "");
4925 Node* raw_obj = alloc_obj->in(1);
4926 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4927
4928 AllocateNode* alloc = nullptr;
4929 if (ReduceBulkZeroing) {
4930 // We will be completely responsible for initializing this object -
4931 // mark Initialize node as complete.
4932 alloc = AllocateNode::Ideal_allocation(alloc_obj);
4933 // The object was just allocated - there should be no any stores!
4934 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
4935 // Mark as complete_with_arraycopy so that on AllocateNode
4936 // expansion, we know this AllocateNode is initialized by an array
4937 // copy and a StoreStore barrier exists after the array copy.
4938 alloc->initialization()->set_complete_with_arraycopy();
4939 }
4940
4941 Node* size = _gvn.transform(obj_size);
4942 access_clone(obj, alloc_obj, size, is_array);
4943
4944 // Do not let reads from the cloned object float above the arraycopy.
4945 if (alloc != nullptr) {
4946 // Do not let stores that initialize this object be reordered with
4947 // a subsequent store that would make this object accessible by
4948 // other threads.
4949 // Record what AllocateNode this StoreStore protects so that
4950 // escape analysis can go from the MemBarStoreStoreNode to the
4951 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4952 // based on the escape status of the AllocateNode.
4953 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4954 } else {
4955 insert_mem_bar(Op_MemBarCPUOrder);
4956 }
4957 }
4958
4959 //------------------------inline_native_clone----------------------------
4960 // protected native Object java.lang.Object.clone();
4961 //
4962 // Here are the simple edge cases:
4963 // null receiver => normal trap
4964 // virtual and clone was overridden => slow path to out-of-line clone
4965 // not cloneable or finalizer => slow path to out-of-line Object.clone
4966 //
4967 // The general case has two steps, allocation and copying.
4968 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4969 //
4970 // Copying also has two cases, oop arrays and everything else.
4971 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4972 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4973 //
4974 // These steps fold up nicely if and when the cloned object's klass
4975 // can be sharply typed as an object array, a type array, or an instance.
4976 //
4977 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4978 PhiNode* result_val;
4979
4980 // Set the reexecute bit for the interpreter to reexecute
4981 // the bytecode that invokes Object.clone if deoptimization happens.
4982 { PreserveReexecuteState preexecs(this);
4983 jvms()->set_should_reexecute(true);
4984
4985 Node* obj = null_check_receiver();
4986 if (stopped()) return true;
4987
4988 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4989
4990 // If we are going to clone an instance, we need its exact type to
4991 // know the number and types of fields to convert the clone to
4992 // loads/stores. Maybe a speculative type can help us.
4993 if (!obj_type->klass_is_exact() &&
4994 obj_type->speculative_type() != nullptr &&
4995 obj_type->speculative_type()->is_instance_klass()) {
4996 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4997 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4998 !spec_ik->has_injected_fields()) {
4999 if (!obj_type->isa_instptr() ||
5000 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5001 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5002 }
5003 }
5004 }
5005
5006 // Conservatively insert a memory barrier on all memory slices.
5007 // Do not let writes into the original float below the clone.
5008 insert_mem_bar(Op_MemBarCPUOrder);
5009
5010 // paths into result_reg:
5011 enum {
5012 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5013 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5014 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5015 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5016 PATH_LIMIT
5017 };
5018 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5019 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5020 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5021 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5022 record_for_igvn(result_reg);
5023
5024 Node* obj_klass = load_object_klass(obj);
5025 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5026 if (array_ctl != nullptr) {
5027 // It's an array.
5028 PreserveJVMState pjvms(this);
5029 set_control(array_ctl);
5030 Node* obj_length = load_array_length(obj);
5031 Node* array_size = nullptr; // Size of the array without object alignment padding.
5032 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5033
5034 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5035 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5036 // If it is an oop array, it requires very special treatment,
5037 // because gc barriers are required when accessing the array.
5038 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5039 if (is_obja != nullptr) {
5040 PreserveJVMState pjvms2(this);
5041 set_control(is_obja);
5042 // Generate a direct call to the right arraycopy function(s).
5043 // Clones are always tightly coupled.
5044 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5045 ac->set_clone_oop_array();
5046 Node* n = _gvn.transform(ac);
5047 assert(n == ac, "cannot disappear");
5048 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5049
5050 result_reg->init_req(_objArray_path, control());
5051 result_val->init_req(_objArray_path, alloc_obj);
5052 result_i_o ->set_req(_objArray_path, i_o());
5053 result_mem ->set_req(_objArray_path, reset_memory());
5054 }
5055 }
5056 // Otherwise, there are no barriers to worry about.
5057 // (We can dispense with card marks if we know the allocation
5058 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5059 // causes the non-eden paths to take compensating steps to
5060 // simulate a fresh allocation, so that no further
5061 // card marks are required in compiled code to initialize
5062 // the object.)
5063
5064 if (!stopped()) {
5065 copy_to_clone(obj, alloc_obj, array_size, true);
5066
5067 // Present the results of the copy.
5068 result_reg->init_req(_array_path, control());
5069 result_val->init_req(_array_path, alloc_obj);
5070 result_i_o ->set_req(_array_path, i_o());
5071 result_mem ->set_req(_array_path, reset_memory());
5072 }
5073 }
5074
5075 // We only go to the instance fast case code if we pass a number of guards.
5076 // The paths which do not pass are accumulated in the slow_region.
5077 RegionNode* slow_region = new RegionNode(1);
5078 record_for_igvn(slow_region);
5079 if (!stopped()) {
5080 // It's an instance (we did array above). Make the slow-path tests.
5081 // If this is a virtual call, we generate a funny guard. We grab
5082 // the vtable entry corresponding to clone() from the target object.
5083 // If the target method which we are calling happens to be the
5084 // Object clone() method, we pass the guard. We do not need this
5085 // guard for non-virtual calls; the caller is known to be the native
5086 // Object clone().
5087 if (is_virtual) {
5088 generate_virtual_guard(obj_klass, slow_region);
5089 }
5090
5091 // The object must be easily cloneable and must not have a finalizer.
5092 // Both of these conditions may be checked in a single test.
5093 // We could optimize the test further, but we don't care.
5094 generate_access_flags_guard(obj_klass,
5095 // Test both conditions:
5096 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5097 // Must be cloneable but not finalizer:
5098 JVM_ACC_IS_CLONEABLE_FAST,
5099 slow_region);
5100 }
5101
5102 if (!stopped()) {
5103 // It's an instance, and it passed the slow-path tests.
5104 PreserveJVMState pjvms(this);
5105 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5106 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5107 // is reexecuted if deoptimization occurs and there could be problems when merging
5108 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5109 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5110
5111 copy_to_clone(obj, alloc_obj, obj_size, false);
5112
5113 // Present the results of the slow call.
5114 result_reg->init_req(_instance_path, control());
5115 result_val->init_req(_instance_path, alloc_obj);
5116 result_i_o ->set_req(_instance_path, i_o());
5117 result_mem ->set_req(_instance_path, reset_memory());
5118 }
5119
5120 // Generate code for the slow case. We make a call to clone().
5121 set_control(_gvn.transform(slow_region));
5122 if (!stopped()) {
5123 PreserveJVMState pjvms(this);
5124 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5125 // We need to deoptimize on exception (see comment above)
5126 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5127 // this->control() comes from set_results_for_java_call
5128 result_reg->init_req(_slow_path, control());
5129 result_val->init_req(_slow_path, slow_result);
5130 result_i_o ->set_req(_slow_path, i_o());
5131 result_mem ->set_req(_slow_path, reset_memory());
5132 }
5133
5134 // Return the combined state.
5135 set_control( _gvn.transform(result_reg));
5136 set_i_o( _gvn.transform(result_i_o));
5137 set_all_memory( _gvn.transform(result_mem));
5138 } // original reexecute is set back here
5139
5140 set_result(_gvn.transform(result_val));
5141 return true;
5142 }
5143
5144 // If we have a tightly coupled allocation, the arraycopy may take care
5145 // of the array initialization. If one of the guards we insert between
5146 // the allocation and the arraycopy causes a deoptimization, an
5147 // uninitialized array will escape the compiled method. To prevent that
5148 // we set the JVM state for uncommon traps between the allocation and
5149 // the arraycopy to the state before the allocation so, in case of
5150 // deoptimization, we'll reexecute the allocation and the
5151 // initialization.
5152 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5153 if (alloc != nullptr) {
5154 ciMethod* trap_method = alloc->jvms()->method();
5155 int trap_bci = alloc->jvms()->bci();
5156
5157 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5158 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5159 // Make sure there's no store between the allocation and the
5160 // arraycopy otherwise visible side effects could be rexecuted
5161 // in case of deoptimization and cause incorrect execution.
5162 bool no_interfering_store = true;
5163 Node* mem = alloc->in(TypeFunc::Memory);
5164 if (mem->is_MergeMem()) {
5165 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5166 Node* n = mms.memory();
5167 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5168 assert(n->is_Store(), "what else?");
5169 no_interfering_store = false;
5170 break;
5171 }
5172 }
5173 } else {
5174 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5175 Node* n = mms.memory();
5176 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5177 assert(n->is_Store(), "what else?");
5178 no_interfering_store = false;
5179 break;
5180 }
5181 }
5182 }
5183
5184 if (no_interfering_store) {
5185 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5186
5187 JVMState* saved_jvms = jvms();
5188 saved_reexecute_sp = _reexecute_sp;
5189
5190 set_jvms(sfpt->jvms());
5191 _reexecute_sp = jvms()->sp();
5192
5193 return saved_jvms;
5194 }
5195 }
5196 }
5197 return nullptr;
5198 }
5199
5200 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5201 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5202 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5203 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5204 uint size = alloc->req();
5205 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5206 old_jvms->set_map(sfpt);
5207 for (uint i = 0; i < size; i++) {
5208 sfpt->init_req(i, alloc->in(i));
5209 }
5210 // re-push array length for deoptimization
5211 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5212 old_jvms->set_sp(old_jvms->sp()+1);
5213 old_jvms->set_monoff(old_jvms->monoff()+1);
5214 old_jvms->set_scloff(old_jvms->scloff()+1);
5215 old_jvms->set_endoff(old_jvms->endoff()+1);
5216 old_jvms->set_should_reexecute(true);
5217
5218 sfpt->set_i_o(map()->i_o());
5219 sfpt->set_memory(map()->memory());
5220 sfpt->set_control(map()->control());
5221 return sfpt;
5222 }
5223
5224 // In case of a deoptimization, we restart execution at the
5225 // allocation, allocating a new array. We would leave an uninitialized
5226 // array in the heap that GCs wouldn't expect. Move the allocation
5227 // after the traps so we don't allocate the array if we
5228 // deoptimize. This is possible because tightly_coupled_allocation()
5229 // guarantees there's no observer of the allocated array at this point
5230 // and the control flow is simple enough.
5231 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5232 int saved_reexecute_sp, uint new_idx) {
5233 if (saved_jvms_before_guards != nullptr && !stopped()) {
5234 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5235
5236 assert(alloc != nullptr, "only with a tightly coupled allocation");
5237 // restore JVM state to the state at the arraycopy
5238 saved_jvms_before_guards->map()->set_control(map()->control());
5239 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5240 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5241 // If we've improved the types of some nodes (null check) while
5242 // emitting the guards, propagate them to the current state
5243 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5244 set_jvms(saved_jvms_before_guards);
5245 _reexecute_sp = saved_reexecute_sp;
5246
5247 // Remove the allocation from above the guards
5248 CallProjections callprojs;
5249 alloc->extract_projections(&callprojs, true);
5250 InitializeNode* init = alloc->initialization();
5251 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5252 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5253 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5254
5255 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5256 // the allocation (i.e. is only valid if the allocation succeeds):
5257 // 1) replace CastIINode with AllocateArrayNode's length here
5258 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5259 //
5260 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5261 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5262 Node* init_control = init->proj_out(TypeFunc::Control);
5263 Node* alloc_length = alloc->Ideal_length();
5264 #ifdef ASSERT
5265 Node* prev_cast = nullptr;
5266 #endif
5267 for (uint i = 0; i < init_control->outcnt(); i++) {
5268 Node* init_out = init_control->raw_out(i);
5269 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5270 #ifdef ASSERT
5271 if (prev_cast == nullptr) {
5272 prev_cast = init_out;
5273 } else {
5274 if (prev_cast->cmp(*init_out) == false) {
5275 prev_cast->dump();
5276 init_out->dump();
5277 assert(false, "not equal CastIINode");
5278 }
5279 }
5280 #endif
5281 C->gvn_replace_by(init_out, alloc_length);
5282 }
5283 }
5284 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5285
5286 // move the allocation here (after the guards)
5287 _gvn.hash_delete(alloc);
5288 alloc->set_req(TypeFunc::Control, control());
5289 alloc->set_req(TypeFunc::I_O, i_o());
5290 Node *mem = reset_memory();
5291 set_all_memory(mem);
5292 alloc->set_req(TypeFunc::Memory, mem);
5293 set_control(init->proj_out_or_null(TypeFunc::Control));
5294 set_i_o(callprojs.fallthrough_ioproj);
5295
5296 // Update memory as done in GraphKit::set_output_for_allocation()
5297 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5298 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5299 if (ary_type->isa_aryptr() && length_type != nullptr) {
5300 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5301 }
5302 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5303 int elemidx = C->get_alias_index(telemref);
5304 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5305 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5306
5307 Node* allocx = _gvn.transform(alloc);
5308 assert(allocx == alloc, "where has the allocation gone?");
5309 assert(dest->is_CheckCastPP(), "not an allocation result?");
5310
5311 _gvn.hash_delete(dest);
5312 dest->set_req(0, control());
5313 Node* destx = _gvn.transform(dest);
5314 assert(destx == dest, "where has the allocation result gone?");
5315
5316 array_ideal_length(alloc, ary_type, true);
5317 }
5318 }
5319
5320 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5321 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5322 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5323 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5324 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5325 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5326 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5327 JVMState* saved_jvms_before_guards) {
5328 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5329 // There is at least one unrelated uncommon trap which needs to be replaced.
