1 /*
2 * Copyright (c) 1999, 2022, 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/objectMonitor.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "utilities/macros.hpp"
58 #include "utilities/powerOfTwo.hpp"
59
60 //---------------------------make_vm_intrinsic----------------------------
61 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
62 vmIntrinsicID id = m->intrinsic_id();
63 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
64
65 if (!m->is_loaded()) {
66 // Do not attempt to inline unloaded methods.
67 return NULL;
68 }
69
70 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
71 bool is_available = false;
72
73 {
74 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
75 // the compiler must transition to '_thread_in_vm' state because both
76 // methods access VM-internal data.
77 VM_ENTRY_MARK;
78 methodHandle mh(THREAD, m->get_Method());
79 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
80 !C->directive()->is_intrinsic_disabled(mh) &&
81 !vmIntrinsics::is_disabled_by_flags(mh);
82
83 }
84
85 if (is_available) {
86 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
87 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
88 return new LibraryIntrinsic(m, is_virtual,
89 vmIntrinsics::predicates_needed(id),
90 vmIntrinsics::does_virtual_dispatch(id),
91 id);
92 } else {
93 return NULL;
94 }
95 }
96
97 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
98 LibraryCallKit kit(jvms, this);
99 Compile* C = kit.C;
100 int nodes = C->unique();
101 #ifndef PRODUCT
102 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
103 char buf[1000];
104 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
105 tty->print_cr("Intrinsic %s", str);
106 }
107 #endif
108 ciMethod* callee = kit.callee();
109 const int bci = kit.bci();
110 #ifdef ASSERT
111 Node* ctrl = kit.control();
112 #endif
113 // Try to inline the intrinsic.
114 if (callee->check_intrinsic_candidate() &&
115 kit.try_to_inline(_last_predicate)) {
116 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
117 : "(intrinsic)";
118 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
119 if (C->print_intrinsics() || C->print_inlining()) {
120 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
121 }
122 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
123 if (C->log()) {
124 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
125 vmIntrinsics::name_at(intrinsic_id()),
126 (is_virtual() ? " virtual='1'" : ""),
127 C->unique() - nodes);
128 }
129 // Push the result from the inlined method onto the stack.
130 kit.push_result();
131 C->print_inlining_update(this);
132 return kit.transfer_exceptions_into_jvms();
133 }
134
135 // The intrinsic bailed out
136 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
137 if (jvms->has_method()) {
138 // Not a root compile.
139 const char* msg;
140 if (callee->intrinsic_candidate()) {
141 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
142 } else {
143 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
144 : "failed to inline (intrinsic), method not annotated";
145 }
146 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
147 if (C->print_intrinsics() || C->print_inlining()) {
148 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
149 }
150 } else {
151 // Root compile
152 ResourceMark rm;
153 stringStream msg_stream;
154 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
155 vmIntrinsics::name_at(intrinsic_id()),
156 is_virtual() ? " (virtual)" : "", bci);
157 const char *msg = msg_stream.freeze();
158 log_debug(jit, inlining)("%s", msg);
159 if (C->print_intrinsics() || C->print_inlining()) {
160 tty->print("%s", msg);
161 }
162 }
163 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
164 C->print_inlining_update(this);
165
166 return NULL;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
190 if (C->print_intrinsics() || C->print_inlining()) {
191 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
192 }
193 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
194 if (C->log()) {
195 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
196 vmIntrinsics::name_at(intrinsic_id()),
197 (is_virtual() ? " virtual='1'" : ""),
198 C->unique() - nodes);
199 }
200 return slow_ctl; // Could be NULL if the check folds.
201 }
202
203 // The intrinsic bailed out
204 if (jvms->has_method()) {
205 // Not a root compile.
206 const char* msg = "failed to generate predicate for intrinsic";
207 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
208 if (C->print_intrinsics() || C->print_inlining()) {
209 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
210 }
211 } else {
212 // Root compile
213 ResourceMark rm;
214 stringStream msg_stream;
215 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
216 vmIntrinsics::name_at(intrinsic_id()),
217 is_virtual() ? " (virtual)" : "", bci);
218 const char *msg = msg_stream.freeze();
219 log_debug(jit, inlining)("%s", msg);
220 if (C->print_intrinsics() || C->print_inlining()) {
221 C->print_inlining_stream()->print("%s", msg);
222 }
223 }
224 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
225 return NULL;
226 }
227
228 bool LibraryCallKit::try_to_inline(int predicate) {
229 // Handle symbolic names for otherwise undistinguished boolean switches:
230 const bool is_store = true;
231 const bool is_compress = true;
232 const bool is_static = true;
233 const bool is_volatile = true;
234
235 if (!jvms()->has_method()) {
236 // Root JVMState has a null method.
237 assert(map()->memory()->Opcode() == Op_Parm, "");
238 // Insert the memory aliasing node
239 set_all_memory(reset_memory());
240 }
241 assert(merged_memory(), "");
242
243 switch (intrinsic_id()) {
244 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
245 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
246 case vmIntrinsics::_getClass: return inline_native_getClass();
247
248 case vmIntrinsics::_ceil:
249 case vmIntrinsics::_floor:
250 case vmIntrinsics::_rint:
251 case vmIntrinsics::_dsin:
252 case vmIntrinsics::_dcos:
253 case vmIntrinsics::_dtan:
254 case vmIntrinsics::_dabs:
255 case vmIntrinsics::_fabs:
256 case vmIntrinsics::_iabs:
257 case vmIntrinsics::_labs:
258 case vmIntrinsics::_datan2:
259 case vmIntrinsics::_dsqrt:
260 case vmIntrinsics::_dsqrt_strict:
261 case vmIntrinsics::_dexp:
262 case vmIntrinsics::_dlog:
263 case vmIntrinsics::_dlog10:
264 case vmIntrinsics::_dpow:
265 case vmIntrinsics::_dcopySign:
266 case vmIntrinsics::_fcopySign:
267 case vmIntrinsics::_dsignum:
268 case vmIntrinsics::_roundF:
269 case vmIntrinsics::_roundD:
270 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
271
272 case vmIntrinsics::_notify:
273 case vmIntrinsics::_notifyAll:
274 return inline_notify(intrinsic_id());
275
276 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
277 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
278 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
279 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
280 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
281 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
282 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
283 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
284 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
285 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
286 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
287 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
288 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
289 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
290
291 case vmIntrinsics::_arraycopy: return inline_arraycopy();
292
293 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
294 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
295 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
296 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
297
298 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
305 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
306
307 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
308 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
309
310 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
311 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
312 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
313 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
314
315 case vmIntrinsics::_compressStringC:
316 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
317 case vmIntrinsics::_inflateStringC:
318 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
319
320 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
321 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
322 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
323 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
324 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
325 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
326 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
327 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
328 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
329
330 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
331 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
332 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
333 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
334 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
335 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
336 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
337 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
338 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
339
340 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
341 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
342 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
343 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
344 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
345 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
346 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
347 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
348 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
349
350 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
351 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
352 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
353 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
354 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
355 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
356 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
357 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
358 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
359
360 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
361 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
362 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
363 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
364
365 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
366 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
367 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
368 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
369
370 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
371 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
372 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
373 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
374 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
375 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
376 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
377 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
378 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
379
380 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
381 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
382 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
383 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
384 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
385 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
386 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
387 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
388 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
389
390 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
391 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
392 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
393 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
394 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
395 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
396 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
397 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
398 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
399
400 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
401 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
402 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
403 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
404 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
405 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
406 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
407 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
408 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
409
410 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
415
416 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
417 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
418 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
419 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
420 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
421 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
422 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
423 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
424 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
436
437 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
438 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
439 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
440 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
452
453 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
455 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
457
458 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
463
464 case vmIntrinsics::_loadFence:
465 case vmIntrinsics::_storeFence:
466 case vmIntrinsics::_storeStoreFence:
467 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
468
469 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
470
471 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
472 case vmIntrinsics::_currentThread: return inline_native_currentThread();
473 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
474
475 case vmIntrinsics::_extentLocalCache: return inline_native_extentLocalCache();
476 case vmIntrinsics::_setExtentLocalCache: return inline_native_setExtentLocalCache();
477
478 #ifdef JFR_HAVE_INTRINSICS
479 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
480 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
481 #endif
482 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
483 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
484 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
485 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
486 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
487 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
488 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
489 case vmIntrinsics::_getLength: return inline_native_getLength();
490 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
491 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
492 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
493 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
494 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
495 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
496 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
497
498 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
499 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
500
501 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
502
503 case vmIntrinsics::_isInstance:
504 case vmIntrinsics::_getModifiers:
505 case vmIntrinsics::_isInterface:
506 case vmIntrinsics::_isArray:
507 case vmIntrinsics::_isPrimitive:
508 case vmIntrinsics::_isHidden:
509 case vmIntrinsics::_getSuperclass:
510 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
511
512 case vmIntrinsics::_floatToRawIntBits:
513 case vmIntrinsics::_floatToIntBits:
514 case vmIntrinsics::_intBitsToFloat:
515 case vmIntrinsics::_doubleToRawLongBits:
516 case vmIntrinsics::_doubleToLongBits:
517 case vmIntrinsics::_longBitsToDouble:
518 case vmIntrinsics::_floatToFloat16:
519 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
520
521 case vmIntrinsics::_floatIsFinite:
522 case vmIntrinsics::_floatIsInfinite:
523 case vmIntrinsics::_doubleIsFinite:
524 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
525
526 case vmIntrinsics::_numberOfLeadingZeros_i:
527 case vmIntrinsics::_numberOfLeadingZeros_l:
528 case vmIntrinsics::_numberOfTrailingZeros_i:
529 case vmIntrinsics::_numberOfTrailingZeros_l:
530 case vmIntrinsics::_bitCount_i:
531 case vmIntrinsics::_bitCount_l:
532 case vmIntrinsics::_reverse_i:
533 case vmIntrinsics::_reverse_l:
534 case vmIntrinsics::_reverseBytes_i:
535 case vmIntrinsics::_reverseBytes_l:
536 case vmIntrinsics::_reverseBytes_s:
537 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
538
539 case vmIntrinsics::_compress_i:
540 case vmIntrinsics::_compress_l:
541 case vmIntrinsics::_expand_i:
542 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
543
544 case vmIntrinsics::_compareUnsigned_i:
545 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
546
547 case vmIntrinsics::_divideUnsigned_i:
548 case vmIntrinsics::_divideUnsigned_l:
549 case vmIntrinsics::_remainderUnsigned_i:
550 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
551
552 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
553
554 case vmIntrinsics::_Reference_get: return inline_reference_get();
555 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
556 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
557
