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