5330 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5331
5332 JVMState* saved_jvms = jvms();
5333 const int saved_reexecute_sp = _reexecute_sp;
5334 set_jvms(sfpt->jvms());
5335 _reexecute_sp = jvms()->sp();
5336
5337 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5338
5339 // Restore state
5340 set_jvms(saved_jvms);
5341 _reexecute_sp = saved_reexecute_sp;
5342 }
5343 }
5344
5345 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5346 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5347 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5348 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5349 while (if_proj->is_IfProj()) {
5350 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5351 if (uncommon_trap != nullptr) {
5352 create_new_uncommon_trap(uncommon_trap);
5353 }
5354 assert(if_proj->in(0)->is_If(), "must be If");
5355 if_proj = if_proj->in(0)->in(0);
5356 }
5357 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5358 "must have reached control projection of init node");
5359 }
5360
5361 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5362 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5363 assert(trap_request != 0, "no valid UCT trap request");
5364 PreserveJVMState pjvms(this);
5365 set_control(uncommon_trap_call->in(0));
5366 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5367 Deoptimization::trap_request_action(trap_request));
5368 assert(stopped(), "Should be stopped");
5369 _gvn.hash_delete(uncommon_trap_call);
5370 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5371 }
5372
5373 //------------------------------inline_arraycopy-----------------------
5374 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5375 // Object dest, int destPos,
5376 // int length);
5377 bool LibraryCallKit::inline_arraycopy() {
5378 // Get the arguments.
5379 Node* src = argument(0); // type: oop
5380 Node* src_offset = argument(1); // type: int
5381 Node* dest = argument(2); // type: oop
5382 Node* dest_offset = argument(3); // type: int
5383 Node* length = argument(4); // type: int
5384
5385 uint new_idx = C->unique();
5386
5387 // Check for allocation before we add nodes that would confuse
5388 // tightly_coupled_allocation()
5389 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5390
5391 int saved_reexecute_sp = -1;
5392 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5393 // See arraycopy_restore_alloc_state() comment
5394 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5395 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5396 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5397 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5398
5399 // The following tests must be performed
5400 // (1) src and dest are arrays.
5401 // (2) src and dest arrays must have elements of the same BasicType
5402 // (3) src and dest must not be null.
5403 // (4) src_offset must not be negative.
5404 // (5) dest_offset must not be negative.
5405 // (6) length must not be negative.
5406 // (7) src_offset + length must not exceed length of src.
5407 // (8) dest_offset + length must not exceed length of dest.
5408 // (9) each element of an oop array must be assignable
5409
5410 // (3) src and dest must not be null.
5411 // always do this here because we need the JVM state for uncommon traps
5412 Node* null_ctl = top();
5413 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5414 assert(null_ctl->is_top(), "no null control here");
5415 dest = null_check(dest, T_ARRAY);
5416
5417 if (!can_emit_guards) {
5418 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5419 // guards but the arraycopy node could still take advantage of a
5420 // tightly allocated allocation. tightly_coupled_allocation() is
5421 // called again to make sure it takes the null check above into
5422 // account: the null check is mandatory and if it caused an
5423 // uncommon trap to be emitted then the allocation can't be
5424 // considered tightly coupled in this context.
5425 alloc = tightly_coupled_allocation(dest);
5426 }
5427
5428 bool validated = false;
5429
5430 const Type* src_type = _gvn.type(src);
5431 const Type* dest_type = _gvn.type(dest);
5432 const TypeAryPtr* top_src = src_type->isa_aryptr();
5433 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5434
5435 // Do we have the type of src?
5436 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5437 // Do we have the type of dest?
5438 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5439 // Is the type for src from speculation?
5440 bool src_spec = false;
5441 // Is the type for dest from speculation?
5442 bool dest_spec = false;
5443
5444 if ((!has_src || !has_dest) && can_emit_guards) {
5445 // We don't have sufficient type information, let's see if
5446 // speculative types can help. We need to have types for both src
5447 // and dest so that it pays off.
5448
5449 // Do we already have or could we have type information for src
5450 bool could_have_src = has_src;
5451 // Do we already have or could we have type information for dest
5452 bool could_have_dest = has_dest;
5453
5454 ciKlass* src_k = nullptr;
5455 if (!has_src) {
5456 src_k = src_type->speculative_type_not_null();
5457 if (src_k != nullptr && src_k->is_array_klass()) {
5458 could_have_src = true;
5459 }
5460 }
5461
5462 ciKlass* dest_k = nullptr;
5463 if (!has_dest) {
5464 dest_k = dest_type->speculative_type_not_null();
5465 if (dest_k != nullptr && dest_k->is_array_klass()) {
5466 could_have_dest = true;
5467 }
5468 }
5469
5470 if (could_have_src && could_have_dest) {
5471 // This is going to pay off so emit the required guards
5472 if (!has_src) {
5473 src = maybe_cast_profiled_obj(src, src_k, true);
5474 src_type = _gvn.type(src);
5475 top_src = src_type->isa_aryptr();
5476 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5477 src_spec = true;
5478 }
5479 if (!has_dest) {
5480 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5481 dest_type = _gvn.type(dest);
5482 top_dest = dest_type->isa_aryptr();
5483 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5484 dest_spec = true;
5485 }
5486 }
5487 }
5488
5489 if (has_src && has_dest && can_emit_guards) {
5490 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5491 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5492 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5493 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5494
5495 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5496 // If both arrays are object arrays then having the exact types
5497 // for both will remove the need for a subtype check at runtime
5498 // before the call and may make it possible to pick a faster copy
5499 // routine (without a subtype check on every element)
5500 // Do we have the exact type of src?
5501 bool could_have_src = src_spec;
5502 // Do we have the exact type of dest?
5503 bool could_have_dest = dest_spec;
5504 ciKlass* src_k = nullptr;
5505 ciKlass* dest_k = nullptr;
5506 if (!src_spec) {
5507 src_k = src_type->speculative_type_not_null();
5508 if (src_k != nullptr && src_k->is_array_klass()) {
5509 could_have_src = true;
5510 }
5511 }
5512 if (!dest_spec) {
5513 dest_k = dest_type->speculative_type_not_null();
5514 if (dest_k != nullptr && dest_k->is_array_klass()) {
5515 could_have_dest = true;
5516 }
5517 }
5518 if (could_have_src && could_have_dest) {
5519 // If we can have both exact types, emit the missing guards
5520 if (could_have_src && !src_spec) {
5521 src = maybe_cast_profiled_obj(src, src_k, true);
5522 }
5523 if (could_have_dest && !dest_spec) {
5524 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5525 }
5526 }
5527 }
5528 }
5529
5530 ciMethod* trap_method = method();
5531 int trap_bci = bci();
5532 if (saved_jvms_before_guards != nullptr) {
5533 trap_method = alloc->jvms()->method();
5534 trap_bci = alloc->jvms()->bci();
5535 }
5536
5537 bool negative_length_guard_generated = false;
5538
5539 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5540 can_emit_guards &&
5541 !src->is_top() && !dest->is_top()) {
5542 // validate arguments: enables transformation the ArrayCopyNode
5543 validated = true;
5544
5545 RegionNode* slow_region = new RegionNode(1);
5546 record_for_igvn(slow_region);
5547
5548 // (1) src and dest are arrays.
5549 generate_non_array_guard(load_object_klass(src), slow_region);
5550 generate_non_array_guard(load_object_klass(dest), slow_region);
5551
5552 // (2) src and dest arrays must have elements of the same BasicType
5553 // done at macro expansion or at Ideal transformation time
5554
5555 // (4) src_offset must not be negative.
5556 generate_negative_guard(src_offset, slow_region);
5557
5558 // (5) dest_offset must not be negative.
5559 generate_negative_guard(dest_offset, slow_region);
5560
5561 // (7) src_offset + length must not exceed length of src.
5562 generate_limit_guard(src_offset, length,
5563 load_array_length(src),
5564 slow_region);
5565
5566 // (8) dest_offset + length must not exceed length of dest.
5567 generate_limit_guard(dest_offset, length,
5568 load_array_length(dest),
5569 slow_region);
5570
5571 // (6) length must not be negative.
5572 // This is also checked in generate_arraycopy() during macro expansion, but
5573 // we also have to check it here for the case where the ArrayCopyNode will
5574 // be eliminated by Escape Analysis.
5575 if (EliminateAllocations) {
5576 generate_negative_guard(length, slow_region);
5577 negative_length_guard_generated = true;
5578 }
5579
5580 // (9) each element of an oop array must be assignable
5581 Node* dest_klass = load_object_klass(dest);
5582 if (src != dest) {
5583 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5584
5585 if (not_subtype_ctrl != top()) {
5586 PreserveJVMState pjvms(this);
5587 set_control(not_subtype_ctrl);
5588 uncommon_trap(Deoptimization::Reason_intrinsic,
5589 Deoptimization::Action_make_not_entrant);
5590 assert(stopped(), "Should be stopped");
5591 }
5592 }
5593 {
5594 PreserveJVMState pjvms(this);
5595 set_control(_gvn.transform(slow_region));
5596 uncommon_trap(Deoptimization::Reason_intrinsic,
5597 Deoptimization::Action_make_not_entrant);
5598 assert(stopped(), "Should be stopped");
5599 }
5600
5601 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5602 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5603 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5604 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5605 }
5606
5607 if (stopped()) {
5608 return true;
5609 }
5610
5611 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5612 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5613 // so the compiler has a chance to eliminate them: during macro expansion,
5614 // we have to set their control (CastPP nodes are eliminated).
5615 load_object_klass(src), load_object_klass(dest),
5616 load_array_length(src), load_array_length(dest));
5617
5618 ac->set_arraycopy(validated);
5619
5620 Node* n = _gvn.transform(ac);
5621 if (n == ac) {
5622 ac->connect_outputs(this);
5623 } else {
5624 assert(validated, "shouldn't transform if all arguments not validated");
5625 set_all_memory(n);
5626 }
5627 clear_upper_avx();
5628
5629
5630 return true;
5631 }
5632
5633
5634 // Helper function which determines if an arraycopy immediately follows
5635 // an allocation, with no intervening tests or other escapes for the object.
5636 AllocateArrayNode*
5637 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5638 if (stopped()) return nullptr; // no fast path
5639 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5640
5641 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
5642 if (alloc == nullptr) return nullptr;
5643
5644 Node* rawmem = memory(Compile::AliasIdxRaw);
5645 // Is the allocation's memory state untouched?
5646 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5647 // Bail out if there have been raw-memory effects since the allocation.
5648 // (Example: There might have been a call or safepoint.)
5649 return nullptr;
5650 }
5651 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5652 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5653 return nullptr;
5654 }
5655
5656 // There must be no unexpected observers of this allocation.
5657 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5658 Node* obs = ptr->fast_out(i);
5659 if (obs != this->map()) {
5660 return nullptr;
5661 }
5662 }
5663
5664 // This arraycopy must unconditionally follow the allocation of the ptr.
5665 Node* alloc_ctl = ptr->in(0);
5666 Node* ctl = control();
5667 while (ctl != alloc_ctl) {
5668 // There may be guards which feed into the slow_region.