558 case vmIntrinsics::_Class_cast: return inline_Class_cast();
559
560 case vmIntrinsics::_aescrypt_encryptBlock:
561 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
562
563 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
564 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
565 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
566
567 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
568 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
569 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
570
571 case vmIntrinsics::_counterMode_AESCrypt:
572 return inline_counterMode_AESCrypt(intrinsic_id());
573
574 case vmIntrinsics::_galoisCounterMode_AESCrypt:
575 return inline_galoisCounterMode_AESCrypt();
576
577 case vmIntrinsics::_md5_implCompress:
578 case vmIntrinsics::_sha_implCompress:
579 case vmIntrinsics::_sha2_implCompress:
580 case vmIntrinsics::_sha5_implCompress:
581 case vmIntrinsics::_sha3_implCompress:
582 return inline_digestBase_implCompress(intrinsic_id());
583
584 case vmIntrinsics::_digestBase_implCompressMB:
585 return inline_digestBase_implCompressMB(predicate);
586
587 case vmIntrinsics::_multiplyToLen:
588 return inline_multiplyToLen();
589
590 case vmIntrinsics::_squareToLen:
591 return inline_squareToLen();
592
593 case vmIntrinsics::_mulAdd:
594 return inline_mulAdd();
595
596 case vmIntrinsics::_montgomeryMultiply:
597 return inline_montgomeryMultiply();
598 case vmIntrinsics::_montgomerySquare:
599 return inline_montgomerySquare();
600
601 case vmIntrinsics::_bigIntegerRightShiftWorker:
602 return inline_bigIntegerShift(true);
603 case vmIntrinsics::_bigIntegerLeftShiftWorker:
604 return inline_bigIntegerShift(false);
605
606 case vmIntrinsics::_vectorizedMismatch:
607 return inline_vectorizedMismatch();
608
609 case vmIntrinsics::_ghash_processBlocks:
610 return inline_ghash_processBlocks();
611 case vmIntrinsics::_base64_encodeBlock:
612 return inline_base64_encodeBlock();
613 case vmIntrinsics::_base64_decodeBlock:
614 return inline_base64_decodeBlock();
615
616 case vmIntrinsics::_encodeISOArray:
617 case vmIntrinsics::_encodeByteISOArray:
618 return inline_encodeISOArray(false);
619 case vmIntrinsics::_encodeAsciiArray:
620 return inline_encodeISOArray(true);
621
622 case vmIntrinsics::_updateCRC32:
623 return inline_updateCRC32();
624 case vmIntrinsics::_updateBytesCRC32:
625 return inline_updateBytesCRC32();
626 case vmIntrinsics::_updateByteBufferCRC32:
627 return inline_updateByteBufferCRC32();
628
629 case vmIntrinsics::_updateBytesCRC32C:
630 return inline_updateBytesCRC32C();
631 case vmIntrinsics::_updateDirectByteBufferCRC32C:
632 return inline_updateDirectByteBufferCRC32C();
633
634 case vmIntrinsics::_updateBytesAdler32:
635 return inline_updateBytesAdler32();
636 case vmIntrinsics::_updateByteBufferAdler32:
637 return inline_updateByteBufferAdler32();
638
639 case vmIntrinsics::_profileBoolean:
640 return inline_profileBoolean();
641 case vmIntrinsics::_isCompileConstant:
642 return inline_isCompileConstant();
643
644 case vmIntrinsics::_countPositives:
645 return inline_countPositives();
646
647 case vmIntrinsics::_fmaD:
648 case vmIntrinsics::_fmaF:
649 return inline_fma(intrinsic_id());
650
651 case vmIntrinsics::_isDigit:
652 case vmIntrinsics::_isLowerCase:
653 case vmIntrinsics::_isUpperCase:
654 case vmIntrinsics::_isWhitespace:
655 return inline_character_compare(intrinsic_id());
656
657 case vmIntrinsics::_min:
658 case vmIntrinsics::_max:
659 case vmIntrinsics::_min_strict:
660 case vmIntrinsics::_max_strict:
661 return inline_min_max(intrinsic_id());
662
663 case vmIntrinsics::_maxF:
664 case vmIntrinsics::_minF:
665 case vmIntrinsics::_maxD:
666 case vmIntrinsics::_minD:
667 case vmIntrinsics::_maxF_strict:
668 case vmIntrinsics::_minF_strict:
669 case vmIntrinsics::_maxD_strict:
670 case vmIntrinsics::_minD_strict:
671 return inline_fp_min_max(intrinsic_id());
672
673 case vmIntrinsics::_VectorUnaryOp:
674 return inline_vector_nary_operation(1);
675 case vmIntrinsics::_VectorBinaryOp:
676 return inline_vector_nary_operation(2);
677 case vmIntrinsics::_VectorTernaryOp:
678 return inline_vector_nary_operation(3);
679 case vmIntrinsics::_VectorFromBitsCoerced:
680 return inline_vector_frombits_coerced();
681 case vmIntrinsics::_VectorShuffleIota:
682 return inline_vector_shuffle_iota();
683 case vmIntrinsics::_VectorMaskOp:
684 return inline_vector_mask_operation();
685 case vmIntrinsics::_VectorShuffleToVector:
686 return inline_vector_shuffle_to_vector();
687 case vmIntrinsics::_VectorLoadOp:
688 return inline_vector_mem_operation(/*is_store=*/false);
689 case vmIntrinsics::_VectorLoadMaskedOp:
690 return inline_vector_mem_masked_operation(/*is_store*/false);
691 case vmIntrinsics::_VectorStoreOp:
692 return inline_vector_mem_operation(/*is_store=*/true);
693 case vmIntrinsics::_VectorStoreMaskedOp:
694 return inline_vector_mem_masked_operation(/*is_store=*/true);
695 case vmIntrinsics::_VectorGatherOp:
696 return inline_vector_gather_scatter(/*is_scatter*/ false);
697 case vmIntrinsics::_VectorScatterOp:
698 return inline_vector_gather_scatter(/*is_scatter*/ true);
699 case vmIntrinsics::_VectorReductionCoerced:
700 return inline_vector_reduction();
701 case vmIntrinsics::_VectorTest:
702 return inline_vector_test();
703 case vmIntrinsics::_VectorBlend:
704 return inline_vector_blend();
705 case vmIntrinsics::_VectorRearrange:
706 return inline_vector_rearrange();
707 case vmIntrinsics::_VectorCompare:
708 return inline_vector_compare();
709 case vmIntrinsics::_VectorBroadcastInt:
710 return inline_vector_broadcast_int();
711 case vmIntrinsics::_VectorConvert:
712 return inline_vector_convert();
713 case vmIntrinsics::_VectorInsert:
714 return inline_vector_insert();
715 case vmIntrinsics::_VectorExtract:
716 return inline_vector_extract();
717 case vmIntrinsics::_VectorCompressExpand:
718 return inline_vector_compress_expand();
719 case vmIntrinsics::_IndexVector:
720 return inline_index_vector();
721
722 case vmIntrinsics::_getObjectSize:
723 return inline_getObjectSize();
724
725 case vmIntrinsics::_blackhole:
726 return inline_blackhole();
727
728 case vmIntrinsics::_shipilev_magic_timestamp:
729 return inline_timestamp(false);
730 case vmIntrinsics::_shipilev_magic_timestamp_serial:
731 return inline_timestamp(true);
732 case vmIntrinsics::_shipilev_magic_sizeOf:
733 return inline_sizeOf();
734 case vmIntrinsics::_shipilev_magic_addressOf:
735 return inline_addressOf();
736
737 default:
738 // If you get here, it may be that someone has added a new intrinsic
739 // to the list in vmIntrinsics.hpp without implementing it here.
740 #ifndef PRODUCT
741 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
742 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
743 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
744 }
745 #endif
746 return false;
747 }
748 }
749
750 Node* LibraryCallKit::try_to_predicate(int predicate) {
751 if (!jvms()->has_method()) {
752 // Root JVMState has a null method.
753 assert(map()->memory()->Opcode() == Op_Parm, "");
754 // Insert the memory aliasing node
755 set_all_memory(reset_memory());
756 }
757 assert(merged_memory(), "");
758
759 switch (intrinsic_id()) {
760 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
761 return inline_cipherBlockChaining_AESCrypt_predicate(false);
762 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
763 return inline_cipherBlockChaining_AESCrypt_predicate(true);
764 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
765 return inline_electronicCodeBook_AESCrypt_predicate(false);
766 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
767 return inline_electronicCodeBook_AESCrypt_predicate(true);
768 case vmIntrinsics::_counterMode_AESCrypt:
769 return inline_counterMode_AESCrypt_predicate();
770 case vmIntrinsics::_digestBase_implCompressMB:
771 return inline_digestBase_implCompressMB_predicate(predicate);
772 case vmIntrinsics::_galoisCounterMode_AESCrypt:
773 return inline_galoisCounterMode_AESCrypt_predicate();
774
775 default:
776 // If you get here, it may be that someone has added a new intrinsic
777 // to the list in vmIntrinsics.hpp without implementing it here.
778 #ifndef PRODUCT
779 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
780 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
781 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
782 }
783 #endif
784 Node* slow_ctl = control();
785 set_control(top()); // No fast path intrinsic
786 return slow_ctl;
787 }
788 }
789
790 //------------------------------set_result-------------------------------
791 // Helper function for finishing intrinsics.
792 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
793 record_for_igvn(region);
794 set_control(_gvn.transform(region));
795 set_result( _gvn.transform(value));
796 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
797 }
798
799 //------------------------------generate_guard---------------------------
800 // Helper function for generating guarded fast-slow graph structures.
801 // The given 'test', if true, guards a slow path. If the test fails
802 // then a fast path can be taken. (We generally hope it fails.)
803 // In all cases, GraphKit::control() is updated to the fast path.
804 // The returned value represents the control for the slow path.
805 // The return value is never 'top'; it is either a valid control
806 // or NULL if it is obvious that the slow path can never be taken.
807 // Also, if region and the slow control are not NULL, the slow edge
808 // is appended to the region.
809 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
810 if (stopped()) {
811 // Already short circuited.
812 return NULL;
813 }
814
815 // Build an if node and its projections.
816 // If test is true we take the slow path, which we assume is uncommon.
817 if (_gvn.type(test) == TypeInt::ZERO) {
818 // The slow branch is never taken. No need to build this guard.
819 return NULL;
820 }
821
822 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
823
824 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
825 if (if_slow == top()) {
826 // The slow branch is never taken. No need to build this guard.
827 return NULL;
828 }
829
830 if (region != NULL)
831 region->add_req(if_slow);
832
833 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
834 set_control(if_fast);
835
836 return if_slow;
837 }
838
839 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
840 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
841 }
842 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
843 return generate_guard(test, region, PROB_FAIR);
844 }
845
846 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
847 Node* *pos_index) {
848 if (stopped())
849 return NULL; // already stopped
850 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
851 return NULL; // index is already adequately typed
852 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
853 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
854 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
855 if (is_neg != NULL && pos_index != NULL) {
856 // Emulate effect of Parse::adjust_map_after_if.
857 Node* ccast = new CastIINode(index, TypeInt::POS);
858 ccast->set_req(0, control());
859 (*pos_index) = _gvn.transform(ccast);
860 }
861 return is_neg;
862 }
863
864 // Make sure that 'position' is a valid limit index, in [0..length].
865 // There are two equivalent plans for checking this:
866 // A. (offset + copyLength) unsigned<= arrayLength
867 // B. offset <= (arrayLength - copyLength)
868 // We require that all of the values above, except for the sum and
869 // difference, are already known to be non-negative.
870 // Plan A is robust in the face of overflow, if offset and copyLength
871 // are both hugely positive.
872 //
873 // Plan B is less direct and intuitive, but it does not overflow at
874 // all, since the difference of two non-negatives is always
875 // representable. Whenever Java methods must perform the equivalent
876 // check they generally use Plan B instead of Plan A.
877 // For the moment we use Plan A.
878 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
879 Node* subseq_length,
880 Node* array_length,
881 RegionNode* region) {
882 if (stopped())
883 return NULL; // already stopped
884 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
885 if (zero_offset && subseq_length->eqv_uncast(array_length))
886 return NULL; // common case of whole-array copy
887 Node* last = subseq_length;
888 if (!zero_offset) // last += offset
889 last = _gvn.transform(new AddINode(last, offset));
890 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
891 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
892 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
893 return is_over;
894 }
895
896 // Emit range checks for the given String.value byte array
897 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
898 if (stopped()) {
899 return; // already stopped
900 }
901 RegionNode* bailout = new RegionNode(1);
902 record_for_igvn(bailout);
903 if (char_count) {
904 // Convert char count to byte count
905 count = _gvn.transform(new LShiftINode(count, intcon(1)));
906 }
907
908 // Offset and count must not be negative
909 generate_negative_guard(offset, bailout);
910 generate_negative_guard(count, bailout);
911 // Offset + count must not exceed length of array
912 generate_limit_guard(offset, count, load_array_length(array), bailout);
913
914 if (bailout->req() > 1) {
915 PreserveJVMState pjvms(this);
916 set_control(_gvn.transform(bailout));
917 uncommon_trap(Deoptimization::Reason_intrinsic,
918 Deoptimization::Action_maybe_recompile);
919 }
920 }
921
922 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
923 bool is_immutable) {
924 ciKlass* thread_klass = env()->Thread_klass();
925 const Type* thread_type
926 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
927
928 Node* thread = _gvn.transform(new ThreadLocalNode());
929 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
930 tls_output = thread;
931
932 Node* thread_obj_handle
933 = (is_immutable
934 ? LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
935 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
936 : make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
937 thread_obj_handle = _gvn.transform(thread_obj_handle);
938
939 DecoratorSet decorators = IN_NATIVE;
940 if (is_immutable) {
941 decorators |= C2_IMMUTABLE_MEMORY;
942 }
943 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
944 }
945
946 //--------------------------generate_current_thread--------------------
947 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
948 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
949 /*is_immutable*/false);
950 }
951
952 //--------------------------generate_virtual_thread--------------------
953 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
954 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
955 !C->method()->changes_current_thread());
956 }
957
958 //------------------------------make_string_method_node------------------------
959 // Helper method for String intrinsic functions. This version is called with
960 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
961 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
962 // containing the lengths of str1 and str2.
963 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
964 Node* result = NULL;
965 switch (opcode) {
966 case Op_StrIndexOf:
967 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
968 str1_start, cnt1, str2_start, cnt2, ae);
969 break;
970 case Op_StrComp:
971 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
972 str1_start, cnt1, str2_start, cnt2, ae);
973 break;
974 case Op_StrEquals:
975 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
976 // Use the constant length if there is one because optimized match rule may exist.
977 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
978 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
979 break;
980 default:
981 ShouldNotReachHere();
982 return NULL;
983 }
984
985 // All these intrinsics have checks.
986 C->set_has_split_ifs(true); // Has chance for split-if optimization
987 clear_upper_avx();
988
989 return _gvn.transform(result);
990 }
991
992 //------------------------------inline_string_compareTo------------------------
993 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
994 Node* arg1 = argument(0);
995 Node* arg2 = argument(1);
996
997 arg1 = must_be_not_null(arg1, true);
998 arg2 = must_be_not_null(arg2, true);
999
1000 // Get start addr and length of first argument
1001 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1002 Node* arg1_cnt = load_array_length(arg1);
1003
1004 // Get start addr and length of second argument
1005 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1006 Node* arg2_cnt = load_array_length(arg2);
1007
1008 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1009 set_result(result);
1010 return true;
1011 }
1012
1013 //------------------------------inline_string_equals------------------------
1014 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1015 Node* arg1 = argument(0);
1016 Node* arg2 = argument(1);
1017
1018 // paths (plus control) merge
1019 RegionNode* region = new RegionNode(3);
1020 Node* phi = new PhiNode(region, TypeInt::BOOL);
1021
1022 if (!stopped()) {
1023
1024 arg1 = must_be_not_null(arg1, true);
1025 arg2 = must_be_not_null(arg2, true);
1026
1027 // Get start addr and length of first argument
1028 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1029 Node* arg1_cnt = load_array_length(arg1);
1030
1031 // Get start addr and length of second argument
1032 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1033 Node* arg2_cnt = load_array_length(arg2);
1034
1035 // Check for arg1_cnt != arg2_cnt
1036 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1037 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1038 Node* if_ne = generate_slow_guard(bol, NULL);
1039 if (if_ne != NULL) {
1040 phi->init_req(2, intcon(0));
1041 region->init_req(2, if_ne);
1042 }
1043
1044 // Check for count == 0 is done by assembler code for StrEquals.
1045
1046 if (!stopped()) {
1047 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1048 phi->init_req(1, equals);
1049 region->init_req(1, control());
1050 }
1051 }
1052
1053 // post merge
1054 set_control(_gvn.transform(region));
1055 record_for_igvn(region);
1056
1057 set_result(_gvn.transform(phi));
1058 return true;
1059 }
1060
1061 //------------------------------inline_array_equals----------------------------
1062 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1063 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1064 Node* arg1 = argument(0);
1065 Node* arg2 = argument(1);
1066
1067 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1068 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1069 clear_upper_avx();
1070
1071 return true;
1072 }
1073
1074 //------------------------------inline_countPositives------------------------------
1075 bool LibraryCallKit::inline_countPositives() {
1076 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1077 return false;
1078 }
1079
1080 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1081 // no receiver since it is static method
1082 Node* ba = argument(0);
1083 Node* offset = argument(1);
1084 Node* len = argument(2);
1085
1086 ba = must_be_not_null(ba, true);
1087
1088 // Range checks
1089 generate_string_range_check(ba, offset, len, false);
1090 if (stopped()) {
1091 return true;
1092 }
1093 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1094 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1095 set_result(_gvn.transform(result));
1096 return true;
1097 }
1098
1099 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1100 Node* index = argument(0);
1101 Node* length = bt == T_INT ? argument(1) : argument(2);
1102 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1103 return false;
1104 }
1105
1106 // check that length is positive
1107 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1108 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1109
1110 {
1111 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1112 uncommon_trap(Deoptimization::Reason_intrinsic,
1113 Deoptimization::Action_make_not_entrant);
1114 }
1115
1116 if (stopped()) {
1117 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1118 return true;
1119 }
1120
1121 // length is now known positive, add a cast node to make this explicit
1122 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1123 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1124 casted_length = _gvn.transform(casted_length);
1125 replace_in_map(length, casted_length);
1126 length = casted_length;
1127
1128 // Use an unsigned comparison for the range check itself
1129 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1130 BoolTest::mask btest = BoolTest::lt;
1131 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1132 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1133 _gvn.set_type(rc, rc->Value(&_gvn));
1134 if (!rc_bool->is_Con()) {
1135 record_for_igvn(rc);
1136 }
1137 set_control(_gvn.transform(new IfTrueNode(rc)));
1138 {
1139 PreserveJVMState pjvms(this);
1140 set_control(_gvn.transform(new IfFalseNode(rc)));
1141 uncommon_trap(Deoptimization::Reason_range_check,
1142 Deoptimization::Action_make_not_entrant);
1143 }
1144
1145 if (stopped()) {
1146 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1147 return true;
1148 }
1149
1150 // index is now known to be >= 0 and < length, cast it
1151 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1152 result = _gvn.transform(result);
1153 set_result(result);
1154 replace_in_map(index, result);
1155 return true;
1156 }
1157
1158 //------------------------------inline_string_indexOf------------------------
1159 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1160 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1161 return false;
1162 }
1163 Node* src = argument(0);
1164 Node* tgt = argument(1);
1165
1166 // Make the merge point
1167 RegionNode* result_rgn = new RegionNode(4);
1168 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1169
1170 src = must_be_not_null(src, true);
1171 tgt = must_be_not_null(tgt, true);
1172
1173 // Get start addr and length of source string
1174 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1175 Node* src_count = load_array_length(src);
1176
1177 // Get start addr and length of substring
1178 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1179 Node* tgt_count = load_array_length(tgt);
1180
1181 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1182 // Divide src size by 2 if String is UTF16 encoded
1183 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1184 }
1185 if (ae == StrIntrinsicNode::UU) {
1186 // Divide substring size by 2 if String is UTF16 encoded
1187 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1188 }
1189
1190 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1191 if (result != NULL) {
1192 result_phi->init_req(3, result);
1193 result_rgn->init_req(3, control());
1194 }
1195 set_control(_gvn.transform(result_rgn));
1196 record_for_igvn(result_rgn);
1197 set_result(_gvn.transform(result_phi));
1198
1199 return true;
1200 }
1201
1202 //-----------------------------inline_string_indexOf-----------------------
1203 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1204 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1205 return false;
1206 }
1207 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1208 return false;
1209 }
1210 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1211 Node* src = argument(0); // byte[]
1212 Node* src_count = argument(1); // char count
1213 Node* tgt = argument(2); // byte[]
1214 Node* tgt_count = argument(3); // char count
1215 Node* from_index = argument(4); // char index
1216
1217 src = must_be_not_null(src, true);
1218 tgt = must_be_not_null(tgt, true);
1219
1220 // Multiply byte array index by 2 if String is UTF16 encoded
1221 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1222 src_count = _gvn.transform(new SubINode(src_count, from_index));
1223 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1224 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1225
1226 // Range checks
1227 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1228 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1229 if (stopped()) {
1230 return true;
1231 }
1232
1233 RegionNode* region = new RegionNode(5);
1234 Node* phi = new PhiNode(region, TypeInt::INT);
1235
1236 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1237 if (result != NULL) {
1238 // The result is index relative to from_index if substring was found, -1 otherwise.
1239 // Generate code which will fold into cmove.
1240 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1241 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1242
1243 Node* if_lt = generate_slow_guard(bol, NULL);
1244 if (if_lt != NULL) {
1245 // result == -1
1246 phi->init_req(3, result);
1247 region->init_req(3, if_lt);
1248 }
1249 if (!stopped()) {
1250 result = _gvn.transform(new AddINode(result, from_index));
1251 phi->init_req(4, result);
1252 region->init_req(4, control());
1253 }
1254 }
1255
1256 set_control(_gvn.transform(region));
1257 record_for_igvn(region);
1258 set_result(_gvn.transform(phi));
1259 clear_upper_avx();
1260
1261 return true;
1262 }
1263
1264 // Create StrIndexOfNode with fast path checks
1265 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1266 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1267 // Check for substr count > string count
1268 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1269 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1270 Node* if_gt = generate_slow_guard(bol, NULL);
1271 if (if_gt != NULL) {
1272 phi->init_req(1, intcon(-1));
1273 region->init_req(1, if_gt);
1274 }
1275 if (!stopped()) {
1276 // Check for substr count == 0
1277 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1278 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1279 Node* if_zero = generate_slow_guard(bol, NULL);
1280 if (if_zero != NULL) {
1281 phi->init_req(2, intcon(0));
1282 region->init_req(2, if_zero);
1283 }
1284 }
1285 if (!stopped()) {
1286 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1287 }
1288 return NULL;
1289 }
1290
1291 //-----------------------------inline_string_indexOfChar-----------------------
1292 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1293 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1294 return false;
1295 }
1296 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1297 return false;
1298 }
1299 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1300 Node* src = argument(0); // byte[]
1301 Node* tgt = argument(1); // tgt is int ch
1302 Node* from_index = argument(2);
1303 Node* max = argument(3);
1304
1305 src = must_be_not_null(src, true);
1306
1307 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1308 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1309 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1310
1311 // Range checks
1312 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1313 if (stopped()) {
1314 return true;
1315 }
1316
1317 RegionNode* region = new RegionNode(3);
1318 Node* phi = new PhiNode(region, TypeInt::INT);
1319
1320 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1321 C->set_has_split_ifs(true); // Has chance for split-if optimization
1322 _gvn.transform(result);
1323
1324 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1325 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1326
1327 Node* if_lt = generate_slow_guard(bol, NULL);
1328 if (if_lt != NULL) {
1329 // result == -1
1330 phi->init_req(2, result);
1331 region->init_req(2, if_lt);
1332 }
1333 if (!stopped()) {
1334 result = _gvn.transform(new AddINode(result, from_index));
1335 phi->init_req(1, result);
1336 region->init_req(1, control());
1337 }
1338 set_control(_gvn.transform(region));
1339 record_for_igvn(region);
1340 set_result(_gvn.transform(phi));
1341
1342 return true;
1343 }
1344 //---------------------------inline_string_copy---------------------
1345 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1346 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1347 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1348 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1349 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1350 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1351 bool LibraryCallKit::inline_string_copy(bool compress) {
1352 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1353 return false;
1354 }
1355 int nargs = 5; // 2 oops, 3 ints
1356 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1357
1358 Node* src = argument(0);
1359 Node* src_offset = argument(1);
1360 Node* dst = argument(2);
1361 Node* dst_offset = argument(3);
1362 Node* length = argument(4);
1363
1364 // Check for allocation before we add nodes that would confuse
1365 // tightly_coupled_allocation()
1366 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1367
1368 // Figure out the size and type of the elements we will be copying.
1369 const Type* src_type = src->Value(&_gvn);
1370 const Type* dst_type = dst->Value(&_gvn);
1371 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
1372 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
1373 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1374 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1375 "Unsupported array types for inline_string_copy");
1376
1377 src = must_be_not_null(src, true);
1378 dst = must_be_not_null(dst, true);
1379
1380 // Convert char[] offsets to byte[] offsets
1381 bool convert_src = (compress && src_elem == T_BYTE);
1382 bool convert_dst = (!compress && dst_elem == T_BYTE);
1383 if (convert_src) {
1384 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1385 } else if (convert_dst) {
1386 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1387 }
1388
1389 // Range checks
1390 generate_string_range_check(src, src_offset, length, convert_src);
1391 generate_string_range_check(dst, dst_offset, length, convert_dst);
1392 if (stopped()) {
1393 return true;
1394 }
1395
1396 Node* src_start = array_element_address(src, src_offset, src_elem);
1397 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1398 // 'src_start' points to src array + scaled offset
1399 // 'dst_start' points to dst array + scaled offset
1400 Node* count = NULL;
1401 if (compress) {
1402 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1403 } else {
1404 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1405 }
1406
1407 if (alloc != NULL) {
1408 if (alloc->maybe_set_complete(&_gvn)) {
1409 // "You break it, you buy it."
1410 InitializeNode* init = alloc->initialization();
1411 assert(init->is_complete(), "we just did this");
1412 init->set_complete_with_arraycopy();
1413 assert(dst->is_CheckCastPP(), "sanity");
1414 assert(dst->in(0)->in(0) == init, "dest pinned");
1415 }
1416 // Do not let stores that initialize this object be reordered with
1417 // a subsequent store that would make this object accessible by
1418 // other threads.
1419 // Record what AllocateNode this StoreStore protects so that
1420 // escape analysis can go from the MemBarStoreStoreNode to the
1421 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1422 // based on the escape status of the AllocateNode.
1423 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1424 }
1425 if (compress) {
1426 set_result(_gvn.transform(count));
1427 }
1428 clear_upper_avx();
1429
1430 return true;
1431 }
1432
1433 #ifdef _LP64
1434 #define XTOP ,top() /*additional argument*/
1435 #else //_LP64
1436 #define XTOP /*no additional argument*/
1437 #endif //_LP64
1438
1439 //------------------------inline_string_toBytesU--------------------------
1440 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1441 bool LibraryCallKit::inline_string_toBytesU() {
1442 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1443 return false;
1444 }
1445 // Get the arguments.
1446 Node* value = argument(0);
1447 Node* offset = argument(1);
1448 Node* length = argument(2);
1449
1450 Node* newcopy = NULL;
1451
1452 // Set the original stack and the reexecute bit for the interpreter to reexecute
1453 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1454 { PreserveReexecuteState preexecs(this);
1455 jvms()->set_should_reexecute(true);
1456
1457 // Check if a null path was taken unconditionally.
1458 value = null_check(value);
1459
1460 RegionNode* bailout = new RegionNode(1);
1461 record_for_igvn(bailout);
1462
1463 // Range checks
1464 generate_negative_guard(offset, bailout);
1465 generate_negative_guard(length, bailout);
1466 generate_limit_guard(offset, length, load_array_length(value), bailout);
1467 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1468 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1469
1470 if (bailout->req() > 1) {
1471 PreserveJVMState pjvms(this);
1472 set_control(_gvn.transform(bailout));
1473 uncommon_trap(Deoptimization::Reason_intrinsic,
1474 Deoptimization::Action_maybe_recompile);
1475 }
1476 if (stopped()) {
1477 return true;
1478 }
1479
1480 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1481 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1482 newcopy = new_array(klass_node, size, 0); // no arguments to push
1483 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1484 guarantee(alloc != NULL, "created above");
1485
1486 // Calculate starting addresses.
1487 Node* src_start = array_element_address(value, offset, T_CHAR);
1488 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1489
1490 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1491 const TypeInt* toffset = gvn().type(offset)->is_int();
1492 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1493
1494 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1495 const char* copyfunc_name = "arraycopy";
1496 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1497 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1498 OptoRuntime::fast_arraycopy_Type(),
1499 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1500 src_start, dst_start, ConvI2X(length) XTOP);
1501 // Do not let reads from the cloned object float above the arraycopy.
1502 if (alloc->maybe_set_complete(&_gvn)) {
1503 // "You break it, you buy it."
1504 InitializeNode* init = alloc->initialization();
1505 assert(init->is_complete(), "we just did this");
1506 init->set_complete_with_arraycopy();
1507 assert(newcopy->is_CheckCastPP(), "sanity");
1508 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1509 }
1510 // Do not let stores that initialize this object be reordered with
1511 // a subsequent store that would make this object accessible by
1512 // other threads.
1513 // Record what AllocateNode this StoreStore protects so that
1514 // escape analysis can go from the MemBarStoreStoreNode to the
1515 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1516 // based on the escape status of the AllocateNode.
1517 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1518 } // original reexecute is set back here
1519
1520 C->set_has_split_ifs(true); // Has chance for split-if optimization
1521 if (!stopped()) {
1522 set_result(newcopy);
1523 }
1524 clear_upper_avx();
1525
1526 return true;
1527 }
1528
1529 //------------------------inline_string_getCharsU--------------------------
1530 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1531 bool LibraryCallKit::inline_string_getCharsU() {
1532 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1533 return false;
1534 }
1535
1536 // Get the arguments.