5669 // Any other control flow means that we might not get a chance
5670 // to finish initializing the allocated object.
5671 // Various low-level checks bottom out in uncommon traps. These
5672 // are considered safe since we've already checked above that
5673 // there is no unexpected observer of this allocation.
5674 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
5675 assert(ctl->in(0)->is_If(), "must be If");
5676 ctl = ctl->in(0)->in(0);
5677 } else {
5678 return nullptr;
5679 }
5680 }
5681
5682 // If we get this far, we have an allocation which immediately
5683 // precedes the arraycopy, and we can take over zeroing the new object.
5684 // The arraycopy will finish the initialization, and provide
5685 // a new control state to which we will anchor the destination pointer.
5686
5687 return alloc;
5688 }
5689
5690 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
5691 if (node->is_IfProj()) {
5692 Node* other_proj = node->as_IfProj()->other_if_proj();
5693 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
5694 Node* obs = other_proj->fast_out(j);
5695 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
5696 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5697 return obs->as_CallStaticJava();
5698 }
5699 }
5700 }
5701 return nullptr;
5702 }
5703
5704 //-------------inline_encodeISOArray-----------------------------------
5705 // encode char[] to byte[] in ISO_8859_1 or ASCII
5706 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5707 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5708 // no receiver since it is static method
5709 Node *src = argument(0);
5710 Node *src_offset = argument(1);
5711 Node *dst = argument(2);
5712 Node *dst_offset = argument(3);
5713 Node *length = argument(4);
5714
5715 src = must_be_not_null(src, true);
5716 dst = must_be_not_null(dst, true);
5717
5718 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
5719 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
5720 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
5721 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
5722 // failed array check
5723 return false;
5724 }
5725
5726 // Figure out the size and type of the elements we will be copying.
5727 BasicType src_elem = src_type->elem()->array_element_basic_type();
5728 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
5729 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5730 return false;
5731 }
5732
5733 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5734 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5735 // 'src_start' points to src array + scaled offset
5736 // 'dst_start' points to dst array + scaled offset
5737
5738 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5739 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5740 enc = _gvn.transform(enc);
5741 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5742 set_memory(res_mem, mtype);
5743 set_result(enc);
5744 clear_upper_avx();
5745
5746 return true;
5747 }
5748
5749 //-------------inline_multiplyToLen-----------------------------------
5750 bool LibraryCallKit::inline_multiplyToLen() {
5751 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5752
5753 address stubAddr = StubRoutines::multiplyToLen();
5754 if (stubAddr == nullptr) {
5755 return false; // Intrinsic's stub is not implemented on this platform
5756 }
5757 const char* stubName = "multiplyToLen";
5758
5759 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5760
5761 // no receiver because it is a static method
5762 Node* x = argument(0);
5763 Node* xlen = argument(1);
5764 Node* y = argument(2);
5765 Node* ylen = argument(3);
5766 Node* z = argument(4);
5767
5768 x = must_be_not_null(x, true);
5769 y = must_be_not_null(y, true);
5770
5771 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5772 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
5773 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5774 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
5775 // failed array check
5776 return false;
5777 }
5778
5779 BasicType x_elem = x_type->elem()->array_element_basic_type();
5780 BasicType y_elem = y_type->elem()->array_element_basic_type();
5781 if (x_elem != T_INT || y_elem != T_INT) {
5782 return false;
5783 }
5784
5785 // Set the original stack and the reexecute bit for the interpreter to reexecute
5786 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5787 // on the return from z array allocation in runtime.
5788 { PreserveReexecuteState preexecs(this);
5789 jvms()->set_should_reexecute(true);
5790
5791 Node* x_start = array_element_address(x, intcon(0), x_elem);
5792 Node* y_start = array_element_address(y, intcon(0), y_elem);
5793 // 'x_start' points to x array + scaled xlen
5794 // 'y_start' points to y array + scaled ylen
5795
5796 // Allocate the result array
5797 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5798 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5799 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5800
5801 IdealKit ideal(this);
5802
5803 #define __ ideal.
5804 Node* one = __ ConI(1);
5805 Node* zero = __ ConI(0);
5806 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5807 __ set(need_alloc, zero);
5808 __ set(z_alloc, z);
5809 __ if_then(z, BoolTest::eq, null()); {
5810 __ increment (need_alloc, one);
5811 } __ else_(); {
5812 // Update graphKit memory and control from IdealKit.
5813 sync_kit(ideal);
5814 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5815 cast->init_req(0, control());
5816 _gvn.set_type(cast, cast->bottom_type());
5817 C->record_for_igvn(cast);
5818
5819 Node* zlen_arg = load_array_length(cast);
5820 // Update IdealKit memory and control from graphKit.
5821 __ sync_kit(this);
5822 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5823 __ increment (need_alloc, one);
5824 } __ end_if();
5825 } __ end_if();
5826
5827 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5828 // Update graphKit memory and control from IdealKit.
5829 sync_kit(ideal);
5830 Node * narr = new_array(klass_node, zlen, 1);
5831 // Update IdealKit memory and control from graphKit.
5832 __ sync_kit(this);
5833 __ set(z_alloc, narr);
5834 } __ end_if();
5835
5836 sync_kit(ideal);
5837 z = __ value(z_alloc);
5838 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5839 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5840 // Final sync IdealKit and GraphKit.
5841 final_sync(ideal);
5842 #undef __
5843
5844 Node* z_start = array_element_address(z, intcon(0), T_INT);
5845
5846 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5847 OptoRuntime::multiplyToLen_Type(),
5848 stubAddr, stubName, TypePtr::BOTTOM,
5849 x_start, xlen, y_start, ylen, z_start, zlen);
5850 } // original reexecute is set back here
5851
5852 C->set_has_split_ifs(true); // Has chance for split-if optimization
5853 set_result(z);
5854 return true;
5855 }
5856
5857 //-------------inline_squareToLen------------------------------------
5858 bool LibraryCallKit::inline_squareToLen() {
5859 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5860
5861 address stubAddr = StubRoutines::squareToLen();
5862 if (stubAddr == nullptr) {
5863 return false; // Intrinsic's stub is not implemented on this platform
5864 }
5865 const char* stubName = "squareToLen";
5866
5867 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5868
5869 Node* x = argument(0);
5870 Node* len = argument(1);
5871 Node* z = argument(2);
5872 Node* zlen = argument(3);
5873
5874 x = must_be_not_null(x, true);
5875 z = must_be_not_null(z, true);
5876
5877 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5878 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
5879 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5880 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
5881 // failed array check
5882 return false;
5883 }
5884
5885 BasicType x_elem = x_type->elem()->array_element_basic_type();
5886 BasicType z_elem = z_type->elem()->array_element_basic_type();
5887 if (x_elem != T_INT || z_elem != T_INT) {
5888 return false;
5889 }
5890
5891
5892 Node* x_start = array_element_address(x, intcon(0), x_elem);
5893 Node* z_start = array_element_address(z, intcon(0), z_elem);
5894
5895 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5896 OptoRuntime::squareToLen_Type(),
5897 stubAddr, stubName, TypePtr::BOTTOM,
5898 x_start, len, z_start, zlen);
5899
5900 set_result(z);
5901 return true;
5902 }
5903
5904 //-------------inline_mulAdd------------------------------------------
5905 bool LibraryCallKit::inline_mulAdd() {
5906 assert(UseMulAddIntrinsic, "not implemented on this platform");
5907
5908 address stubAddr = StubRoutines::mulAdd();
5909 if (stubAddr == nullptr) {
5910 return false; // Intrinsic's stub is not implemented on this platform
5911 }
5912 const char* stubName = "mulAdd";
5913
5914 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5915
5916 Node* out = argument(0);
5917 Node* in = argument(1);
5918 Node* offset = argument(2);
5919 Node* len = argument(3);
5920 Node* k = argument(4);
5921
5922 in = must_be_not_null(in, true);
5923 out = must_be_not_null(out, true);
5924
5925 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
5926 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
5927 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
5928 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
5929 // failed array check
5930 return false;
5931 }
5932
5933 BasicType out_elem = out_type->elem()->array_element_basic_type();
5934 BasicType in_elem = in_type->elem()->array_element_basic_type();
5935 if (out_elem != T_INT || in_elem != T_INT) {
5936 return false;
5937 }
5938
5939 Node* outlen = load_array_length(out);
5940 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5941 Node* out_start = array_element_address(out, intcon(0), out_elem);
5942 Node* in_start = array_element_address(in, intcon(0), in_elem);
5943
5944 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5945 OptoRuntime::mulAdd_Type(),
5946 stubAddr, stubName, TypePtr::BOTTOM,
5947 out_start,in_start, new_offset, len, k);
5948 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5949 set_result(result);
5950 return true;
5951 }
5952
5953 //-------------inline_montgomeryMultiply-----------------------------------
5954 bool LibraryCallKit::inline_montgomeryMultiply() {
5955 address stubAddr = StubRoutines::montgomeryMultiply();
5956 if (stubAddr == nullptr) {
5957 return false; // Intrinsic's stub is not implemented on this platform
5958 }
5959
5960 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5961 const char* stubName = "montgomery_multiply";
5962
5963 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5964
5965 Node* a = argument(0);
5966 Node* b = argument(1);
5967 Node* n = argument(2);
5968 Node* len = argument(3);
5969 Node* inv = argument(4);
5970 Node* m = argument(6);
5971
5972 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
5973 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
5974 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
5975 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
5976 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
5977 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
5978 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
5979 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
5980 // failed array check
5981 return false;
5982 }
5983
5984 BasicType a_elem = a_type->elem()->array_element_basic_type();
5985 BasicType b_elem = b_type->elem()->array_element_basic_type();
5986 BasicType n_elem = n_type->elem()->array_element_basic_type();
5987 BasicType m_elem = m_type->elem()->array_element_basic_type();
5988 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5989 return false;
5990 }
5991
5992 // Make the call
5993 {
5994 Node* a_start = array_element_address(a, intcon(0), a_elem);
5995 Node* b_start = array_element_address(b, intcon(0), b_elem);
5996 Node* n_start = array_element_address(n, intcon(0), n_elem);
5997 Node* m_start = array_element_address(m, intcon(0), m_elem);
5998
5999 Node* call = make_runtime_call(RC_LEAF,
6000 OptoRuntime::montgomeryMultiply_Type(),
6001 stubAddr, stubName, TypePtr::BOTTOM,
6002 a_start, b_start, n_start, len, inv, top(),
6003 m_start);
6004 set_result(m);
6005 }
6006
6007 return true;
6008 }
6009
6010 bool LibraryCallKit::inline_montgomerySquare() {
6011 address stubAddr = StubRoutines::montgomerySquare();
6012 if (stubAddr == nullptr) {
6013 return false; // Intrinsic's stub is not implemented on this platform
6014 }
6015
6016 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6017 const char* stubName = "montgomery_square";
6018
6019 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6020
6021 Node* a = argument(0);
6022 Node* n = argument(1);
6023 Node* len = argument(2);
6024 Node* inv = argument(3);
6025 Node* m = argument(5);
6026
6027 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6028 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6029 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6030 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6031 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6032 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6033 // failed array check
6034 return false;
6035 }
6036
6037 BasicType a_elem = a_type->elem()->array_element_basic_type();
6038 BasicType n_elem = n_type->elem()->array_element_basic_type();
6039 BasicType m_elem = m_type->elem()->array_element_basic_type();
6040 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6041 return false;
6042 }
6043
6044 // Make the call
6045 {
6046 Node* a_start = array_element_address(a, intcon(0), a_elem);
6047 Node* n_start = array_element_address(n, intcon(0), n_elem);
6048 Node* m_start = array_element_address(m, intcon(0), m_elem);
6049
6050 Node* call = make_runtime_call(RC_LEAF,
6051 OptoRuntime::montgomerySquare_Type(),
6052 stubAddr, stubName, TypePtr::BOTTOM,
6053 a_start, n_start, len, inv, top(),
6054 m_start);
6055 set_result(m);
6056 }
6057
6058 return true;
6059 }
6060
6061 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6062 address stubAddr = nullptr;
6063 const char* stubName = nullptr;
6064
6065 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6066 if (stubAddr == nullptr) {
6067 return false; // Intrinsic's stub is not implemented on this platform
6068 }
6069
6070 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6071
6072 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6073
6074 Node* newArr = argument(0);
6075 Node* oldArr = argument(1);
6076 Node* newIdx = argument(2);
6077 Node* shiftCount = argument(3);
6078 Node* numIter = argument(4);
6079
6080 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6081 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6082 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6083 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6084 return false;
6085 }
6086
6087 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6088 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6089 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6090 return false;
6091 }
6092
6093 // Make the call
6094 {
6095 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6096 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6097