1537 Node* src = argument(0);
1538 Node* src_begin = argument(1);
1539 Node* src_end = argument(2); // exclusive offset (i < src_end)
1540 Node* dst = argument(3);
1541 Node* dst_begin = argument(4);
1542
1543 // Check for allocation before we add nodes that would confuse
1544 // tightly_coupled_allocation()
1545 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1546
1547 // Check if a null path was taken unconditionally.
1548 src = null_check(src);
1549 dst = null_check(dst);
1550 if (stopped()) {
1551 return true;
1552 }
1553
1554 // Get length and convert char[] offset to byte[] offset
1555 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1556 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1557
1558 // Range checks
1559 generate_string_range_check(src, src_begin, length, true);
1560 generate_string_range_check(dst, dst_begin, length, false);
1561 if (stopped()) {
1562 return true;
1563 }
1564
1565 if (!stopped()) {
1566 // Calculate starting addresses.
1567 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1568 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1569
1570 // Check if array addresses are aligned to HeapWordSize
1571 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1572 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1573 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1574 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1575
1576 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1577 const char* copyfunc_name = "arraycopy";
1578 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1579 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1580 OptoRuntime::fast_arraycopy_Type(),
1581 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1582 src_start, dst_start, ConvI2X(length) XTOP);
1583 // Do not let reads from the cloned object float above the arraycopy.
1584 if (alloc != NULL) {
1585 if (alloc->maybe_set_complete(&_gvn)) {
1586 // "You break it, you buy it."
1587 InitializeNode* init = alloc->initialization();
1588 assert(init->is_complete(), "we just did this");
1589 init->set_complete_with_arraycopy();
1590 assert(dst->is_CheckCastPP(), "sanity");
1591 assert(dst->in(0)->in(0) == init, "dest pinned");
1592 }
1593 // Do not let stores that initialize this object be reordered with
1594 // a subsequent store that would make this object accessible by
1595 // other threads.
1596 // Record what AllocateNode this StoreStore protects so that
1597 // escape analysis can go from the MemBarStoreStoreNode to the
1598 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1599 // based on the escape status of the AllocateNode.
1600 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1601 } else {
1602 insert_mem_bar(Op_MemBarCPUOrder);
1603 }
1604 }
1605
1606 C->set_has_split_ifs(true); // Has chance for split-if optimization
1607 return true;
1608 }
1609
1610 //----------------------inline_string_char_access----------------------------
1611 // Store/Load char to/from byte[] array.
1612 // static void StringUTF16.putChar(byte[] val, int index, int c)
1613 // static char StringUTF16.getChar(byte[] val, int index)
1614 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1615 Node* value = argument(0);
1616 Node* index = argument(1);
1617 Node* ch = is_store ? argument(2) : NULL;
1618
1619 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1620 // correctly requires matched array shapes.
1621 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1622 "sanity: byte[] and char[] bases agree");
1623 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1624 "sanity: byte[] and char[] scales agree");
1625
1626 // Bail when getChar over constants is requested: constant folding would
1627 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1628 // Java method would constant fold nicely instead.
1629 if (!is_store && value->is_Con() && index->is_Con()) {
1630 return false;
1631 }
1632
1633 // Save state and restore on bailout
1634 uint old_sp = sp();
1635 SafePointNode* old_map = clone_map();
1636
1637 value = must_be_not_null(value, true);
1638
1639 Node* adr = array_element_address(value, index, T_CHAR);
1640 if (adr->is_top()) {
1641 set_map(old_map);
1642 set_sp(old_sp);
1643 return false;
1644 }
1645 old_map->destruct(&_gvn);
1646 if (is_store) {
1647 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1648 } else {
1649 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);
1650 set_result(ch);
1651 }
1652 return true;
1653 }
1654
1655 //--------------------------round_double_node--------------------------------
1656 // Round a double node if necessary.
1657 Node* LibraryCallKit::round_double_node(Node* n) {
1658 if (Matcher::strict_fp_requires_explicit_rounding) {
1659 #ifdef IA32
1660 if (UseSSE < 2) {
1661 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1662 }
1663 #else
1664 Unimplemented();
1665 #endif // IA32
1666 }
1667 return n;
1668 }
1669
1670 //------------------------------inline_math-----------------------------------
1671 // public static double Math.abs(double)
1672 // public static double Math.sqrt(double)
1673 // public static double Math.log(double)
1674 // public static double Math.log10(double)
1675 // public static double Math.round(double)
1676 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1677 Node* arg = round_double_node(argument(0));
1678 Node* n = NULL;
1679 switch (id) {
1680 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1681 case vmIntrinsics::_dsqrt:
1682 case vmIntrinsics::_dsqrt_strict:
1683 n = new SqrtDNode(C, control(), arg); break;
1684 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1685 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1686 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1687 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1688 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1689 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1690 default: fatal_unexpected_iid(id); break;
1691 }
1692 set_result(_gvn.transform(n));
1693 return true;
1694 }
1695
1696 //------------------------------inline_math-----------------------------------
1697 // public static float Math.abs(float)
1698 // public static int Math.abs(int)
1699 // public static long Math.abs(long)
1700 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1701 Node* arg = argument(0);
1702 Node* n = NULL;
1703 switch (id) {
1704 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1705 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1706 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1707 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1708 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1709 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1710 default: fatal_unexpected_iid(id); break;
1711 }
1712 set_result(_gvn.transform(n));
1713 return true;
1714 }
1715
1716 //------------------------------runtime_math-----------------------------
1717 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1718 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1719 "must be (DD)D or (D)D type");
1720
1721 // Inputs
1722 Node* a = round_double_node(argument(0));
1723 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1724
1725 const TypePtr* no_memory_effects = NULL;
1726 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1727 no_memory_effects,
1728 a, top(), b, b ? top() : NULL);
1729 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1730 #ifdef ASSERT
1731 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1732 assert(value_top == top(), "second value must be top");
1733 #endif
1734
1735 set_result(value);
1736 return true;
1737 }
1738
1739 //------------------------------inline_math_pow-----------------------------
1740 bool LibraryCallKit::inline_math_pow() {
1741 Node* exp = round_double_node(argument(2));
1742 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1743 if (d != NULL) {
1744 if (d->getd() == 2.0) {
1745 // Special case: pow(x, 2.0) => x * x
1746 Node* base = round_double_node(argument(0));
1747 set_result(_gvn.transform(new MulDNode(base, base)));
1748 return true;
1749 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1750 // Special case: pow(x, 0.5) => sqrt(x)
1751 Node* base = round_double_node(argument(0));
1752 Node* zero = _gvn.zerocon(T_DOUBLE);
1753
1754 RegionNode* region = new RegionNode(3);
1755 Node* phi = new PhiNode(region, Type::DOUBLE);
1756
1757 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1758 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1759 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1760 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1761 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1762
1763 Node* if_pow = generate_slow_guard(test, NULL);
1764 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1765 phi->init_req(1, value_sqrt);
1766 region->init_req(1, control());
1767
1768 if (if_pow != NULL) {
1769 set_control(if_pow);
1770 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1771 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1772 const TypePtr* no_memory_effects = NULL;
1773 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1774 no_memory_effects, base, top(), exp, top());
1775 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1776 #ifdef ASSERT
1777 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1778 assert(value_top == top(), "second value must be top");
1779 #endif
1780 phi->init_req(2, value_pow);
1781 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1782 }
1783
1784 C->set_has_split_ifs(true); // Has chance for split-if optimization
1785 set_control(_gvn.transform(region));
1786 record_for_igvn(region);
1787 set_result(_gvn.transform(phi));
1788
1789 return true;
1790 }
1791 }
1792
1793 return StubRoutines::dpow() != NULL ?
1794 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1795 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1796 }
1797
1798 //------------------------------inline_math_native-----------------------------
1799 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1800 switch (id) {
1801 case vmIntrinsics::_dsin:
1802 return StubRoutines::dsin() != NULL ?
1803 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1804 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1805 case vmIntrinsics::_dcos:
1806 return StubRoutines::dcos() != NULL ?
1807 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1808 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1809 case vmIntrinsics::_dtan:
1810 return StubRoutines::dtan() != NULL ?
1811 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1812 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1813 case vmIntrinsics::_dexp:
1814 return StubRoutines::dexp() != NULL ?
1815 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1816 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1817 case vmIntrinsics::_dlog:
1818 return StubRoutines::dlog() != NULL ?
1819 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1820 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1821 case vmIntrinsics::_dlog10:
1822 return StubRoutines::dlog10() != NULL ?
1823 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1824 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1825
1826 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1827 case vmIntrinsics::_ceil:
1828 case vmIntrinsics::_floor:
1829 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1830
1831 case vmIntrinsics::_dsqrt:
1832 case vmIntrinsics::_dsqrt_strict:
1833 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1834 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1835 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1836 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1837 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1838
1839 case vmIntrinsics::_dpow: return inline_math_pow();
1840 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1841 case vmIntrinsics::_fcopySign: return inline_math(id);
1842 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1843 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1844 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1845
1846 // These intrinsics are not yet correctly implemented
1847 case vmIntrinsics::_datan2:
1848 return false;
1849
1850 default:
1851 fatal_unexpected_iid(id);
1852 return false;
1853 }
1854 }
1855
1856 //----------------------------inline_notify-----------------------------------*
1857 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1858 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1859 address func;
1860 if (id == vmIntrinsics::_notify) {
1861 func = OptoRuntime::monitor_notify_Java();
1862 } else {
1863 func = OptoRuntime::monitor_notifyAll_Java();
1864 }
1865 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1866 make_slow_call_ex(call, env()->Throwable_klass(), false);
1867 return true;
1868 }
1869
1870
1871 //----------------------------inline_min_max-----------------------------------
1872 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1873 set_result(generate_min_max(id, argument(0), argument(1)));
1874 return true;
1875 }
1876
1877 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1878 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1879 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1880 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1881 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1882
1883 {
1884 PreserveJVMState pjvms(this);
1885 PreserveReexecuteState preexecs(this);
1886 jvms()->set_should_reexecute(true);
1887
1888 set_control(slow_path);
1889 set_i_o(i_o());
1890
1891 uncommon_trap(Deoptimization::Reason_intrinsic,
1892 Deoptimization::Action_none);
1893 }
1894
1895 set_control(fast_path);
1896 set_result(math);
1897 }
1898
1899 template <typename OverflowOp>
1900 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1901 typedef typename OverflowOp::MathOp MathOp;
1902
1903 MathOp* mathOp = new MathOp(arg1, arg2);
1904 Node* operation = _gvn.transform( mathOp );
1905 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1906 inline_math_mathExact(operation, ofcheck);
1907 return true;
1908 }
1909
1910 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1911 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1912 }
1913
1914 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1915 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1916 }
1917
1918 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1919 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1920 }
1921
1922 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1923 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1924 }
1925
1926 bool LibraryCallKit::inline_math_negateExactI() {
1927 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1928 }
1929
1930 bool LibraryCallKit::inline_math_negateExactL() {
1931 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1932 }
1933
1934 bool LibraryCallKit::inline_math_multiplyExactI() {
1935 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1936 }
1937
1938 bool LibraryCallKit::inline_math_multiplyExactL() {
1939 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1940 }
1941
1942 bool LibraryCallKit::inline_math_multiplyHigh() {
1943 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1944 return true;
1945 }
1946
1947 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1948 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1949 return true;
1950 }
1951
1952 Node*
1953 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1954 Node* result_val = NULL;
1955 switch (id) {
1956 case vmIntrinsics::_min:
1957 case vmIntrinsics::_min_strict:
1958 result_val = _gvn.transform(new MinINode(x0, y0));
1959 break;
1960 case vmIntrinsics::_max:
1961 case vmIntrinsics::_max_strict:
1962 result_val = _gvn.transform(new MaxINode(x0, y0));
1963 break;
1964 default:
1965 fatal_unexpected_iid(id);
1966 break;
1967 }
1968 return result_val;
1969 }
1970
1971 inline int
1972 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
1973 const TypePtr* base_type = TypePtr::NULL_PTR;
1974 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
1975 if (base_type == NULL) {
1976 // Unknown type.
1977 return Type::AnyPtr;
1978 } else if (base_type == TypePtr::NULL_PTR) {
1979 // Since this is a NULL+long form, we have to switch to a rawptr.
1980 base = _gvn.transform(new CastX2PNode(offset));
1981 offset = MakeConX(0);
1982 return Type::RawPtr;
1983 } else if (base_type->base() == Type::RawPtr) {
1984 return Type::RawPtr;
1985 } else if (base_type->isa_oopptr()) {
1986 // Base is never null => always a heap address.
1987 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
1988 return Type::OopPtr;
1989 }
1990 // Offset is small => always a heap address.
1991 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
1992 if (offset_type != NULL &&
1993 base_type->offset() == 0 && // (should always be?)
1994 offset_type->_lo >= 0 &&
1995 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
1996 return Type::OopPtr;
1997 } else if (type == T_OBJECT) {
1998 // off heap access to an oop doesn't make any sense. Has to be on
1999 // heap.
2000 return Type::OopPtr;
2001 }
2002 // Otherwise, it might either be oop+off or NULL+addr.
2003 return Type::AnyPtr;
2004 } else {
2005 // No information:
2006 return Type::AnyPtr;
2007 }
2008 }
2009
2010 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2011 Node* uncasted_base = base;
2012 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2013 if (kind == Type::RawPtr) {
2014 return basic_plus_adr(top(), uncasted_base, offset);
2015 } else if (kind == Type::AnyPtr) {
2016 assert(base == uncasted_base, "unexpected base change");
2017 if (can_cast) {
2018 if (!_gvn.type(base)->speculative_maybe_null() &&
2019 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2020 // According to profiling, this access is always on
2021 // heap. Casting the base to not null and thus avoiding membars
2022 // around the access should allow better optimizations
2023 Node* null_ctl = top();
2024 base = null_check_oop(base, &null_ctl, true, true, true);
2025 assert(null_ctl->is_top(), "no null control here");
2026 return basic_plus_adr(base, offset);
2027 } else if (_gvn.type(base)->speculative_always_null() &&
2028 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2029 // According to profiling, this access is always off
2030 // heap.
2031 base = null_assert(base);
2032 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2033 offset = MakeConX(0);
2034 return basic_plus_adr(top(), raw_base, offset);
2035 }
2036 }
2037 // We don't know if it's an on heap or off heap access. Fall back
2038 // to raw memory access.
2039 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2040 return basic_plus_adr(top(), raw, offset);
2041 } else {
2042 assert(base == uncasted_base, "unexpected base change");
2043 // We know it's an on heap access so base can't be null
2044 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2045 base = must_be_not_null(base, true);
2046 }
2047 return basic_plus_adr(base, offset);
2048 }
2049 }
2050
2051 //--------------------------inline_number_methods-----------------------------
2052 // inline int Integer.numberOfLeadingZeros(int)
2053 // inline int Long.numberOfLeadingZeros(long)
2054 //
2055 // inline int Integer.numberOfTrailingZeros(int)
2056 // inline int Long.numberOfTrailingZeros(long)
2057 //
2058 // inline int Integer.bitCount(int)
2059 // inline int Long.bitCount(long)
2060 //
2061 // inline char Character.reverseBytes(char)
2062 // inline short Short.reverseBytes(short)
2063 // inline int Integer.reverseBytes(int)
2064 // inline long Long.reverseBytes(long)
2065 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2066 Node* arg = argument(0);
2067 Node* n = NULL;
2068 switch (id) {
2069 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2070 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2071 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2072 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2073 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2074 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2075 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2076 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2077 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2078 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2079 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2080 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2081 default: fatal_unexpected_iid(id); break;
2082 }
2083 set_result(_gvn.transform(n));
2084 return true;
2085 }
2086
2087 //--------------------------inline_bitshuffle_methods-----------------------------
2088 // inline int Integer.compress(int, int)
2089 // inline int Integer.expand(int, int)
2090 // inline long Long.compress(long, long)
2091 // inline long Long.expand(long, long)
2092 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2093 Node* n = NULL;
2094 switch (id) {
2095 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2096 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2097 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2098 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2099 default: fatal_unexpected_iid(id); break;
2100 }
2101 set_result(_gvn.transform(n));
2102 return true;
2103 }
2104
2105 //--------------------------inline_number_methods-----------------------------
2106 // inline int Integer.compareUnsigned(int, int)
2107 // inline int Long.compareUnsigned(long, long)
2108 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2109 Node* arg1 = argument(0);
2110 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2111 Node* n = NULL;
2112 switch (id) {
2113 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2114 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2115 default: fatal_unexpected_iid(id); break;
2116 }
2117 set_result(_gvn.transform(n));
2118 return true;
2119 }
2120
2121 //--------------------------inline_unsigned_divmod_methods-----------------------------
2122 // inline int Integer.divideUnsigned(int, int)
2123 // inline int Integer.remainderUnsigned(int, int)
2124 // inline long Long.divideUnsigned(long, long)
2125 // inline long Long.remainderUnsigned(long, long)
2126 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2127 Node* n = NULL;
2128 switch (id) {
2129 case vmIntrinsics::_divideUnsigned_i: {
2130 zero_check_int(argument(1));
2131 // Compile-time detect of null-exception
2132 if (stopped()) {
2133 return true; // keep the graph constructed so far
2134 }
2135 n = new UDivINode(control(), argument(0), argument(1));
2136 break;
2137 }
2138 case vmIntrinsics::_divideUnsigned_l: {
2139 zero_check_long(argument(2));
2140 // Compile-time detect of null-exception
2141 if (stopped()) {
2142 return true; // keep the graph constructed so far
2143 }
2144 n = new UDivLNode(control(), argument(0), argument(2));
2145 break;
2146 }
2147 case vmIntrinsics::_remainderUnsigned_i: {
2148 zero_check_int(argument(1));
2149 // Compile-time detect of null-exception
2150 if (stopped()) {
2151 return true; // keep the graph constructed so far
2152 }
2153 n = new UModINode(control(), argument(0), argument(1));
2154 break;
2155 }
2156 case vmIntrinsics::_remainderUnsigned_l: {
2157 zero_check_long(argument(2));
2158 // Compile-time detect of null-exception
2159 if (stopped()) {
2160 return true; // keep the graph constructed so far
2161 }
2162 n = new UModLNode(control(), argument(0), argument(2));
2163 break;
2164 }
2165 default: fatal_unexpected_iid(id); break;
2166 }
2167 set_result(_gvn.transform(n));
2168 return true;
2169 }
2170
2171 //----------------------------inline_unsafe_access----------------------------
2172
2173 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2174 // Attempt to infer a sharper value type from the offset and base type.
2175 ciKlass* sharpened_klass = NULL;
2176
2177 // See if it is an instance field, with an object type.
2178 if (alias_type->field() != NULL) {
2179 if (alias_type->field()->type()->is_klass()) {
2180 sharpened_klass = alias_type->field()->type()->as_klass();
2181 }
2182 }
2183
2184 const TypeOopPtr* result = NULL;
2185 // See if it is a narrow oop array.
2186 if (adr_type->isa_aryptr()) {
2187 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2188 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2189 if (elem_type != NULL && elem_type->is_loaded()) {
2190 // Sharpen the value type.
2191 result = elem_type;
2192 }
2193 }
2194 }
2195
2196 // The sharpened class might be unloaded if there is no class loader
2197 // contraint in place.
2198 if (result == NULL && sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2199 // Sharpen the value type.
2200 result = TypeOopPtr::make_from_klass(sharpened_klass);
2201 }
2202 if (result != NULL) {
2203 #ifndef PRODUCT
2204 if (C->print_intrinsics() || C->print_inlining()) {
2205 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2206 tty->print(" sharpened value: "); result->dump(); tty->cr();
2207 }
2208 #endif
2209 }
2210 return result;
2211 }
2212
2213 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2214 switch (kind) {
2215 case Relaxed:
2216 return MO_UNORDERED;
2217 case Opaque:
2218 return MO_RELAXED;
2219 case Acquire:
2220 return MO_ACQUIRE;
2221 case Release:
2222 return MO_RELEASE;
2223 case Volatile:
2224 return MO_SEQ_CST;
2225 default:
2226 ShouldNotReachHere();
2227 return 0;
2228 }
2229 }
2230
2231 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2232 if (callee()->is_static()) return false; // caller must have the capability!
2233 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2234 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2235 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2236 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2237
2238 if (is_reference_type(type)) {
2239 decorators |= ON_UNKNOWN_OOP_REF;
2240 }
2241
2242 if (unaligned) {
2243 decorators |= C2_UNALIGNED;
2244 }
2245
2246 #ifndef PRODUCT
2247 {
2248 ResourceMark rm;
2249 // Check the signatures.
2250 ciSignature* sig = callee()->signature();
2251 #ifdef ASSERT
2252 if (!is_store) {
2253 // Object getReference(Object base, int/long offset), etc.
2254 BasicType rtype = sig->return_type()->basic_type();
2255 assert(rtype == type, "getter must return the expected value");
2256 assert(sig->count() == 2, "oop getter has 2 arguments");
2257 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2258 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2259 } else {
2260 // void putReference(Object base, int/long offset, Object x), etc.
2261 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2262 assert(sig->count() == 3, "oop putter has 3 arguments");
2263 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2264 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2265 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2266 assert(vtype == type, "putter must accept the expected value");
2267 }
2268 #endif // ASSERT
2269 }
2270 #endif //PRODUCT
2271
2272 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2273
2274 Node* receiver = argument(0); // type: oop
2275
2276 // Build address expression.
2277 Node* heap_base_oop = top();
2278
2279 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2280 Node* base = argument(1); // type: oop
2281 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2282 Node* offset = argument(2); // type: long
2283 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2284 // to be plain byte offsets, which are also the same as those accepted
2285 // by oopDesc::field_addr.
2286 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2287 "fieldOffset must be byte-scaled");
2288 // 32-bit machines ignore the high half!
2289 offset = ConvL2X(offset);
2290
2291 // Save state and restore on bailout
2292 uint old_sp = sp();
2293 SafePointNode* old_map = clone_map();
2294
2295 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2296
2297 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2298 if (type != T_OBJECT) {
2299 decorators |= IN_NATIVE; // off-heap primitive access
2300 } else {
2301 set_map(old_map);
2302 set_sp(old_sp);
2303 return false; // off-heap oop accesses are not supported
2304 }
2305 } else {
2306 heap_base_oop = base; // on-heap or mixed access
2307 }
2308
2309 // Can base be NULL? Otherwise, always on-heap access.
2310 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2311
2312 if (!can_access_non_heap) {
2313 decorators |= IN_HEAP;
2314 }
2315
2316 Node* val = is_store ? argument(4) : NULL;
2317
2318 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2319 if (adr_type == TypePtr::NULL_PTR) {
2320 set_map(old_map);
2321 set_sp(old_sp);
2322 return false; // off-heap access with zero address
2323 }
2324
2325 // Try to categorize the address.
2326 Compile::AliasType* alias_type = C->alias_type(adr_type);
2327 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2328
2329 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2330 alias_type->adr_type() == TypeAryPtr::RANGE) {
2331 set_map(old_map);
2332 set_sp(old_sp);
2333 return false; // not supported
2334 }
2335
2336 bool mismatched = false;
2337 BasicType bt = alias_type->basic_type();
2338 if (bt != T_ILLEGAL) {
2339 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2340 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2341 // Alias type doesn't differentiate between byte[] and boolean[]).
2342 // Use address type to get the element type.
2343 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2344 }
2345 if (is_reference_type(bt, true)) {
2346 // accessing an array field with getReference is not a mismatch
2347 bt = T_OBJECT;
2348 }
2349 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2350 // Don't intrinsify mismatched object accesses
2351 set_map(old_map);
2352 set_sp(old_sp);
2353 return false;
2354 }
2355 mismatched = (bt != type);
2356 } else if (alias_type->adr_type()->isa_oopptr()) {
2357 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2358 }
2359
2360 old_map->destruct(&_gvn);
2361 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2362
2363 if (mismatched) {
2364 decorators |= C2_MISMATCHED;
2365 }
2366
2367 // First guess at the value type.
2368 const Type *value_type = Type::get_const_basic_type(type);
2369
2370 // Figure out the memory ordering.
2371 decorators |= mo_decorator_for_access_kind(kind);
2372
2373 if (!is_store && type == T_OBJECT) {
2374 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2375 if (tjp != NULL) {
2376 value_type = tjp;
2377 }
2378 }
2379
2380 receiver = null_check(receiver);
2381 if (stopped()) {
2382 return true;
2383 }
2384 // Heap pointers get a null-check from the interpreter,
2385 // as a courtesy. However, this is not guaranteed by Unsafe,
2386 // and it is not possible to fully distinguish unintended nulls
2387 // from intended ones in this API.
2388
2389 if (!is_store) {
2390 Node* p = NULL;
2391 // Try to constant fold a load from a constant field
2392 ciField* field = alias_type->field();
2393 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2394 // final or stable field
2395 p = make_constant_from_field(field, heap_base_oop);
2396 }
2397
2398 if (p == NULL) { // Could not constant fold the load
2399 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2400 // Normalize the value returned by getBoolean in the following cases
2401 if (type == T_BOOLEAN &&
2402 (mismatched ||
2403 heap_base_oop == top() || // - heap_base_oop is NULL or
2404 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2405 // and the unsafe access is made to large offset
2406 // (i.e., larger than the maximum offset necessary for any
2407 // field access)
2408 ) {
2409 IdealKit ideal = IdealKit(this);
2410 #define __ ideal.
2411 IdealVariable normalized_result(ideal);
2412 __ declarations_done();
2413 __ set(normalized_result, p);
2414 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2415 __ set(normalized_result, ideal.ConI(1));
2416 ideal.end_if();
2417 final_sync(ideal);
2418 p = __ value(normalized_result);
2419 #undef __
2420 }
2421 }
2422 if (type == T_ADDRESS) {
2423 p = gvn().transform(new CastP2XNode(NULL, p));
2424 p = ConvX2UL(p);
2425 }
2426 // The load node has the control of the preceding MemBarCPUOrder. All
2427 // following nodes will have the control of the MemBarCPUOrder inserted at
2428 // the end of this method. So, pushing the load onto the stack at a later
2429 // point is fine.
2430 set_result(p);
2431 } else {
2432 if (bt == T_ADDRESS) {
2433 // Repackage the long as a pointer.