6098 Node* call = make_runtime_call(RC_LEAF,
6099 OptoRuntime::bigIntegerShift_Type(),
6100 stubAddr,
6101 stubName,
6102 TypePtr::BOTTOM,
6103 newArr_start,
6104 oldArr_start,
6105 newIdx,
6106 shiftCount,
6107 numIter);
6108 }
6109
6110 return true;
6111 }
6112
6113 //-------------inline_vectorizedMismatch------------------------------
6114 bool LibraryCallKit::inline_vectorizedMismatch() {
6115 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6116
6117 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6118 Node* obja = argument(0); // Object
6119 Node* aoffset = argument(1); // long
6120 Node* objb = argument(3); // Object
6121 Node* boffset = argument(4); // long
6122 Node* length = argument(6); // int
6123 Node* scale = argument(7); // int
6124
6125 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6126 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6127 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6128 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6129 scale == top()) {
6130 return false; // failed input validation
6131 }
6132
6133 Node* obja_adr = make_unsafe_address(obja, aoffset);
6134 Node* objb_adr = make_unsafe_address(objb, boffset);
6135
6136 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6137 //
6138 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6139 // if (length <= inline_limit) {
6140 // inline_path:
6141 // vmask = VectorMaskGen length
6142 // vload1 = LoadVectorMasked obja, vmask
6143 // vload2 = LoadVectorMasked objb, vmask
6144 // result1 = VectorCmpMasked vload1, vload2, vmask
6145 // } else {
6146 // call_stub_path:
6147 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6148 // }
6149 // exit_block:
6150 // return Phi(result1, result2);
6151 //
6152 enum { inline_path = 1, // input is small enough to process it all at once
6153 stub_path = 2, // input is too large; call into the VM
6154 PATH_LIMIT = 3
6155 };
6156
6157 Node* exit_block = new RegionNode(PATH_LIMIT);
6158 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6159 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6160
6161 Node* call_stub_path = control();
6162
6163 BasicType elem_bt = T_ILLEGAL;
6164
6165 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6166 if (scale_t->is_con()) {
6167 switch (scale_t->get_con()) {
6168 case 0: elem_bt = T_BYTE; break;
6169 case 1: elem_bt = T_SHORT; break;
6170 case 2: elem_bt = T_INT; break;
6171 case 3: elem_bt = T_LONG; break;
6172
6173 default: elem_bt = T_ILLEGAL; break; // not supported
6174 }
6175 }
6176
6177 int inline_limit = 0;
6178 bool do_partial_inline = false;
6179
6180 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6181 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6182 do_partial_inline = inline_limit >= 16;
6183 }
6184
6185 if (do_partial_inline) {
6186 assert(elem_bt != T_ILLEGAL, "sanity");
6187
6188 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6189 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6190 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6191
6192 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6193 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6194 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6195
6196 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6197
6198 if (!stopped()) {
6199 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6200
6201 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6202 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6203 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6204 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6205
6206 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6207 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6208 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6209 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6210
6211 exit_block->init_req(inline_path, control());
6212 memory_phi->init_req(inline_path, map()->memory());
6213 result_phi->init_req(inline_path, result);
6214
6215 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6216 clear_upper_avx();
6217 }
6218 }
6219 }
6220
6221 if (call_stub_path != nullptr) {
6222 set_control(call_stub_path);
6223
6224 Node* call = make_runtime_call(RC_LEAF,
6225 OptoRuntime::vectorizedMismatch_Type(),
6226 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6227 obja_adr, objb_adr, length, scale);
6228
6229 exit_block->init_req(stub_path, control());
6230 memory_phi->init_req(stub_path, map()->memory());
6231 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6232 }
6233
6234 exit_block = _gvn.transform(exit_block);
6235 memory_phi = _gvn.transform(memory_phi);
6236 result_phi = _gvn.transform(result_phi);
6237
6238 set_control(exit_block);
6239 set_all_memory(memory_phi);
6240 set_result(result_phi);
6241
6242 return true;
6243 }
6244
6245 //------------------------------inline_vectorizedHashcode----------------------------
6246 bool LibraryCallKit::inline_vectorizedHashCode() {
6247 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6248
6249 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6250 Node* array = argument(0);
6251 Node* offset = argument(1);
6252 Node* length = argument(2);
6253 Node* initialValue = argument(3);
6254 Node* basic_type = argument(4);
6255
6256 if (basic_type == top()) {
6257 return false; // failed input validation
6258 }
6259
6260 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6261 if (!basic_type_t->is_con()) {
6262 return false; // Only intrinsify if mode argument is constant
6263 }
6264
6265 array = must_be_not_null(array, true);
6266
6267 BasicType bt = (BasicType)basic_type_t->get_con();
6268
6269 // Resolve address of first element
6270 Node* array_start = array_element_address(array, offset, bt);
6271
6272 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6273 array_start, length, initialValue, basic_type)));
6274 clear_upper_avx();
6275
6276 return true;
6277 }
6278
6279 /**
6280 * Calculate CRC32 for byte.
6281 * int java.util.zip.CRC32.update(int crc, int b)
6282 */
6283 bool LibraryCallKit::inline_updateCRC32() {
6284 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6285 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6286 // no receiver since it is static method
6287 Node* crc = argument(0); // type: int
6288 Node* b = argument(1); // type: int
6289
6290 /*
6291 * int c = ~ crc;
6292 * b = timesXtoThe32[(b ^ c) & 0xFF];
6293 * b = b ^ (c >>> 8);
6294 * crc = ~b;
6295 */
6296
6297 Node* M1 = intcon(-1);
6298 crc = _gvn.transform(new XorINode(crc, M1));
6299 Node* result = _gvn.transform(new XorINode(crc, b));
6300 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6301
6302 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6303 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6304 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6305 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6306
6307 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6308 result = _gvn.transform(new XorINode(crc, result));
6309 result = _gvn.transform(new XorINode(result, M1));
6310 set_result(result);
6311 return true;
6312 }
6313
6314 /**
6315 * Calculate CRC32 for byte[] array.
6316 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6317 */
6318 bool LibraryCallKit::inline_updateBytesCRC32() {
6319 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6320 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6321 // no receiver since it is static method
6322 Node* crc = argument(0); // type: int
6323 Node* src = argument(1); // type: oop
6324 Node* offset = argument(2); // type: int
6325 Node* length = argument(3); // type: int
6326
6327 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6328 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6329 // failed array check
6330 return false;
6331 }
6332
6333 // Figure out the size and type of the elements we will be copying.
6334 BasicType src_elem = src_type->elem()->array_element_basic_type();
6335 if (src_elem != T_BYTE) {
6336 return false;
6337 }
6338
6339 // 'src_start' points to src array + scaled offset
6340 src = must_be_not_null(src, true);
6341 Node* src_start = array_element_address(src, offset, src_elem);
6342
6343 // We assume that range check is done by caller.
6344 // TODO: generate range check (offset+length < src.length) in debug VM.
6345
6346 // Call the stub.
6347 address stubAddr = StubRoutines::updateBytesCRC32();
6348 const char *stubName = "updateBytesCRC32";
6349
6350 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6351 stubAddr, stubName, TypePtr::BOTTOM,
6352 crc, src_start, length);
6353 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6354 set_result(result);
6355 return true;
6356 }
6357
6358 /**
6359 * Calculate CRC32 for ByteBuffer.
6360 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6361 */
6362 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6363 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6364 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6365 // no receiver since it is static method
6366 Node* crc = argument(0); // type: int
6367 Node* src = argument(1); // type: long
6368 Node* offset = argument(3); // type: int
6369 Node* length = argument(4); // type: int
6370
6371 src = ConvL2X(src); // adjust Java long to machine word
6372 Node* base = _gvn.transform(new CastX2PNode(src));
6373 offset = ConvI2X(offset);
6374
6375 // 'src_start' points to src array + scaled offset
6376 Node* src_start = basic_plus_adr(top(), base, offset);
6377
6378 // Call the stub.
6379 address stubAddr = StubRoutines::updateBytesCRC32();
6380 const char *stubName = "updateBytesCRC32";
6381
6382 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6383 stubAddr, stubName, TypePtr::BOTTOM,
6384 crc, src_start, length);
6385 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6386 set_result(result);
6387 return true;
6388 }
6389
6390 //------------------------------get_table_from_crc32c_class-----------------------
6391 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6392 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6393 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6394
6395 return table;
6396 }
6397
6398 //------------------------------inline_updateBytesCRC32C-----------------------
6399 //
6400 // Calculate CRC32C for byte[] array.
6401 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6402 //
6403 bool LibraryCallKit::inline_updateBytesCRC32C() {
6404 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6405 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6406 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6407 // no receiver since it is a static method
6408 Node* crc = argument(0); // type: int
6409 Node* src = argument(1); // type: oop
6410 Node* offset = argument(2); // type: int
6411 Node* end = argument(3); // type: int
6412
6413 Node* length = _gvn.transform(new SubINode(end, offset));
6414
6415 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6416 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6417 // failed array check
6418 return false;
6419 }
6420
6421 // Figure out the size and type of the elements we will be copying.
6422 BasicType src_elem = src_type->elem()->array_element_basic_type();
6423 if (src_elem != T_BYTE) {
6424 return false;
6425 }
6426
6427 // 'src_start' points to src array + scaled offset
6428 src = must_be_not_null(src, true);
6429 Node* src_start = array_element_address(src, offset, src_elem);
6430
6431 // static final int[] byteTable in class CRC32C
6432 Node* table = get_table_from_crc32c_class(callee()->holder());
6433 table = must_be_not_null(table, true);
6434 Node* table_start = array_element_address(table, intcon(0), T_INT);
6435
6436 // We assume that range check is done by caller.
6437 // TODO: generate range check (offset+length < src.length) in debug VM.
6438
6439 // Call the stub.
6440 address stubAddr = StubRoutines::updateBytesCRC32C();
6441 const char *stubName = "updateBytesCRC32C";
6442
6443 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6444 stubAddr, stubName, TypePtr::BOTTOM,
6445 crc, src_start, length, table_start);
6446 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6447 set_result(result);
6448 return true;
6449 }
6450
6451 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6452 //
6453 // Calculate CRC32C for DirectByteBuffer.
6454 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6455 //
6456 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6457 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6458 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6459 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6460 // no receiver since it is a static method
6461 Node* crc = argument(0); // type: int
6462 Node* src = argument(1); // type: long
6463 Node* offset = argument(3); // type: int
6464 Node* end = argument(4); // type: int
6465
6466 Node* length = _gvn.transform(new SubINode(end, offset));
6467
6468 src = ConvL2X(src); // adjust Java long to machine word
6469 Node* base = _gvn.transform(new CastX2PNode(src));
6470 offset = ConvI2X(offset);
6471
6472 // 'src_start' points to src array + scaled offset
6473 Node* src_start = basic_plus_adr(top(), base, offset);
6474
6475 // static final int[] byteTable in class CRC32C
6476 Node* table = get_table_from_crc32c_class(callee()->holder());
6477 table = must_be_not_null(table, true);
6478 Node* table_start = array_element_address(table, intcon(0), T_INT);
6479
6480 // Call the stub.
6481 address stubAddr = StubRoutines::updateBytesCRC32C();
6482 const char *stubName = "updateBytesCRC32C";
6483
6484 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6485 stubAddr, stubName, TypePtr::BOTTOM,
6486 crc, src_start, length, table_start);
6487 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6488 set_result(result);
6489 return true;
6490 }
6491
6492 //------------------------------inline_updateBytesAdler32----------------------
6493 //
6494 // Calculate Adler32 checksum for byte[] array.
6495 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6496 //
6497 bool LibraryCallKit::inline_updateBytesAdler32() {
6498 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6499 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6500 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6501 // no receiver since it is static method
6502 Node* crc = argument(0); // type: int
6503 Node* src = argument(1); // type: oop
6504 Node* offset = argument(2); // type: int
6505 Node* length = argument(3); // type: int
6506
6507 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6508 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6509 // failed array check
6510 return false;
6511 }
6512
6513 // Figure out the size and type of the elements we will be copying.
6514 BasicType src_elem = src_type->elem()->array_element_basic_type();
6515 if (src_elem != T_BYTE) {
6516 return false;
6517 }
6518
6519 // 'src_start' points to src array + scaled offset
6520 Node* src_start = array_element_address(src, offset, src_elem);
6521
6522 // We assume that range check is done by caller.
6523 // TODO: generate range check (offset+length < src.length) in debug VM.
6524
6525 // Call the stub.
6526 address stubAddr = StubRoutines::updateBytesAdler32();
6527 const char *stubName = "updateBytesAdler32";
6528
6529 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6530 stubAddr, stubName, TypePtr::BOTTOM,
6531 crc, src_start, length);
6532 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6533 set_result(result);
6534 return true;
6535 }
6536
6537 //------------------------------inline_updateByteBufferAdler32---------------
6538 //
6539 // Calculate Adler32 checksum for DirectByteBuffer.
6540 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6541 //
6542 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6543 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6544 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6545 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6546 // no receiver since it is static method
6547 Node* crc = argument(0); // type: int
6548 Node* src = argument(1); // type: long
6549 Node* offset = argument(3); // type: int
6550 Node* length = argument(4); // type: int
6551
6552 src = ConvL2X(src); // adjust Java long to machine word
6553 Node* base = _gvn.transform(new CastX2PNode(src));
6554 offset = ConvI2X(offset);
6555
6556 // 'src_start' points to src array + scaled offset
6557 Node* src_start = basic_plus_adr(top(), base, offset);
6558
6559 // Call the stub.
6560 address stubAddr = StubRoutines::updateBytesAdler32();
6561 const char *stubName = "updateBytesAdler32";
6562
6563 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6564 stubAddr, stubName, TypePtr::BOTTOM,
6565 crc, src_start, length);
6566
6567 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6568 set_result(result);
6569 return true;
6570 }
6571
6572 //----------------------------inline_reference_get----------------------------
6573 // public T java.lang.ref.Reference.get();
6574 bool LibraryCallKit::inline_reference_get() {
6575 const int referent_offset = java_lang_ref_Reference::referent_offset();
6576
6577 // Get the argument:
6578 Node* reference_obj = null_check_receiver();
6579 if (stopped()) return true;
6580
6581 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6582 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6583 decorators, /*is_static*/ false, nullptr);
6584 if (result == nullptr) return false;
6585
6586 // Add memory barrier to prevent commoning reads from this field
6587 // across safepoint since GC can change its value.
6588 insert_mem_bar(Op_MemBarCPUOrder);
6589
6590 set_result(result);
6591 return true;
6592 }
6593
6594 //----------------------------inline_reference_refersTo0----------------------------
6595 // bool java.lang.ref.Reference.refersTo0();
6596 // bool java.lang.ref.PhantomReference.refersTo0();
6597 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6598 // Get arguments:
6599 Node* reference_obj = null_check_receiver();
6600 Node* other_obj = argument(1);
6601 if (stopped()) return true;
6602
6603 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6604 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6605 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6606 decorators, /*is_static*/ false, nullptr);
6607 if (referent == nullptr) return false;
6608
6609 // Add memory barrier to prevent commoning reads from this field
6610 // across safepoint since GC can change its value.
6611 insert_mem_bar(Op_MemBarCPUOrder);
6612
6613 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6614 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6615 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6616
6617 RegionNode* region = new RegionNode(3);
6618 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6619
6620 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6621 region->init_req(1, if_true);
6622 phi->init_req(1, intcon(1));
6623
6624 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6625 region->init_req(2, if_false);
6626 phi->init_req(2, intcon(0));
6627
6628 set_control(_gvn.transform(region));
6629 record_for_igvn(region);
6630 set_result(_gvn.transform(phi));
6631 return true;
6632 }
6633
6634
6635 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6636 DecoratorSet decorators, bool is_static,
6637 ciInstanceKlass* fromKls) {
6638 if (fromKls == nullptr) {
6639 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6640 assert(tinst != nullptr, "obj is null");
6641 assert(tinst->is_loaded(), "obj is not loaded");
6642 fromKls = tinst->instance_klass();
6643 } else {
6644 assert(is_static, "only for static field access");
6645 }
6646 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6647 ciSymbol::make(fieldTypeString),
6648 is_static);
6649
6650 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6651 if (field == nullptr) return (Node *) nullptr;
6652
6653 if (is_static) {
6654 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6655 fromObj = makecon(tip);
6656 }
6657
6658 // Next code copied from Parse::do_get_xxx():
6659
6660 // Compute address and memory type.
6661 int offset = field->offset_in_bytes();
6662 bool is_vol = field->is_volatile();
6663 ciType* field_klass = field->type();
6664 assert(field_klass->is_loaded(), "should be loaded");
6665 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6666 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6667 BasicType bt = field->layout_type();
6668
6669 // Build the resultant type of the load
6670 const Type *type;
6671 if (bt == T_OBJECT) {
6672 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6673 } else {
6674 type = Type::get_const_basic_type(bt);
6675 }
6676
6677 if (is_vol) {
6678 decorators |= MO_SEQ_CST;
6679 }
6680
6681 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6682 }
6683
6684 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6685 bool is_exact /* true */, bool is_static /* false */,
6686 ciInstanceKlass * fromKls /* nullptr */) {
6687 if (fromKls == nullptr) {
6688 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6689 assert(tinst != nullptr, "obj is null");
6690 assert(tinst->is_loaded(), "obj is not loaded");
6691 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6692 fromKls = tinst->instance_klass();
6693 }
6694 else {
6695 assert(is_static, "only for static field access");
6696 }
6697 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6698 ciSymbol::make(fieldTypeString),
6699 is_static);
6700
6701 assert(field != nullptr, "undefined field");
6702 assert(!field->is_volatile(), "not defined for volatile fields");
6703
6704 if (is_static) {
6705 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6706 fromObj = makecon(tip);
6707 }
6708
6709 // Next code copied from Parse::do_get_xxx():
6710
6711 // Compute address and memory type.
6712 int offset = field->offset_in_bytes();
6713 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6714
6715 return adr;
6716 }
6717
6718 //------------------------------inline_aescrypt_Block-----------------------
6719 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6720 address stubAddr = nullptr;
6721 const char *stubName;
6722 assert(UseAES, "need AES instruction support");
6723
6724 switch(id) {
6725 case vmIntrinsics::_aescrypt_encryptBlock:
6726 stubAddr = StubRoutines::aescrypt_encryptBlock();
6727 stubName = "aescrypt_encryptBlock";
6728 break;
6729 case vmIntrinsics::_aescrypt_decryptBlock:
6730 stubAddr = StubRoutines::aescrypt_decryptBlock();
6731 stubName = "aescrypt_decryptBlock";
6732 break;
6733 default:
6734 break;
6735 }
6736 if (stubAddr == nullptr) return false;
6737
6738 Node* aescrypt_object = argument(0);
6739 Node* src = argument(1);
6740 Node* src_offset = argument(2);
6741 Node* dest = argument(3);
6742 Node* dest_offset = argument(4);
6743
6744 src = must_be_not_null(src, true);
6745 dest = must_be_not_null(dest, true);
6746
6747 // (1) src and dest are arrays.
6748 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6749 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6750 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6751 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6752
6753 // for the quick and dirty code we will skip all the checks.
6754 // we are just trying to get the call to be generated.
6755 Node* src_start = src;
6756 Node* dest_start = dest;
6757 if (src_offset != nullptr || dest_offset != nullptr) {
6758 assert(src_offset != nullptr && dest_offset != nullptr, "");
6759 src_start = array_element_address(src, src_offset, T_BYTE);
6760 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6761 }
6762
6763 // now need to get the start of its expanded key array
6764 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6765 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6766 if (k_start == nullptr) return false;
6767
6768 // Call the stub.
6769 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6770 stubAddr, stubName, TypePtr::BOTTOM,
6771 src_start, dest_start, k_start);
6772
6773 return true;
6774 }
6775
6776 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6777 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6778 address stubAddr = nullptr;
6779 const char *stubName = nullptr;
6780
6781 assert(UseAES, "need AES instruction support");
6782
6783 switch(id) {
6784 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6785 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6786 stubName = "cipherBlockChaining_encryptAESCrypt";
6787 break;
6788 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6789 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6790 stubName = "cipherBlockChaining_decryptAESCrypt";
6791 break;
6792 default:
6793 break;
6794 }
6795 if (stubAddr == nullptr) return false;
6796
6797 Node* cipherBlockChaining_object = argument(0);
6798 Node* src = argument(1);
6799 Node* src_offset = argument(2);
6800 Node* len = argument(3);
6801 Node* dest = argument(4);
6802 Node* dest_offset = argument(5);
6803
6804 src = must_be_not_null(src, false);
6805 dest = must_be_not_null(dest, false);
6806
6807 // (1) src and dest are arrays.
6808 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6809 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6810 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6811 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6812
6813 // checks are the responsibility of the caller
6814 Node* src_start = src;
6815 Node* dest_start = dest;
6816 if (src_offset != nullptr || dest_offset != nullptr) {
6817 assert(src_offset != nullptr && dest_offset != nullptr, "");
6818 src_start = array_element_address(src, src_offset, T_BYTE);
6819 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6820 }
6821
6822 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6823 // (because of the predicated logic executed earlier).
6824 // so we cast it here safely.
6825 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6826
6827 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6828 if (embeddedCipherObj == nullptr) return false;
6829
6830 // cast it to what we know it will be at runtime
6831 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6832 assert(tinst != nullptr, "CBC obj is null");
6833 assert(tinst->is_loaded(), "CBC obj is not loaded");
6834 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6835 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6836
6837 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6838 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6839 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6840 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6841 aescrypt_object = _gvn.transform(aescrypt_object);
6842
6843 // we need to get the start of the aescrypt_object's expanded key array
6844 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6845 if (k_start == nullptr) return false;
6846
6847 // similarly, get the start address of the r vector
6848 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6849 if (objRvec == nullptr) return false;
6850 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6851
6852 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6853 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6854 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6855 stubAddr, stubName, TypePtr::BOTTOM,
6856 src_start, dest_start, k_start, r_start, len);
6857
6858 // return cipher length (int)
6859 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6860 set_result(retvalue);
6861 return true;
6862 }
6863
6864 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6865 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6866 address stubAddr = nullptr;
6867 const char *stubName = nullptr;
6868
6869 assert(UseAES, "need AES instruction support");
6870
6871 switch (id) {
6872 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6873 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6874 stubName = "electronicCodeBook_encryptAESCrypt";
6875 break;
6876 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6877 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6878 stubName = "electronicCodeBook_decryptAESCrypt";
6879 break;
6880 default:
6881 break;
6882 }
6883
6884 if (stubAddr == nullptr) return false;
6885
6886 Node* electronicCodeBook_object = argument(0);
6887 Node* src = argument(1);
6888 Node* src_offset = argument(2);
6889 Node* len = argument(3);
6890 Node* dest = argument(4);
6891 Node* dest_offset = argument(5);
6892
6893 // (1) src and dest are arrays.
6894 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6895 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6896 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6897 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6898
6899 // checks are the responsibility of the caller
6900 Node* src_start = src;
6901 Node* dest_start = dest;
6902 if (src_offset != nullptr || dest_offset != nullptr) {
6903 assert(src_offset != nullptr && dest_offset != nullptr, "");
6904 src_start = array_element_address(src, src_offset, T_BYTE);
6905 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6906 }
6907
6908 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6909 // (because of the predicated logic executed earlier).
6910 // so we cast it here safely.
6911 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6912
6913 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6914 if (embeddedCipherObj == nullptr) return false;
6915
6916 // cast it to what we know it will be at runtime
6917 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6918 assert(tinst != nullptr, "ECB obj is null");
6919 assert(tinst->is_loaded(), "ECB obj is not loaded");
6920 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6921 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6922
6923 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6924 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6925 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6926 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6927 aescrypt_object = _gvn.transform(aescrypt_object);
6928
6929 // we need to get the start of the aescrypt_object's expanded key array
6930 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6931 if (k_start == nullptr) return false;
6932
6933 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6934 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6935 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6936 stubAddr, stubName, TypePtr::BOTTOM,
6937 src_start, dest_start, k_start, len);
6938
6939 // return cipher length (int)
6940 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6941 set_result(retvalue);
6942 return true;
6943 }
6944
6945 //------------------------------inline_counterMode_AESCrypt-----------------------
6946 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6947 assert(UseAES, "need AES instruction support");
6948 if (!UseAESCTRIntrinsics) return false;
6949
6950 address stubAddr = nullptr;
6951 const char *stubName = nullptr;
6952 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6953 stubAddr = StubRoutines::counterMode_AESCrypt();
6954 stubName = "counterMode_AESCrypt";
6955 }
6956 if (stubAddr == nullptr) return false;
6957
6958 Node* counterMode_object = argument(0);
6959 Node* src = argument(1);
6960 Node* src_offset = argument(2);
6961 Node* len = argument(3);
6962 Node* dest = argument(4);
6963 Node* dest_offset = argument(5);
6964
6965 // (1) src and dest are arrays.