2434 val = ConvL2X(val);
2435 val = gvn().transform(new CastX2PNode(val));
2436 }
2437 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2438 }
2439
2440 return true;
2441 }
2442
2443 //----------------------------inline_unsafe_load_store----------------------------
2444 // This method serves a couple of different customers (depending on LoadStoreKind):
2445 //
2446 // LS_cmp_swap:
2447 //
2448 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2449 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2450 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2451 //
2452 // LS_cmp_swap_weak:
2453 //
2454 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2455 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2456 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2457 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2458 //
2459 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2460 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2461 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2462 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2463 //
2464 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2465 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2466 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2467 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2468 //
2469 // LS_cmp_exchange:
2470 //
2471 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2472 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2473 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2474 //
2475 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2476 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2477 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2478 //
2479 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2480 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2481 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2482 //
2483 // LS_get_add:
2484 //
2485 // int getAndAddInt( Object o, long offset, int delta)
2486 // long getAndAddLong(Object o, long offset, long delta)
2487 //
2488 // LS_get_set:
2489 //
2490 // int getAndSet(Object o, long offset, int newValue)
2491 // long getAndSet(Object o, long offset, long newValue)
2492 // Object getAndSet(Object o, long offset, Object newValue)
2493 //
2494 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2495 // This basic scheme here is the same as inline_unsafe_access, but
2496 // differs in enough details that combining them would make the code
2497 // overly confusing. (This is a true fact! I originally combined
2498 // them, but even I was confused by it!) As much code/comments as
2499 // possible are retained from inline_unsafe_access though to make
2500 // the correspondences clearer. - dl
2501
2502 if (callee()->is_static()) return false; // caller must have the capability!
2503
2504 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2505 decorators |= mo_decorator_for_access_kind(access_kind);
2506
2507 #ifndef PRODUCT
2508 BasicType rtype;
2509 {
2510 ResourceMark rm;
2511 // Check the signatures.
2512 ciSignature* sig = callee()->signature();
2513 rtype = sig->return_type()->basic_type();
2514 switch(kind) {
2515 case LS_get_add:
2516 case LS_get_set: {
2517 // Check the signatures.
2518 #ifdef ASSERT
2519 assert(rtype == type, "get and set must return the expected type");
2520 assert(sig->count() == 3, "get and set has 3 arguments");
2521 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2522 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2523 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2524 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2525 #endif // ASSERT
2526 break;
2527 }
2528 case LS_cmp_swap:
2529 case LS_cmp_swap_weak: {
2530 // Check the signatures.
2531 #ifdef ASSERT
2532 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2533 assert(sig->count() == 4, "CAS has 4 arguments");
2534 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2535 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2536 #endif // ASSERT
2537 break;
2538 }
2539 case LS_cmp_exchange: {
2540 // Check the signatures.
2541 #ifdef ASSERT
2542 assert(rtype == type, "CAS must return the expected type");
2543 assert(sig->count() == 4, "CAS has 4 arguments");
2544 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2545 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2546 #endif // ASSERT
2547 break;
2548 }
2549 default:
2550 ShouldNotReachHere();
2551 }
2552 }
2553 #endif //PRODUCT
2554
2555 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2556
2557 // Get arguments:
2558 Node* receiver = NULL;
2559 Node* base = NULL;
2560 Node* offset = NULL;
2561 Node* oldval = NULL;
2562 Node* newval = NULL;
2563 switch(kind) {
2564 case LS_cmp_swap:
2565 case LS_cmp_swap_weak:
2566 case LS_cmp_exchange: {
2567 const bool two_slot_type = type2size[type] == 2;
2568 receiver = argument(0); // type: oop
2569 base = argument(1); // type: oop
2570 offset = argument(2); // type: long
2571 oldval = argument(4); // type: oop, int, or long
2572 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2573 break;
2574 }
2575 case LS_get_add:
2576 case LS_get_set: {
2577 receiver = argument(0); // type: oop
2578 base = argument(1); // type: oop
2579 offset = argument(2); // type: long
2580 oldval = NULL;
2581 newval = argument(4); // type: oop, int, or long
2582 break;
2583 }
2584 default:
2585 ShouldNotReachHere();
2586 }
2587
2588 // Build field offset expression.
2589 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2590 // to be plain byte offsets, which are also the same as those accepted
2591 // by oopDesc::field_addr.
2592 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2593 // 32-bit machines ignore the high half of long offsets
2594 offset = ConvL2X(offset);
2595 // Save state and restore on bailout
2596 uint old_sp = sp();
2597 SafePointNode* old_map = clone_map();
2598 Node* adr = make_unsafe_address(base, offset,type, false);
2599 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2600
2601 Compile::AliasType* alias_type = C->alias_type(adr_type);
2602 BasicType bt = alias_type->basic_type();
2603 if (bt != T_ILLEGAL &&
2604 (is_reference_type(bt) != (type == T_OBJECT))) {
2605 // Don't intrinsify mismatched object accesses.
2606 set_map(old_map);
2607 set_sp(old_sp);
2608 return false;
2609 }
2610
2611 old_map->destruct(&_gvn);
2612
2613 // For CAS, unlike inline_unsafe_access, there seems no point in
2614 // trying to refine types. Just use the coarse types here.
2615 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2616 const Type *value_type = Type::get_const_basic_type(type);
2617
2618 switch (kind) {
2619 case LS_get_set:
2620 case LS_cmp_exchange: {
2621 if (type == T_OBJECT) {
2622 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2623 if (tjp != NULL) {
2624 value_type = tjp;
2625 }
2626 }
2627 break;
2628 }
2629 case LS_cmp_swap:
2630 case LS_cmp_swap_weak:
2631 case LS_get_add:
2632 break;
2633 default:
2634 ShouldNotReachHere();
2635 }
2636
2637 // Null check receiver.
2638 receiver = null_check(receiver);
2639 if (stopped()) {
2640 return true;
2641 }
2642
2643 int alias_idx = C->get_alias_index(adr_type);
2644
2645 if (is_reference_type(type)) {
2646 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2647
2648 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2649 // could be delayed during Parse (for example, in adjust_map_after_if()).
2650 // Execute transformation here to avoid barrier generation in such case.
2651 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2652 newval = _gvn.makecon(TypePtr::NULL_PTR);
2653
2654 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2655 // Refine the value to a null constant, when it is known to be null
2656 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2657 }
2658 }
2659
2660 Node* result = NULL;
2661 switch (kind) {
2662 case LS_cmp_exchange: {
2663 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2664 oldval, newval, value_type, type, decorators);
2665 break;
2666 }
2667 case LS_cmp_swap_weak:
2668 decorators |= C2_WEAK_CMPXCHG;
2669 case LS_cmp_swap: {
2670 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2671 oldval, newval, value_type, type, decorators);
2672 break;
2673 }
2674 case LS_get_set: {
2675 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2676 newval, value_type, type, decorators);
2677 break;
2678 }
2679 case LS_get_add: {
2680 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2681 newval, value_type, type, decorators);
2682 break;
2683 }
2684 default:
2685 ShouldNotReachHere();
2686 }
2687
2688 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2689 set_result(result);
2690 return true;
2691 }
2692
2693 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2694 // Regardless of form, don't allow previous ld/st to move down,
2695 // then issue acquire, release, or volatile mem_bar.
2696 insert_mem_bar(Op_MemBarCPUOrder);
2697 switch(id) {
2698 case vmIntrinsics::_loadFence:
2699 insert_mem_bar(Op_LoadFence);
2700 return true;
2701 case vmIntrinsics::_storeFence:
2702 insert_mem_bar(Op_StoreFence);
2703 return true;
2704 case vmIntrinsics::_storeStoreFence:
2705 insert_mem_bar(Op_StoreStoreFence);
2706 return true;
2707 case vmIntrinsics::_fullFence:
2708 insert_mem_bar(Op_MemBarVolatile);
2709 return true;
2710 default:
2711 fatal_unexpected_iid(id);
2712 return false;
2713 }
2714 }
2715
2716 bool LibraryCallKit::inline_onspinwait() {
2717 insert_mem_bar(Op_OnSpinWait);
2718 return true;
2719 }
2720
2721 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2722 if (!kls->is_Con()) {
2723 return true;
2724 }
2725 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2726 if (klsptr == NULL) {
2727 return true;
2728 }
2729 ciInstanceKlass* ik = klsptr->instance_klass();
2730 // don't need a guard for a klass that is already initialized
2731 return !ik->is_initialized();
2732 }
2733
2734 //----------------------------inline_unsafe_writeback0-------------------------
2735 // public native void Unsafe.writeback0(long address)
2736 bool LibraryCallKit::inline_unsafe_writeback0() {
2737 if (!Matcher::has_match_rule(Op_CacheWB)) {
2738 return false;
2739 }
2740 #ifndef PRODUCT
2741 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2742 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2743 ciSignature* sig = callee()->signature();
2744 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2745 #endif
2746 null_check_receiver(); // null-check, then ignore
2747 Node *addr = argument(1);
2748 addr = new CastX2PNode(addr);
2749 addr = _gvn.transform(addr);
2750 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2751 flush = _gvn.transform(flush);
2752 set_memory(flush, TypeRawPtr::BOTTOM);
2753 return true;
2754 }
2755
2756 //----------------------------inline_unsafe_writeback0-------------------------
2757 // public native void Unsafe.writeback0(long address)
2758 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2759 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2760 return false;
2761 }
2762 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2763 return false;
2764 }
2765 #ifndef PRODUCT
2766 assert(Matcher::has_match_rule(Op_CacheWB),
2767 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2768 : "found match rule for CacheWBPostSync but not CacheWB"));
2769
2770 #endif
2771 null_check_receiver(); // null-check, then ignore
2772 Node *sync;
2773 if (is_pre) {
2774 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2775 } else {
2776 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2777 }
2778 sync = _gvn.transform(sync);
2779 set_memory(sync, TypeRawPtr::BOTTOM);
2780 return true;
2781 }
2782
2783 //----------------------------inline_unsafe_allocate---------------------------
2784 // public native Object Unsafe.allocateInstance(Class<?> cls);
2785 bool LibraryCallKit::inline_unsafe_allocate() {
2786 if (callee()->is_static()) return false; // caller must have the capability!
2787
2788 null_check_receiver(); // null-check, then ignore
2789 Node* cls = null_check(argument(1));
2790 if (stopped()) return true;
2791
2792 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2793 kls = null_check(kls);
2794 if (stopped()) return true; // argument was like int.class
2795
2796 Node* test = NULL;
2797 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2798 // Note: The argument might still be an illegal value like
2799 // Serializable.class or Object[].class. The runtime will handle it.
2800 // But we must make an explicit check for initialization.
2801 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2802 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2803 // can generate code to load it as unsigned byte.
2804 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2805 Node* bits = intcon(InstanceKlass::fully_initialized);
2806 test = _gvn.transform(new SubINode(inst, bits));
2807 // The 'test' is non-zero if we need to take a slow path.
2808 }
2809
2810 Node* obj = new_instance(kls, test);
2811 set_result(obj);
2812 return true;
2813 }
2814
2815 //------------------------inline_native_time_funcs--------------
2816 // inline code for System.currentTimeMillis() and System.nanoTime()
2817 // these have the same type and signature
2818 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2819 const TypeFunc* tf = OptoRuntime::void_long_Type();
2820 const TypePtr* no_memory_effects = NULL;
2821 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2822 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2823 #ifdef ASSERT
2824 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2825 assert(value_top == top(), "second value must be top");
2826 #endif
2827 set_result(value);
2828 return true;
2829 }
2830
2831 #ifdef JFR_HAVE_INTRINSICS
2832
2833 /**
2834 * if oop->klass != null
2835 * // normal class
2836 * epoch = _epoch_state ? 2 : 1
2837 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2838 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2839 * }
2840 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2841 * else
2842 * // primitive class
2843 * if oop->array_klass != null
2844 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2845 * else
2846 * id = LAST_TYPE_ID + 1 // void class path
2847 * if (!signaled)
2848 * signaled = true
2849 */
2850 bool LibraryCallKit::inline_native_classID() {
2851 Node* cls = argument(0);
2852
2853 IdealKit ideal(this);
2854 #define __ ideal.
2855 IdealVariable result(ideal); __ declarations_done();
2856 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2857 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2858 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2859
2860
2861 __ if_then(kls, BoolTest::ne, null()); {
2862 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2863 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2864
2865 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
2866 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2867 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2868 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2869 mask = _gvn.transform(new OrLNode(mask, epoch));
2870 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2871
2872 float unlikely = PROB_UNLIKELY(0.999);
2873 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2874 sync_kit(ideal);
2875 make_runtime_call(RC_LEAF,
2876 OptoRuntime::class_id_load_barrier_Type(),
2877 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
2878 "class id load barrier",
2879 TypePtr::BOTTOM,
2880 kls);
2881 ideal.sync_kit(this);
2882 } __ end_if();
2883
2884 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
2885 } __ else_(); {
2886 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2887 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
2888 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2889 __ if_then(array_kls, BoolTest::ne, null()); {
2890 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2891 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2892 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
2893 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
2894 } __ else_(); {
2895 // void class case
2896 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
2897 } __ end_if();
2898
2899 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
2900 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
2901 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
2902 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
2903 } __ end_if();
2904 } __ end_if();
2905
2906 final_sync(ideal);
2907 set_result(ideal.value(result));
2908 #undef __
2909 return true;
2910 }
2911
2912 /*
2913 * The intrinsic is a model of this pseudo-code:
2914 *
2915 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
2916 * jobject h_event_writer = tl->java_event_writer();
2917 * if (h_event_writer == NULL) {
2918 * return NULL;
2919 * }
2920 * oop threadObj = Thread::threadObj();
2921 * oop vthread = java_lang_Thread::vthread(threadObj);
2922 * traceid tid;
2923 * bool excluded;
2924 * if (vthread != threadObj) { // i.e. current thread is virtual
2925 * tid = java_lang_Thread::tid(vthread);
2926 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
2927 * excluded = vthread_epoch_raw & excluded_mask;
2928 * if (!excluded) {
2929 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
2930 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
2931 * if (vthread_epoch != current_epoch) {
2932 * write_checkpoint();
2933 * }
2934 * }
2935 * } else {
2936 * tid = java_lang_Thread::tid(threadObj);
2937 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
2938 * excluded = thread_epoch_raw & excluded_mask;
2939 * }
2940 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
2941 * traceid tid_in_event_writer = getField(event_writer, "threadID");
2942 * if (tid_in_event_writer != tid) {
2943 * setField(event_writer, "threadID", tid);
2944 * setField(event_writer, "excluded", excluded);
2945 * }
2946 * return event_writer
2947 */
2948 bool LibraryCallKit::inline_native_getEventWriter() {
2949 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
2950
2951 // Save input memory and i_o state.
2952 Node* input_memory_state = reset_memory();
2953 set_all_memory(input_memory_state);
2954 Node* input_io_state = i_o();
2955
2956 Node* excluded_mask = _gvn.intcon(32768);
2957 Node* epoch_mask = _gvn.intcon(32767);
2958
2959 // TLS
2960 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2961
2962 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
2963 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2964
2965 // Load the eventwriter jobject handle.
2966 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2967
2968 // Null check the jobject handle.
2969 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
2970 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
2971 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
2972
2973 // False path, jobj is null.
2974 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
2975
2976 // True path, jobj is not null.
2977 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
2978
2979 set_control(jobj_is_not_null);
2980
2981 // Load the threadObj for the CarrierThread.
2982 Node* threadObj = generate_current_thread(tls_ptr);
2983
2984 // Load the vthread.
2985 Node* vthread = generate_virtual_thread(tls_ptr);
2986
2987 // If vthread != threadObj, this is a virtual thread.
2988 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
2989 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
2990 IfNode* iff_vthread_not_equal_threadObj =
2991 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
2992
2993 // False branch, fallback to threadObj.
2994 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
2995 set_control(vthread_equal_threadObj);
2996
2997 // Load the tid field from the vthread object.
2998 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
2999
3000 // Load the raw epoch value from the threadObj.
3001 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3002 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3003 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3004
3005 // Mask off the excluded information from the epoch.
3006 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3007
3008 // True branch, this is a virtual thread.
3009 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3010 set_control(vthread_not_equal_threadObj);
3011
3012 // Load the tid field from the vthread object.
3013 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3014
3015 // Load the raw epoch value from the vthread.
3016 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3017 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3018 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3019
3020 // Mask off the excluded information from the epoch.
3021 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3022
3023 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3024 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3025 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3026 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3027
3028 // False branch, vthread is excluded, no need to write epoch info.
3029 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3030
3031 // True branch, vthread is included, update epoch info.
3032 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3033 set_control(included);
3034
3035 // Get epoch value.
3036 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3037
3038 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3039 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3040 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3041
3042 // Compare the epoch in the vthread to the current epoch generation.
3043 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3044 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3045 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3046
3047 // False path, epoch is equal, checkpoint information is valid.
3048 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3049
3050 // True path, epoch is not equal, write a checkpoint for the vthread.
3051 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3052
3053 set_control(epoch_is_not_equal);
3054
3055 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3056 // The call also updates the native thread local thread id and the vthread with the current epoch.
3057 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3058 OptoRuntime::jfr_write_checkpoint_Type(),
3059 StubRoutines::jfr_write_checkpoint(),
3060 "write_checkpoint", TypePtr::BOTTOM);
3061 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3062
3063 // vthread epoch != current epoch
3064 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3065 record_for_igvn(epoch_compare_rgn);
3066 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3067 record_for_igvn(epoch_compare_mem);
3068 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3069 record_for_igvn(epoch_compare_io);
3070
3071 // Update control and phi nodes.
3072 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3073 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3074 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3075 epoch_compare_mem->init_req(_false_path, input_memory_state);
3076 epoch_compare_io->init_req(_true_path, i_o());
3077 epoch_compare_io->init_req(_false_path, input_io_state);
3078
3079 // excluded != true
3080 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3081 record_for_igvn(exclude_compare_rgn);
3082 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3083 record_for_igvn(exclude_compare_mem);
3084 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3085 record_for_igvn(exclude_compare_io);
3086
3087 // Update control and phi nodes.
3088 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3089 exclude_compare_rgn->init_req(_false_path, excluded);
3090 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3091 exclude_compare_mem->init_req(_false_path, input_memory_state);
3092 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3093 exclude_compare_io->init_req(_false_path, input_io_state);
3094
3095 // vthread != threadObj
3096 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3097 record_for_igvn(vthread_compare_rgn);
3098 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3099 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3100 record_for_igvn(vthread_compare_io);
3101 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3102 record_for_igvn(tid);
3103 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3104 record_for_igvn(exclusion);
3105
3106 // Update control and phi nodes.
3107 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3108 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3109 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3110 vthread_compare_mem->init_req(_false_path, input_memory_state);
3111 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3112 vthread_compare_io->init_req(_false_path, input_io_state);
3113 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3114 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3115 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3116 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3117
3118 // Update branch state.
3119 set_control(_gvn.transform(vthread_compare_rgn));
3120 set_all_memory(_gvn.transform(vthread_compare_mem));
3121 set_i_o(_gvn.transform(vthread_compare_io));
3122
3123 // Load the event writer oop by dereferencing the jobject handle.
3124 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3125 assert(klass_EventWriter->is_loaded(), "invariant");
3126 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3127 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3128 const TypeOopPtr* const xtype = aklass->as_instance_type();
3129 Node* event_writer = access_load(jobj, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3130
3131 // Load the current thread id from the event writer object.
3132 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3133 // Get the field offset to, conditionally, store an updated tid value later.
3134 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3135 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3136 // Get the field offset to, conditionally, store an updated exclusion value later.
3137 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3138 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3139
3140 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3141 record_for_igvn(event_writer_tid_compare_rgn);
3142 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3143 record_for_igvn(event_writer_tid_compare_mem);
3144 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3145 record_for_igvn(event_writer_tid_compare_io);
3146
3147 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3148 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3149 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3150 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3151
3152 // False path, tids are the same.
3153 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3154
3155 // True path, tid is not equal, need to update the tid in the event writer.
3156 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3157 record_for_igvn(tid_is_not_equal);
3158
3159 // Store the exclusion state to the event writer.
3160 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3161
3162 // Store the tid to the event writer.
3163 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3164
3165 // Update control and phi nodes.
3166 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3167 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3168 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3169 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3170 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3171 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3172
3173 // Result of top level CFG, Memory, IO and Value.
3174 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3175 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3176 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3177 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3178
3179 // Result control.
3180 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3181 result_rgn->init_req(_false_path, jobj_is_null);
3182
3183 // Result memory.
3184 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3185 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3186
3187 // Result IO.
3188 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3189 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3190
3191 // Result value.
3192 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3193 result_value->init_req(_false_path, null()); // return NULL
3194
3195 // Set output state.
3196 set_control(_gvn.transform(result_rgn));
3197 set_all_memory(_gvn.transform(result_mem));
3198 set_i_o(_gvn.transform(result_io));
3199 set_result(result_rgn, result_value);
3200 return true;
3201 }
3202
3203 /*
3204 * The intrinsic is a model of this pseudo-code:
3205 *
3206 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3207 * if (carrierThread != thread) { // is virtual thread
3208 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3209 * bool excluded = vthread_epoch_raw & excluded_mask;
3210 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3211 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3212 * if (!excluded) {
3213 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3214 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3215 * }
3216 * Atomic::release_store(&tl->_vthread, true);
3217 * return;
3218 * }
3219 * Atomic::release_store(&tl->_vthread, false);
3220 */
3221 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3222 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3223
3224 Node* input_memory_state = reset_memory();
3225 set_all_memory(input_memory_state);
3226
3227 Node* excluded_mask = _gvn.intcon(32768);
3228 Node* epoch_mask = _gvn.intcon(32767);
3229
3230 Node* const carrierThread = generate_current_thread(jt);
3231 // If thread != carrierThread, this is a virtual thread.
3232 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3233 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3234 IfNode* iff_thread_not_equal_carrierThread =
3235 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3236
3237 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3238
3239 // False branch, is carrierThread.
3240 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3241 // Store release
3242 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3243
3244 set_all_memory(input_memory_state);
3245
3246 // True branch, is virtual thread.
3247 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3248 set_control(thread_not_equal_carrierThread);
3249
3250 // Load the raw epoch value from the vthread.
3251 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3252 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3253 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3254
3255 // Mask off the excluded information from the epoch.
3256 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3257
3258 // Load the tid field from the thread.
3259 Node* tid = load_field_from_object(thread, "tid", "J");
3260
3261 // Store the vthread tid to the jfr thread local.
3262 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3263 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3264
3265 // Branch is_excluded to conditionalize updating the epoch .
3266 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3267 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3268 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3269
3270 // True branch, vthread is excluded, no need to write epoch info.
3271 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3272 set_control(excluded);
3273 Node* vthread_is_excluded = _gvn.intcon(1);
3274
3275 // False branch, vthread is included, update epoch info.
3276 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3277 set_control(included);
3278 Node* vthread_is_included = _gvn.intcon(0);
3279
3280 // Get epoch value.
3281 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3282
3283 // Store the vthread epoch to the jfr thread local.
3284 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3285 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3286
3287 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3288 record_for_igvn(excluded_rgn);
3289 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3290 record_for_igvn(excluded_mem);
3291 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3292 record_for_igvn(exclusion);
3293
3294 // Merge the excluded control and memory.
3295 excluded_rgn->init_req(_true_path, excluded);
3296 excluded_rgn->init_req(_false_path, included);
3297 excluded_mem->init_req(_true_path, tid_memory);
3298 excluded_mem->init_req(_false_path, included_memory);
3299 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3300 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3301
3302 // Set intermediate state.
3303 set_control(_gvn.transform(excluded_rgn));
3304 set_all_memory(excluded_mem);
3305
3306 // Store the vthread exclusion state to the jfr thread local.
3307 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3308 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3309
3310 // Store release
3311 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3312
3313 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3314 record_for_igvn(thread_compare_rgn);
3315 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3316 record_for_igvn(thread_compare_mem);
3317 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3318 record_for_igvn(vthread);
3319
3320 // Merge the thread_compare control and memory.
3321 thread_compare_rgn->init_req(_true_path, control());
3322 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3323 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3324 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3325
3326 // Set output state.
3327 set_control(_gvn.transform(thread_compare_rgn));
3328 set_all_memory(_gvn.transform(thread_compare_mem));
3329 }
3330
3331 #endif // JFR_HAVE_INTRINSICS
3332
3333 //------------------------inline_native_currentCarrierThread------------------
3334 bool LibraryCallKit::inline_native_currentCarrierThread() {
3335 Node* junk = NULL;
3336 set_result(generate_current_thread(junk));
3337 return true;
3338 }
3339
3340 //------------------------inline_native_currentThread------------------
3341 bool LibraryCallKit::inline_native_currentThread() {
3342 Node* junk = NULL;
3343 set_result(generate_virtual_thread(junk));
3344 return true;
3345 }
3346
3347 //------------------------inline_native_setVthread------------------
3348 bool LibraryCallKit::inline_native_setCurrentThread() {
3349 assert(C->method()->changes_current_thread(),
3350 "method changes current Thread but is not annotated ChangesCurrentThread");
3351 Node* arr = argument(1);
3352 Node* thread = _gvn.transform(new ThreadLocalNode());
3353 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3354 Node* thread_obj_handle
3355 = make_load(NULL, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3356 thread_obj_handle = _gvn.transform(thread_obj_handle);
3357 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3358 // Stores of oops to native memory not supported yet by BarrierSetC2::store_at_resolved
3359 // access_store_at(NULL, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3360 store_to_memory(control(), thread_obj_handle, arr, T_OBJECT, adr_type, MemNode::unordered);
3361 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3362 return true;
3363 }
3364
3365 Node* LibraryCallKit::extentLocalCache_helper() {
3366 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3367 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3368
3369 bool xk = etype->klass_is_exact();
3370
3371 Node* thread = _gvn.transform(new ThreadLocalNode());
3372 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::extentLocalCache_offset()));
3373 return _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
3374 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3375 }
3376
3377 //------------------------inline_native_extentLocalCache------------------
3378 bool LibraryCallKit::inline_native_extentLocalCache() {
3379 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3380 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3381 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3382
3383 // Because we create the extentLocal cache lazily we have to make the
3384 // type of the result BotPTR.
3385 bool xk = etype->klass_is_exact();
3386 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3387 Node* cache_obj_handle = extentLocalCache_helper();
3388 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3389
3390 return true;
3391 }
3392
3393 //------------------------inline_native_setExtentLocalCache------------------
3394 bool LibraryCallKit::inline_native_setExtentLocalCache() {
3395 Node* arr = argument(0);
3396 Node* cache_obj_handle = extentLocalCache_helper();
3397
3398 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3399 store_to_memory(control(), cache_obj_handle, arr, T_OBJECT, adr_type,
3400 MemNode::unordered);
3401
3402 return true;
3403 }
3404
3405 //---------------------------load_mirror_from_klass----------------------------
3406 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3407 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3408 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3409 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3410 // mirror = ((OopHandle)mirror)->resolve();
3411 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3412 }
3413
3414 //-----------------------load_klass_from_mirror_common-------------------------
3415 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3416 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3417 // and branch to the given path on the region.
3418 // If never_see_null, take an uncommon trap on null, so we can optimistically
3419 // compile for the non-null case.
3420 // If the region is NULL, force never_see_null = true.
3421 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3422 bool never_see_null,
3423 RegionNode* region,
3424 int null_path,
3425 int offset) {
3426 if (region == NULL) never_see_null = true;
3427 Node* p = basic_plus_adr(mirror, offset);
3428 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3429 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3430 Node* null_ctl = top();
3431 kls = null_check_oop(kls, &null_ctl, never_see_null);
3432 if (region != NULL) {
3433 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3434 region->init_req(null_path, null_ctl);
3435 } else {
3436 assert(null_ctl == top(), "no loose ends");
3437 }
3438 return kls;
3439 }
3440
3441 //--------------------(inline_native_Class_query helpers)---------------------
3442 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3443 // Fall through if (mods & mask) == bits, take the guard otherwise.