6966 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6967 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6968 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6969 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6970
6971 // checks are the responsibility of the caller
6972 Node* src_start = src;
6973 Node* dest_start = dest;
6974 if (src_offset != nullptr || dest_offset != nullptr) {
6975 assert(src_offset != nullptr && dest_offset != nullptr, "");
6976 src_start = array_element_address(src, src_offset, T_BYTE);
6977 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6978 }
6979
6980 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6981 // (because of the predicated logic executed earlier).
6982 // so we cast it here safely.
6983 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6984 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6985 if (embeddedCipherObj == nullptr) return false;
6986 // cast it to what we know it will be at runtime
6987 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6988 assert(tinst != nullptr, "CTR obj is null");
6989 assert(tinst->is_loaded(), "CTR obj is not loaded");
6990 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6991 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6992 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6993 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6994 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6995 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6996 aescrypt_object = _gvn.transform(aescrypt_object);
6997 // we need to get the start of the aescrypt_object's expanded key array
6998 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6999 if (k_start == nullptr) return false;
7000 // similarly, get the start address of the r vector
7001 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7002 if (obj_counter == nullptr) return false;
7003 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7004
7005 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7006 if (saved_encCounter == nullptr) return false;
7007 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7008 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7009
7010 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7011 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7012 OptoRuntime::counterMode_aescrypt_Type(),
7013 stubAddr, stubName, TypePtr::BOTTOM,
7014 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7015
7016 // return cipher length (int)
7017 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7018 set_result(retvalue);
7019 return true;
7020 }
7021
7022 //------------------------------get_key_start_from_aescrypt_object-----------------------
7023 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7024 #if defined(PPC64) || defined(S390)
7025 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7026 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7027 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7028 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7029 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7030 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7031 if (objSessionK == nullptr) {
7032 return (Node *) nullptr;
7033 }
7034 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7035 #else
7036 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7037 #endif // PPC64
7038 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7039 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7040
7041 // now have the array, need to get the start address of the K array
7042 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7043 return k_start;
7044 }
7045
7046 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7047 // Return node representing slow path of predicate check.
7048 // the pseudo code we want to emulate with this predicate is:
7049 // for encryption:
7050 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7051 // for decryption:
7052 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7053 // note cipher==plain is more conservative than the original java code but that's OK
7054 //
7055 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7056 // The receiver was checked for null already.
7057 Node* objCBC = argument(0);
7058
7059 Node* src = argument(1);
7060 Node* dest = argument(4);
7061
7062 // Load embeddedCipher field of CipherBlockChaining object.
7063 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7064
7065 // get AESCrypt klass for instanceOf check
7066 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7067 // will have same classloader as CipherBlockChaining object
7068 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7069 assert(tinst != nullptr, "CBCobj is null");
7070 assert(tinst->is_loaded(), "CBCobj is not loaded");
7071
7072 // we want to do an instanceof comparison against the AESCrypt class
7073 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7074 if (!klass_AESCrypt->is_loaded()) {
7075 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7076 Node* ctrl = control();
7077 set_control(top()); // no regular fast path
7078 return ctrl;
7079 }
7080
7081 src = must_be_not_null(src, true);
7082 dest = must_be_not_null(dest, true);
7083
7084 // Resolve oops to stable for CmpP below.
7085 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7086
7087 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7088 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7089 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7090
7091 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7092
7093 // for encryption, we are done
7094 if (!decrypting)
7095 return instof_false; // even if it is null
7096
7097 // for decryption, we need to add a further check to avoid
7098 // taking the intrinsic path when cipher and plain are the same
7099 // see the original java code for why.
7100 RegionNode* region = new RegionNode(3);
7101 region->init_req(1, instof_false);
7102
7103 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7104 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7105 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7106 region->init_req(2, src_dest_conjoint);
7107
7108 record_for_igvn(region);
7109 return _gvn.transform(region);
7110 }
7111
7112 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7113 // Return node representing slow path of predicate check.
7114 // the pseudo code we want to emulate with this predicate is:
7115 // for encryption:
7116 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7117 // for decryption:
7118 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7119 // note cipher==plain is more conservative than the original java code but that's OK
7120 //
7121 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7122 // The receiver was checked for null already.
7123 Node* objECB = argument(0);
7124
7125 // Load embeddedCipher field of ElectronicCodeBook object.
7126 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7127
7128 // get AESCrypt klass for instanceOf check
7129 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7130 // will have same classloader as ElectronicCodeBook object
7131 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7132 assert(tinst != nullptr, "ECBobj is null");
7133 assert(tinst->is_loaded(), "ECBobj is not loaded");
7134
7135 // we want to do an instanceof comparison against the AESCrypt class
7136 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7137 if (!klass_AESCrypt->is_loaded()) {
7138 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7139 Node* ctrl = control();
7140 set_control(top()); // no regular fast path
7141 return ctrl;
7142 }
7143 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7144
7145 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7146 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7147 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7148
7149 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7150
7151 // for encryption, we are done
7152 if (!decrypting)
7153 return instof_false; // even if it is null
7154
7155 // for decryption, we need to add a further check to avoid
7156 // taking the intrinsic path when cipher and plain are the same
7157 // see the original java code for why.
7158 RegionNode* region = new RegionNode(3);
7159 region->init_req(1, instof_false);
7160 Node* src = argument(1);
7161 Node* dest = argument(4);
7162 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7163 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7164 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7165 region->init_req(2, src_dest_conjoint);
7166
7167 record_for_igvn(region);
7168 return _gvn.transform(region);
7169 }
7170
7171 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7172 // Return node representing slow path of predicate check.
7173 // the pseudo code we want to emulate with this predicate is:
7174 // for encryption:
7175 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7176 // for decryption:
7177 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7178 // note cipher==plain is more conservative than the original java code but that's OK
7179 //
7180
7181 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7182 // The receiver was checked for null already.
7183 Node* objCTR = argument(0);
7184
7185 // Load embeddedCipher field of CipherBlockChaining object.
7186 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7187
7188 // get AESCrypt klass for instanceOf check
7189 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7190 // will have same classloader as CipherBlockChaining object
7191 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7192 assert(tinst != nullptr, "CTRobj is null");
7193 assert(tinst->is_loaded(), "CTRobj is not loaded");
7194
7195 // we want to do an instanceof comparison against the AESCrypt class
7196 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7197 if (!klass_AESCrypt->is_loaded()) {
7198 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7199 Node* ctrl = control();
7200 set_control(top()); // no regular fast path
7201 return ctrl;
7202 }
7203
7204 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7205 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7206 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7207 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7208 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7209
7210 return instof_false; // even if it is null
7211 }
7212
7213 //------------------------------inline_ghash_processBlocks
7214 bool LibraryCallKit::inline_ghash_processBlocks() {
7215 address stubAddr;
7216 const char *stubName;
7217 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7218
7219 stubAddr = StubRoutines::ghash_processBlocks();
7220 stubName = "ghash_processBlocks";
7221
7222 Node* data = argument(0);
7223 Node* offset = argument(1);
7224 Node* len = argument(2);
7225 Node* state = argument(3);
7226 Node* subkeyH = argument(4);
7227
7228 state = must_be_not_null(state, true);
7229 subkeyH = must_be_not_null(subkeyH, true);
7230 data = must_be_not_null(data, true);
7231
7232 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7233 assert(state_start, "state is null");
7234 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7235 assert(subkeyH_start, "subkeyH is null");
7236 Node* data_start = array_element_address(data, offset, T_BYTE);
7237 assert(data_start, "data is null");
7238
7239 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7240 OptoRuntime::ghash_processBlocks_Type(),
7241 stubAddr, stubName, TypePtr::BOTTOM,
7242 state_start, subkeyH_start, data_start, len);
7243 return true;
7244 }
7245
7246 //------------------------------inline_chacha20Block
7247 bool LibraryCallKit::inline_chacha20Block() {
7248 address stubAddr;
7249 const char *stubName;
7250 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7251
7252 stubAddr = StubRoutines::chacha20Block();
7253 stubName = "chacha20Block";
7254
7255 Node* state = argument(0);
7256 Node* result = argument(1);
7257
7258 state = must_be_not_null(state, true);
7259 result = must_be_not_null(result, true);
7260
7261 Node* state_start = array_element_address(state, intcon(0), T_INT);
7262 assert(state_start, "state is null");
7263 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7264 assert(result_start, "result is null");
7265
7266 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7267 OptoRuntime::chacha20Block_Type(),
7268 stubAddr, stubName, TypePtr::BOTTOM,
7269 state_start, result_start);
7270 // return key stream length (int)
7271 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7272 set_result(retvalue);
7273 return true;
7274 }
7275
7276 bool LibraryCallKit::inline_base64_encodeBlock() {
7277 address stubAddr;
7278 const char *stubName;
7279 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7280 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7281 stubAddr = StubRoutines::base64_encodeBlock();
7282 stubName = "encodeBlock";
7283
7284 if (!stubAddr) return false;
7285 Node* base64obj = argument(0);
7286 Node* src = argument(1);
7287 Node* offset = argument(2);
7288 Node* len = argument(3);
7289 Node* dest = argument(4);
7290 Node* dp = argument(5);
7291 Node* isURL = argument(6);
7292
7293 src = must_be_not_null(src, true);
7294 dest = must_be_not_null(dest, true);
7295
7296 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7297 assert(src_start, "source array is null");
7298 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7299 assert(dest_start, "destination array is null");
7300
7301 Node* base64 = make_runtime_call(RC_LEAF,
7302 OptoRuntime::base64_encodeBlock_Type(),
7303 stubAddr, stubName, TypePtr::BOTTOM,
7304 src_start, offset, len, dest_start, dp, isURL);
7305 return true;
7306 }
7307
7308 bool LibraryCallKit::inline_base64_decodeBlock() {
7309 address stubAddr;
7310 const char *stubName;
7311 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7312 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7313 stubAddr = StubRoutines::base64_decodeBlock();
7314 stubName = "decodeBlock";
7315
7316 if (!stubAddr) return false;
7317 Node* base64obj = argument(0);
7318 Node* src = argument(1);
7319 Node* src_offset = argument(2);
7320 Node* len = argument(3);
7321 Node* dest = argument(4);
7322 Node* dest_offset = argument(5);
7323 Node* isURL = argument(6);
7324 Node* isMIME = argument(7);
7325
7326 src = must_be_not_null(src, true);
7327 dest = must_be_not_null(dest, true);
7328
7329 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7330 assert(src_start, "source array is null");
7331 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7332 assert(dest_start, "destination array is null");
7333
7334 Node* call = make_runtime_call(RC_LEAF,
7335 OptoRuntime::base64_decodeBlock_Type(),
7336 stubAddr, stubName, TypePtr::BOTTOM,
7337 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7338 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7339 set_result(result);
7340 return true;
7341 }
7342
7343 bool LibraryCallKit::inline_poly1305_processBlocks() {
7344 address stubAddr;
7345 const char *stubName;
7346 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7347 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7348 stubAddr = StubRoutines::poly1305_processBlocks();
7349 stubName = "poly1305_processBlocks";
7350
7351 if (!stubAddr) return false;
7352 null_check_receiver(); // null-check receiver
7353 if (stopped()) return true;
7354
7355 Node* input = argument(1);
7356 Node* input_offset = argument(2);
7357 Node* len = argument(3);
7358 Node* alimbs = argument(4);
7359 Node* rlimbs = argument(5);
7360
7361 input = must_be_not_null(input, true);
7362 alimbs = must_be_not_null(alimbs, true);
7363 rlimbs = must_be_not_null(rlimbs, true);
7364
7365 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7366 assert(input_start, "input array is null");
7367 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7368 assert(acc_start, "acc array is null");
7369 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7370 assert(r_start, "r array is null");
7371
7372 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7373 OptoRuntime::poly1305_processBlocks_Type(),
7374 stubAddr, stubName, TypePtr::BOTTOM,
7375 input_start, len, acc_start, r_start);
7376 return true;
7377 }
7378
7379 //------------------------------inline_digestBase_implCompress-----------------------
7380 //
7381 // Calculate MD5 for single-block byte[] array.
7382 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7383 //
7384 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7385 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7386 //
7387 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7388 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7389 //
7390 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7391 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7392 //
7393 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7394 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7395 //
7396 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7397 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7398
7399 Node* digestBase_obj = argument(0);
7400 Node* src = argument(1); // type oop
7401 Node* ofs = argument(2); // type int
7402
7403 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7404 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7405 // failed array check
7406 return false;
7407 }
7408 // Figure out the size and type of the elements we will be copying.
7409 BasicType src_elem = src_type->elem()->array_element_basic_type();
7410 if (src_elem != T_BYTE) {
7411 return false;
7412 }
7413 // 'src_start' points to src array + offset
7414 src = must_be_not_null(src, true);
7415 Node* src_start = array_element_address(src, ofs, src_elem);
7416 Node* state = nullptr;
7417 Node* block_size = nullptr;
7418 address stubAddr;
7419 const char *stubName;
7420
7421 switch(id) {
7422 case vmIntrinsics::_md5_implCompress:
7423 assert(UseMD5Intrinsics, "need MD5 instruction support");
7424 state = get_state_from_digest_object(digestBase_obj, T_INT);
7425 stubAddr = StubRoutines::md5_implCompress();
7426 stubName = "md5_implCompress";
7427 break;
7428 case vmIntrinsics::_sha_implCompress:
7429 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7430 state = get_state_from_digest_object(digestBase_obj, T_INT);
7431 stubAddr = StubRoutines::sha1_implCompress();
7432 stubName = "sha1_implCompress";
7433 break;
7434 case vmIntrinsics::_sha2_implCompress:
7435 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7436 state = get_state_from_digest_object(digestBase_obj, T_INT);
7437 stubAddr = StubRoutines::sha256_implCompress();
7438 stubName = "sha256_implCompress";
7439 break;
7440 case vmIntrinsics::_sha5_implCompress:
7441 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7442 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7443 stubAddr = StubRoutines::sha512_implCompress();
7444 stubName = "sha512_implCompress";
7445 break;
7446 case vmIntrinsics::_sha3_implCompress:
7447 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7448 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7449 stubAddr = StubRoutines::sha3_implCompress();
7450 stubName = "sha3_implCompress";
7451 block_size = get_block_size_from_digest_object(digestBase_obj);
7452 if (block_size == nullptr) return false;
7453 break;
7454 default:
7455 fatal_unexpected_iid(id);
7456 return false;
7457 }
7458 if (state == nullptr) return false;
7459
7460 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7461 if (stubAddr == nullptr) return false;
7462
7463 // Call the stub.
7464 Node* call;
7465 if (block_size == nullptr) {
7466 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7467 stubAddr, stubName, TypePtr::BOTTOM,
7468 src_start, state);
7469 } else {
7470 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7471 stubAddr, stubName, TypePtr::BOTTOM,
7472 src_start, state, block_size);
7473 }
7474
7475 return true;
7476 }
7477
7478 //------------------------------inline_digestBase_implCompressMB-----------------------
7479 //
7480 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7481 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7482 //
7483 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7484 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7485 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7486 assert((uint)predicate < 5, "sanity");
7487 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7488
7489 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7490 Node* src = argument(1); // byte[] array
7491 Node* ofs = argument(2); // type int
7492 Node* limit = argument(3); // type int
7493
7494 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7495 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7496 // failed array check
7497 return false;
7498 }
7499 // Figure out the size and type of the elements we will be copying.
7500 BasicType src_elem = src_type->elem()->array_element_basic_type();
7501 if (src_elem != T_BYTE) {
7502 return false;
7503 }
7504 // 'src_start' points to src array + offset
7505 src = must_be_not_null(src, false);
7506 Node* src_start = array_element_address(src, ofs, src_elem);
7507
7508 const char* klass_digestBase_name = nullptr;
7509 const char* stub_name = nullptr;
7510 address stub_addr = nullptr;
7511 BasicType elem_type = T_INT;
7512
7513 switch (predicate) {
7514 case 0:
7515 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7516 klass_digestBase_name = "sun/security/provider/MD5";
7517 stub_name = "md5_implCompressMB";
7518 stub_addr = StubRoutines::md5_implCompressMB();
7519 }
7520 break;
7521 case 1:
7522 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7523 klass_digestBase_name = "sun/security/provider/SHA";
7524 stub_name = "sha1_implCompressMB";
7525 stub_addr = StubRoutines::sha1_implCompressMB();
7526 }
7527 break;
7528 case 2:
7529 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7530 klass_digestBase_name = "sun/security/provider/SHA2";
7531 stub_name = "sha256_implCompressMB";
7532 stub_addr = StubRoutines::sha256_implCompressMB();
7533 }
7534 break;
7535 case 3:
7536 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7537 klass_digestBase_name = "sun/security/provider/SHA5";
7538 stub_name = "sha512_implCompressMB";
7539 stub_addr = StubRoutines::sha512_implCompressMB();
7540 elem_type = T_LONG;
7541 }
7542 break;
7543 case 4:
7544 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7545 klass_digestBase_name = "sun/security/provider/SHA3";
7546 stub_name = "sha3_implCompressMB";
7547 stub_addr = StubRoutines::sha3_implCompressMB();
7548 elem_type = T_BYTE;
7549 }
7550 break;
7551 default:
7552 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7553 }
7554 if (klass_digestBase_name != nullptr) {
7555 assert(stub_addr != nullptr, "Stub is generated");
7556 if (stub_addr == nullptr) return false;
7557
7558 // get DigestBase klass to lookup for SHA klass
7559 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7560 assert(tinst != nullptr, "digestBase_obj is not instance???");
7561 assert(tinst->is_loaded(), "DigestBase is not loaded");
7562
7563 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7564 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7565 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7566 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7567 }
7568 return false;
7569 }
7570
7571 //------------------------------inline_digestBase_implCompressMB-----------------------
7572 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7573 BasicType elem_type, address stubAddr, const char *stubName,
7574 Node* src_start, Node* ofs, Node* limit) {
7575 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7576 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7577 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7578 digest_obj = _gvn.transform(digest_obj);
7579
7580 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7581 if (state == nullptr) return false;
7582
7583 Node* block_size = nullptr;
7584 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7585 block_size = get_block_size_from_digest_object(digest_obj);
7586 if (block_size == nullptr) return false;
7587 }
7588
7589 // Call the stub.
7590 Node* call;
7591 if (block_size == nullptr) {
7592 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7593 OptoRuntime::digestBase_implCompressMB_Type(false),
7594 stubAddr, stubName, TypePtr::BOTTOM,
7595 src_start, state, ofs, limit);
7596 } else {
7597 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7598 OptoRuntime::digestBase_implCompressMB_Type(true),
7599 stubAddr, stubName, TypePtr::BOTTOM,
7600 src_start, state, block_size, ofs, limit);
7601 }
7602
7603 // return ofs (int)
7604 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7605 set_result(result);
7606
7607 return true;
7608 }
7609
7610 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7611 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7612 assert(UseAES, "need AES instruction support");
7613 address stubAddr = nullptr;
7614 const char *stubName = nullptr;
7615 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7616 stubName = "galoisCounterMode_AESCrypt";
7617
7618 if (stubAddr == nullptr) return false;
7619
7620 Node* in = argument(0);
7621 Node* inOfs = argument(1);
7622 Node* len = argument(2);
7623 Node* ct = argument(3);
7624 Node* ctOfs = argument(4);
7625 Node* out = argument(5);
7626 Node* outOfs = argument(6);
7627 Node* gctr_object = argument(7);
7628 Node* ghash_object = argument(8);
7629
7630 // (1) in, ct and out are arrays.
7631 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7632 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
7633 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7634 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
7635 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
7636 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
7637
7638 // checks are the responsibility of the caller
7639 Node* in_start = in;
7640 Node* ct_start = ct;
7641 Node* out_start = out;
7642 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
7643 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
7644 in_start = array_element_address(in, inOfs, T_BYTE);
7645 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7646 out_start = array_element_address(out, outOfs, T_BYTE);
7647 }
7648
7649 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7650 // (because of the predicated logic executed earlier).
7651 // so we cast it here safely.
7652 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7653 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7654 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7655 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7656 Node* state = load_field_from_object(ghash_object, "state", "[J");
7657
7658 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
7659 return false;
7660 }
7661 // cast it to what we know it will be at runtime
7662 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7663 assert(tinst != nullptr, "GCTR obj is null");
7664 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7665 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7666 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7667 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7668 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7669 const TypeOopPtr* xtype = aklass->as_instance_type();
7670 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7671 aescrypt_object = _gvn.transform(aescrypt_object);
7672 // we need to get the start of the aescrypt_object's expanded key array
7673 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7674 if (k_start == nullptr) return false;
7675 // similarly, get the start address of the r vector
7676 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7677 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7678 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7679
7680
7681 // Call the stub, passing params
7682 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7683 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7684 stubAddr, stubName, TypePtr::BOTTOM,
7685 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7686
7687 // return cipher length (int)
7688 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7689 set_result(retvalue);
7690
7691 return true;
7692 }
7693
7694 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7695 // Return node representing slow path of predicate check.
7696 // the pseudo code we want to emulate with this predicate is:
7697 // for encryption:
7698 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7699 // for decryption:
7700 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7701 // note cipher==plain is more conservative than the original java code but that's OK
7702 //
7703
7704 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7705 // The receiver was checked for null already.
7706 Node* objGCTR = argument(7);
7707 // Load embeddedCipher field of GCTR object.
7708 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7709 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
7710
7711 // get AESCrypt klass for instanceOf check
7712 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7713 // will have same classloader as CipherBlockChaining object
7714 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7715 assert(tinst != nullptr, "GCTR obj is null");
7716 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7717
7718 // we want to do an instanceof comparison against the AESCrypt class
7719 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7720 if (!klass_AESCrypt->is_loaded()) {
7721 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7722 Node* ctrl = control();
7723 set_control(top()); // no regular fast path
7724 return ctrl;
7725 }
7726
7727 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7728 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7729 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7730 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7731 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7732
7733 return instof_false; // even if it is null
7734 }
7735
7736 //------------------------------get_state_from_digest_object-----------------------
7737 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7738 const char* state_type;
7739 switch (elem_type) {
7740 case T_BYTE: state_type = "[B"; break;
7741 case T_INT: state_type = "[I"; break;
7742 case T_LONG: state_type = "[J"; break;
7743 default: ShouldNotReachHere();
7744 }
7745 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7746 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7747 if (digest_state == nullptr) return (Node *) nullptr;
7748
7749 // now have the array, need to get the start address of the state array
7750 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7751 return state;
7752 }
7753
7754 //------------------------------get_block_size_from_sha3_object----------------------------------
7755 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7756 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7757 assert (block_size != nullptr, "sanity");
7758 return block_size;
7759 }
7760
7761 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7762 // Return node representing slow path of predicate check.
7763 // the pseudo code we want to emulate with this predicate is:
7764 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7765 //
7766 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7767 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7768 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7769 assert((uint)predicate < 5, "sanity");
7770
7771 // The receiver was checked for null already.
7772 Node* digestBaseObj = argument(0);
7773
7774 // get DigestBase klass for instanceOf check
7775 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7776 assert(tinst != nullptr, "digestBaseObj is null");
7777 assert(tinst->is_loaded(), "DigestBase is not loaded");
7778
7779 const char* klass_name = nullptr;
7780 switch (predicate) {
7781 case 0:
7782 if (UseMD5Intrinsics) {
7783 // we want to do an instanceof comparison against the MD5 class
7784 klass_name = "sun/security/provider/MD5";
7785 }
7786 break;
7787 case 1:
7788 if (UseSHA1Intrinsics) {
7789 // we want to do an instanceof comparison against the SHA class
7790 klass_name = "sun/security/provider/SHA";
7791 }
7792 break;
7793 case 2:
7794 if (UseSHA256Intrinsics) {
7795 // we want to do an instanceof comparison against the SHA2 class
7796 klass_name = "sun/security/provider/SHA2";
7797 }
7798 break;
7799 case 3:
7800 if (UseSHA512Intrinsics) {
7801 // we want to do an instanceof comparison against the SHA5 class
7802 klass_name = "sun/security/provider/SHA5";
7803 }
7804 break;
7805 case 4:
7806 if (UseSHA3Intrinsics) {
7807 // we want to do an instanceof comparison against the SHA3 class
7808 klass_name = "sun/security/provider/SHA3";
7809 }
7810 break;
7811 default:
7812 fatal("unknown SHA intrinsic predicate: %d", predicate);
7813 }
7814
7815 ciKlass* klass = nullptr;
7816 if (klass_name != nullptr) {
7817 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7818 }
7819 if ((klass == nullptr) || !klass->is_loaded()) {
7820 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7821 Node* ctrl = control();
7822 set_control(top()); // no intrinsic path
7823 return ctrl;
7824 }
7825 ciInstanceKlass* instklass = klass->as_instance_klass();
7826
7827 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7828 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7829 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7830 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7831
7832 return instof_false; // even if it is null
7833 }
7834
7835 //-------------inline_fma-----------------------------------
7836 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7837 Node *a = nullptr;
7838 Node *b = nullptr;
7839 Node *c = nullptr;
7840 Node* result = nullptr;
7841 switch (id) {
7842 case vmIntrinsics::_fmaD:
7843 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7844 // no receiver since it is static method
7845 a = round_double_node(argument(0));
7846 b = round_double_node(argument(2));
7847 c = round_double_node(argument(4));
7848 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7849 break;
7850 case vmIntrinsics::_fmaF:
7851 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7852 a = argument(0);
7853 b = argument(1);
7854 c = argument(2);
7855 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7856 break;
7857 default:
7858 fatal_unexpected_iid(id); break;
7859 }
7860 set_result(result);
7861 return true;
7862 }
7863
7864 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7865 // argument(0) is receiver
7866 Node* codePoint = argument(1);
7867 Node* n = nullptr;
7868
7869 switch (id) {
7870 case vmIntrinsics::_isDigit :
7871 n = new DigitNode(control(), codePoint);
7872 break;
7873 case vmIntrinsics::_isLowerCase :
7874 n = new LowerCaseNode(control(), codePoint);
7875 break;
7876 case vmIntrinsics::_isUpperCase :
7877 n = new UpperCaseNode(control(), codePoint);
7878 break;
7879 case vmIntrinsics::_isWhitespace :
7880 n = new WhitespaceNode(control(), codePoint);
7881 break;
7882 default:
7883 fatal_unexpected_iid(id);
7884 }
7885
7886 set_result(_gvn.transform(n));
7887 return true;
7888 }
7889
7890 //------------------------------inline_fp_min_max------------------------------
7891 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7892 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7893
7894 // The intrinsic should be used only when the API branches aren't predictable,
7895 // the last one performing the most important comparison. The following heuristic
7896 // uses the branch statistics to eventually bail out if necessary.
7897
7898 ciMethodData *md = callee()->method_data();
7899
7900 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
7901 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7902
7903 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7904 // Bail out if the call-site didn't contribute enough to the statistics.
7905 return false;
7906 }
7907
7908 uint taken = 0, not_taken = 0;
7909
7910 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7911 if (p->is_BranchData()) {
7912 taken = ((ciBranchData*)p)->taken();
7913 not_taken = ((ciBranchData*)p)->not_taken();
7914 }
7915 }
7916
7917 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7918 balance = balance < 0 ? -balance : balance;
7919 if ( balance > 0.2 ) {
7920 // Bail out if the most important branch is predictable enough.
7921 return false;
7922 }
7923 }
7924 */
7925
7926 Node *a = nullptr;
7927 Node *b = nullptr;
7928 Node *n = nullptr;
7929 switch (id) {
7930 case vmIntrinsics::_maxF:
7931 case vmIntrinsics::_minF:
7932 case vmIntrinsics::_maxF_strict:
7933 case vmIntrinsics::_minF_strict:
7934 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7935 a = argument(0);
7936 b = argument(1);
7937 break;
7938 case vmIntrinsics::_maxD:
7939 case vmIntrinsics::_minD:
7940 case vmIntrinsics::_maxD_strict:
7941 case vmIntrinsics::_minD_strict:
7942 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7943 a = round_double_node(argument(0));
7944 b = round_double_node(argument(2));
7945 break;
7946 default:
7947 fatal_unexpected_iid(id);
7948 break;
7949 }
7950 switch (id) {
7951 case vmIntrinsics::_maxF:
7952 case vmIntrinsics::_maxF_strict:
7953 n = new MaxFNode(a, b);
7954 break;
7955 case vmIntrinsics::_minF:
7956 case vmIntrinsics::_minF_strict:
7957 n = new MinFNode(a, b);
7958 break;
7959 case vmIntrinsics::_maxD:
7960 case vmIntrinsics::_maxD_strict:
7961 n = new MaxDNode(a, b);
7962 break;
7963 case vmIntrinsics::_minD:
7964 case vmIntrinsics::_minD_strict:
7965 n = new MinDNode(a, b);
7966 break;
7967 default:
7968 fatal_unexpected_iid(id);
7969 break;
7970 }
7971 set_result(_gvn.transform(n));
7972 return true;
7973 }
7974
7975 bool LibraryCallKit::inline_profileBoolean() {
7976 Node* counts = argument(1);
7977 const TypeAryPtr* ary = nullptr;
7978 ciArray* aobj = nullptr;
7979 if (counts->is_Con()
7980 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
7981 && (aobj = ary->const_oop()->as_array()) != nullptr
7982 && (aobj->length() == 2)) {
7983 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7984 jint false_cnt = aobj->element_value(0).as_int();
7985 jint true_cnt = aobj->element_value(1).as_int();
7986
7987 if (C->log() != nullptr) {
7988 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7989 false_cnt, true_cnt);
7990 }
7991
7992 if (false_cnt + true_cnt == 0) {
7993 // According to profile, never executed.
7994 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7995 Deoptimization::Action_reinterpret);
7996 return true;
7997 }
7998
7999 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8000 // is a number of each value occurrences.
8001 Node* result = argument(0);
8002 if (false_cnt == 0 || true_cnt == 0) {
8003 // According to profile, one value has been never seen.
8004 int expected_val = (false_cnt == 0) ? 1 : 0;
8005
8006 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8007 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8008
8009 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8010 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8011 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8012
8013 { // Slow path: uncommon trap for never seen value and then reexecute
8014 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8015 // the value has been seen at least once.
8016 PreserveJVMState pjvms(this);
8017 PreserveReexecuteState preexecs(this);
8018 jvms()->set_should_reexecute(true);
8019
8020 set_control(slow_path);
8021 set_i_o(i_o());
8022
8023 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8024 Deoptimization::Action_reinterpret);
8025 }
8026 // The guard for never seen value enables sharpening of the result and
8027 // returning a constant. It allows to eliminate branches on the same value
8028 // later on.
8029 set_control(fast_path);
8030 result = intcon(expected_val);
8031 }
8032 // Stop profiling.
8033 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8034 // By replacing method body with profile data (represented as ProfileBooleanNode
8035 // on IR level) we effectively disable profiling.
8036 // It enables full speed execution once optimized code is generated.
8037 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8038 C->record_for_igvn(profile);
8039 set_result(profile);
8040 return true;
8041 } else {
8042 // Continue profiling.
8043 // Profile data isn't available at the moment. So, execute method's bytecode version.
8044 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8045 // is compiled and counters aren't available since corresponding MethodHandle
8046 // isn't a compile-time constant.
8047 return false;
8048 }
8049 }
8050
8051 bool LibraryCallKit::inline_isCompileConstant() {
8052 Node* n = argument(0);
8053 set_result(n->is_Con() ? intcon(1) : intcon(0));
8054 return true;
8055 }
8056
8057 //------------------------------- inline_getObjectSize/sizeOf --------------------------------------
8058 //
8059 // Calculate the runtime size of the object/array.
8060 //
8061 // long java.lang.Runtime.sizeOf(Object obj)
8062 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8063 //
8064 bool LibraryCallKit::inline_sizeOf() {
8065 Node* obj = argument(0);
8066 return inline_sizeOf_impl(obj);
8067 }
8068
8069 bool LibraryCallKit::inline_getObjectSize() {
8070 Node* obj = argument(3);
8071 return inline_sizeOf_impl(obj);
8072 }
8073
8074 bool LibraryCallKit::inline_sizeOf_impl(Node* obj) {
8075 Node* klass_node = load_object_klass(obj);
8076
8077 jint layout_con = Klass::_lh_neutral_value;
8078 Node* layout_val = get_layout_helper(klass_node, layout_con);
8079 int layout_is_con = (layout_val == nullptr);
8080
8081 if (layout_is_con) {
8082 // Layout helper is constant, can figure out things at compile time.
8083
8084 if (Klass::layout_helper_is_instance(layout_con)) {
8085 // Instance case: layout_con contains the size itself.
8086 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8087 set_result(size);
8088 } else {
8089 // Array case: size is round(header + element_size*arraylength).
8090 // Since arraylength is different for every array instance, we have to
8091 // compute the whole thing at runtime.
8092
8093 Node* arr_length = load_array_length(obj);
8094
8095 int round_mask = MinObjAlignmentInBytes - 1;
8096 int hsize = Klass::layout_helper_header_size(layout_con);
8097 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8098
8099 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8100 round_mask = 0; // strength-reduce it if it goes away completely
8101 }
8102 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8103 Node* header_size = intcon(hsize + round_mask);
8104
8105 Node* lengthx = ConvI2X(arr_length);
8106 Node* headerx = ConvI2X(header_size);
8107
8108 Node* abody = lengthx;
8109 if (eshift != 0) {
8110 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8111 }
8112 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8113 if (round_mask != 0) {
8114 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8115 }
8116 size = ConvX2L(size);
8117 set_result(size);
8118 }
8119 } else {
8120 // Layout helper is not constant, need to test for array-ness at runtime.
8121
8122 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8123 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8124 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8125 record_for_igvn(result_reg);
8126
8127 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8128 if (array_ctl != nullptr) {
8129 // Array case: size is round(header + element_size*arraylength).
8130 // Since arraylength is different for every array instance, we have to
8131 // compute the whole thing at runtime.
8132
8133 PreserveJVMState pjvms(this);
8134 set_control(array_ctl);
8135 Node* arr_length = load_array_length(obj);
8136
8137 int round_mask = MinObjAlignmentInBytes - 1;
8138 Node* mask = intcon(round_mask);
8139
8140 Node* hss = intcon(Klass::_lh_header_size_shift);
8141 Node* hsm = intcon(Klass::_lh_header_size_mask);
8142 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8143 header_size = _gvn.transform(new AndINode(header_size, hsm));
8144 header_size = _gvn.transform(new AddINode(header_size, mask));
8145
8146 // There is no need to mask or shift this value.
8147 // The semantics of LShiftINode include an implicit mask to 0x1F.
8148 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8149 Node* elem_shift = layout_val;
8150
8151 Node* lengthx = ConvI2X(arr_length);
8152 Node* headerx = ConvI2X(header_size);
8153
8154 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8155 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8156 if (round_mask != 0) {
8157 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8158 }
8159 size = ConvX2L(size);
8160
8161 result_reg->init_req(_array_path, control());
8162 result_val->init_req(_array_path, size);
8163 }
8164
8165 if (!stopped()) {
8166 // Instance case: the layout helper gives us instance size almost directly,
8167 // but we need to mask out the _lh_instance_slow_path_bit.
8168 Node* size = ConvI2X(layout_val);
8169 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8170 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8171 size = _gvn.transform(new AndXNode(size, mask));
8172 size = ConvX2L(size);
8173
8174 result_reg->init_req(_instance_path, control());
8175 result_val->init_req(_instance_path, size);
8176 }
8177
8178 set_result(result_reg, result_val);
8179 }
8180
8181 return true;
8182 }
8183
8184 //------------------------------- inline_blackhole --------------------------------------
8185 //
8186 // Make sure all arguments to this node are alive.
8187 // This matches methods that were requested to be blackholed through compile commands.
8188 //
8189 bool LibraryCallKit::inline_blackhole() {
8190 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8191 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8192 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8193
8194 // Blackhole node pinches only the control, not memory. This allows
8195 // the blackhole to be pinned in the loop that computes blackholed
8196 // values, but have no other side effects, like breaking the optimizations
8197 // across the blackhole.
8198
8199 Node* bh = _gvn.transform(new BlackholeNode(control()));
8200 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8201
8202 // Bind call arguments as blackhole arguments to keep them alive
8203 uint nargs = callee()->arg_size();
8204 for (uint i = 0; i < nargs; i++) {
8205 bh->add_req(argument(i));
8206 }
8207
8208 return true;
8209 }
8210
8211 bool LibraryCallKit::inline_timestamp(bool serial) {
8212 insert_mem_bar(Op_MemBarCPUOrder);
8213 Node* node = serial ?
8214 _gvn.transform(new TimestampSerialNode(control())) :
8215 _gvn.transform(new TimestampNode(control()));
8216 set_result(node);
8217 insert_mem_bar(Op_MemBarCPUOrder);
8218 return true;
8219 }
8220
8221 bool LibraryCallKit::inline_addressOf() {
8222 Node* obj = argument(0);
8223 Node* raw_val = _gvn.transform(new CastP2XNode(NULL, obj));
8224 Node* long_val = ConvX2L(raw_val);
8225 set_result(long_val);
8226 return true;
8227 }