3444 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3445 // Branch around if the given klass has the given modifier bit set.
3446 // Like generate_guard, adds a new path onto the region.
3447 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3448 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3449 Node* mask = intcon(modifier_mask);
3450 Node* bits = intcon(modifier_bits);
3451 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3452 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3453 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3454 return generate_fair_guard(bol, region);
3455 }
3456 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3457 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3458 }
3459 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3460 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3461 }
3462
3463 //-------------------------inline_native_Class_query-------------------
3464 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3465 const Type* return_type = TypeInt::BOOL;
3466 Node* prim_return_value = top(); // what happens if it's a primitive class?
3467 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3468 bool expect_prim = false; // most of these guys expect to work on refs
3469
3470 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3471
3472 Node* mirror = argument(0);
3473 Node* obj = top();
3474
3475 switch (id) {
3476 case vmIntrinsics::_isInstance:
3477 // nothing is an instance of a primitive type
3478 prim_return_value = intcon(0);
3479 obj = argument(1);
3480 break;
3481 case vmIntrinsics::_getModifiers:
3482 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3483 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3484 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3485 break;
3486 case vmIntrinsics::_isInterface:
3487 prim_return_value = intcon(0);
3488 break;
3489 case vmIntrinsics::_isArray:
3490 prim_return_value = intcon(0);
3491 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3492 break;
3493 case vmIntrinsics::_isPrimitive:
3494 prim_return_value = intcon(1);
3495 expect_prim = true; // obviously
3496 break;
3497 case vmIntrinsics::_isHidden:
3498 prim_return_value = intcon(0);
3499 break;
3500 case vmIntrinsics::_getSuperclass:
3501 prim_return_value = null();
3502 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3503 break;
3504 case vmIntrinsics::_getClassAccessFlags:
3505 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3506 return_type = TypeInt::INT; // not bool! 6297094
3507 break;
3508 default:
3509 fatal_unexpected_iid(id);
3510 break;
3511 }
3512
3513 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3514 if (mirror_con == NULL) return false; // cannot happen?
3515
3516 #ifndef PRODUCT
3517 if (C->print_intrinsics() || C->print_inlining()) {
3518 ciType* k = mirror_con->java_mirror_type();
3519 if (k) {
3520 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3521 k->print_name();
3522 tty->cr();
3523 }
3524 }
3525 #endif
3526
3527 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3528 RegionNode* region = new RegionNode(PATH_LIMIT);
3529 record_for_igvn(region);
3530 PhiNode* phi = new PhiNode(region, return_type);
3531
3532 // The mirror will never be null of Reflection.getClassAccessFlags, however
3533 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3534 // if it is. See bug 4774291.
3535
3536 // For Reflection.getClassAccessFlags(), the null check occurs in
3537 // the wrong place; see inline_unsafe_access(), above, for a similar
3538 // situation.
3539 mirror = null_check(mirror);
3540 // If mirror or obj is dead, only null-path is taken.
3541 if (stopped()) return true;
3542
3543 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3544
3545 // Now load the mirror's klass metaobject, and null-check it.
3546 // Side-effects region with the control path if the klass is null.
3547 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3548 // If kls is null, we have a primitive mirror.
3549 phi->init_req(_prim_path, prim_return_value);
3550 if (stopped()) { set_result(region, phi); return true; }
3551 bool safe_for_replace = (region->in(_prim_path) == top());
3552
3553 Node* p; // handy temp
3554 Node* null_ctl;
3555
3556 // Now that we have the non-null klass, we can perform the real query.
3557 // For constant classes, the query will constant-fold in LoadNode::Value.
3558 Node* query_value = top();
3559 switch (id) {
3560 case vmIntrinsics::_isInstance:
3561 // nothing is an instance of a primitive type
3562 query_value = gen_instanceof(obj, kls, safe_for_replace);
3563 break;
3564
3565 case vmIntrinsics::_getModifiers:
3566 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3567 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3568 break;
3569
3570 case vmIntrinsics::_isInterface:
3571 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3572 if (generate_interface_guard(kls, region) != NULL)
3573 // A guard was added. If the guard is taken, it was an interface.
3574 phi->add_req(intcon(1));
3575 // If we fall through, it's a plain class.
3576 query_value = intcon(0);
3577 break;
3578
3579 case vmIntrinsics::_isArray:
3580 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3581 if (generate_array_guard(kls, region) != NULL)
3582 // A guard was added. If the guard is taken, it was an array.
3583 phi->add_req(intcon(1));
3584 // If we fall through, it's a plain class.
3585 query_value = intcon(0);
3586 break;
3587
3588 case vmIntrinsics::_isPrimitive:
3589 query_value = intcon(0); // "normal" path produces false
3590 break;
3591
3592 case vmIntrinsics::_isHidden:
3593 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3594 if (generate_hidden_class_guard(kls, region) != NULL)
3595 // A guard was added. If the guard is taken, it was an hidden class.
3596 phi->add_req(intcon(1));
3597 // If we fall through, it's a plain class.
3598 query_value = intcon(0);
3599 break;
3600
3601
3602 case vmIntrinsics::_getSuperclass:
3603 // The rules here are somewhat unfortunate, but we can still do better
3604 // with random logic than with a JNI call.
3605 // Interfaces store null or Object as _super, but must report null.
3606 // Arrays store an intermediate super as _super, but must report Object.
3607 // Other types can report the actual _super.
3608 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3609 if (generate_interface_guard(kls, region) != NULL)
3610 // A guard was added. If the guard is taken, it was an interface.
3611 phi->add_req(null());
3612 if (generate_array_guard(kls, region) != NULL)
3613 // A guard was added. If the guard is taken, it was an array.
3614 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3615 // If we fall through, it's a plain class. Get its _super.
3616 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3617 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3618 null_ctl = top();
3619 kls = null_check_oop(kls, &null_ctl);
3620 if (null_ctl != top()) {
3621 // If the guard is taken, Object.superClass is null (both klass and mirror).
3622 region->add_req(null_ctl);
3623 phi ->add_req(null());
3624 }
3625 if (!stopped()) {
3626 query_value = load_mirror_from_klass(kls);
3627 }
3628 break;
3629
3630 case vmIntrinsics::_getClassAccessFlags:
3631 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3632 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3633 break;
3634
3635 default:
3636 fatal_unexpected_iid(id);
3637 break;
3638 }
3639
3640 // Fall-through is the normal case of a query to a real class.
3641 phi->init_req(1, query_value);
3642 region->init_req(1, control());
3643
3644 C->set_has_split_ifs(true); // Has chance for split-if optimization
3645 set_result(region, phi);
3646 return true;
3647 }
3648
3649 //-------------------------inline_Class_cast-------------------
3650 bool LibraryCallKit::inline_Class_cast() {
3651 Node* mirror = argument(0); // Class
3652 Node* obj = argument(1);
3653 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3654 if (mirror_con == NULL) {
3655 return false; // dead path (mirror->is_top()).
3656 }
3657 if (obj == NULL || obj->is_top()) {
3658 return false; // dead path
3659 }
3660 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3661
3662 // First, see if Class.cast() can be folded statically.
3663 // java_mirror_type() returns non-null for compile-time Class constants.
3664 ciType* tm = mirror_con->java_mirror_type();
3665 if (tm != NULL && tm->is_klass() &&
3666 tp != NULL) {
3667 if (!tp->is_loaded()) {
3668 // Don't use intrinsic when class is not loaded.
3669 return false;
3670 } else {
3671 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass()), tp->as_klass_type());
3672 if (static_res == Compile::SSC_always_true) {
3673 // isInstance() is true - fold the code.
3674 set_result(obj);
3675 return true;
3676 } else if (static_res == Compile::SSC_always_false) {
3677 // Don't use intrinsic, have to throw ClassCastException.
3678 // If the reference is null, the non-intrinsic bytecode will
3679 // be optimized appropriately.
3680 return false;
3681 }
3682 }
3683 }
3684
3685 // Bailout intrinsic and do normal inlining if exception path is frequent.
3686 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3687 return false;
3688 }
3689
3690 // Generate dynamic checks.
3691 // Class.cast() is java implementation of _checkcast bytecode.
3692 // Do checkcast (Parse::do_checkcast()) optimizations here.
3693
3694 mirror = null_check(mirror);
3695 // If mirror is dead, only null-path is taken.
3696 if (stopped()) {
3697 return true;
3698 }
3699
3700 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3701 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3702 RegionNode* region = new RegionNode(PATH_LIMIT);
3703 record_for_igvn(region);
3704
3705 // Now load the mirror's klass metaobject, and null-check it.
3706 // If kls is null, we have a primitive mirror and
3707 // nothing is an instance of a primitive type.
3708 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3709
3710 Node* res = top();
3711 if (!stopped()) {
3712 Node* bad_type_ctrl = top();
3713 // Do checkcast optimizations.
3714 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3715 region->init_req(_bad_type_path, bad_type_ctrl);
3716 }
3717 if (region->in(_prim_path) != top() ||
3718 region->in(_bad_type_path) != top()) {
3719 // Let Interpreter throw ClassCastException.
3720 PreserveJVMState pjvms(this);
3721 set_control(_gvn.transform(region));
3722 uncommon_trap(Deoptimization::Reason_intrinsic,
3723 Deoptimization::Action_maybe_recompile);
3724 }
3725 if (!stopped()) {
3726 set_result(res);
3727 }
3728 return true;
3729 }
3730
3731
3732 //--------------------------inline_native_subtype_check------------------------
3733 // This intrinsic takes the JNI calls out of the heart of
3734 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3735 bool LibraryCallKit::inline_native_subtype_check() {
3736 // Pull both arguments off the stack.
3737 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3738 args[0] = argument(0);
3739 args[1] = argument(1);
3740 Node* klasses[2]; // corresponding Klasses: superk, subk
3741 klasses[0] = klasses[1] = top();
3742
3743 enum {
3744 // A full decision tree on {superc is prim, subc is prim}:
3745 _prim_0_path = 1, // {P,N} => false
3746 // {P,P} & superc!=subc => false
3747 _prim_same_path, // {P,P} & superc==subc => true
3748 _prim_1_path, // {N,P} => false
3749 _ref_subtype_path, // {N,N} & subtype check wins => true
3750 _both_ref_path, // {N,N} & subtype check loses => false
3751 PATH_LIMIT
3752 };
3753
3754 RegionNode* region = new RegionNode(PATH_LIMIT);
3755 Node* phi = new PhiNode(region, TypeInt::BOOL);
3756 record_for_igvn(region);
3757
3758 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3759 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3760 int class_klass_offset = java_lang_Class::klass_offset();
3761
3762 // First null-check both mirrors and load each mirror's klass metaobject.
3763 int which_arg;
3764 for (which_arg = 0; which_arg <= 1; which_arg++) {
3765 Node* arg = args[which_arg];
3766 arg = null_check(arg);
3767 if (stopped()) break;
3768 args[which_arg] = arg;
3769
3770 Node* p = basic_plus_adr(arg, class_klass_offset);
3771 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3772 klasses[which_arg] = _gvn.transform(kls);
3773 }
3774
3775 // Having loaded both klasses, test each for null.
3776 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3777 for (which_arg = 0; which_arg <= 1; which_arg++) {
3778 Node* kls = klasses[which_arg];
3779 Node* null_ctl = top();
3780 kls = null_check_oop(kls, &null_ctl, never_see_null);
3781 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3782 region->init_req(prim_path, null_ctl);
3783 if (stopped()) break;
3784 klasses[which_arg] = kls;
3785 }
3786
3787 if (!stopped()) {
3788 // now we have two reference types, in klasses[0..1]
3789 Node* subk = klasses[1]; // the argument to isAssignableFrom
3790 Node* superk = klasses[0]; // the receiver
3791 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3792 // now we have a successful reference subtype check
3793 region->set_req(_ref_subtype_path, control());
3794 }
3795
3796 // If both operands are primitive (both klasses null), then
3797 // we must return true when they are identical primitives.
3798 // It is convenient to test this after the first null klass check.
3799 set_control(region->in(_prim_0_path)); // go back to first null check
3800 if (!stopped()) {
3801 // Since superc is primitive, make a guard for the superc==subc case.
3802 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3803 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3804 generate_guard(bol_eq, region, PROB_FAIR);
3805 if (region->req() == PATH_LIMIT+1) {
3806 // A guard was added. If the added guard is taken, superc==subc.
3807 region->swap_edges(PATH_LIMIT, _prim_same_path);
3808 region->del_req(PATH_LIMIT);
3809 }
3810 region->set_req(_prim_0_path, control()); // Not equal after all.
3811 }
3812
3813 // these are the only paths that produce 'true':
3814 phi->set_req(_prim_same_path, intcon(1));
3815 phi->set_req(_ref_subtype_path, intcon(1));
3816
3817 // pull together the cases:
3818 assert(region->req() == PATH_LIMIT, "sane region");
3819 for (uint i = 1; i < region->req(); i++) {
3820 Node* ctl = region->in(i);
3821 if (ctl == NULL || ctl == top()) {
3822 region->set_req(i, top());
3823 phi ->set_req(i, top());
3824 } else if (phi->in(i) == NULL) {
3825 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3826 }
3827 }
3828
3829 set_control(_gvn.transform(region));
3830 set_result(_gvn.transform(phi));
3831 return true;
3832 }
3833
3834 //---------------------generate_array_guard_common------------------------
3835 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3836 bool obj_array, bool not_array) {
3837
3838 if (stopped()) {
3839 return NULL;
3840 }
3841
3842 // If obj_array/non_array==false/false:
3843 // Branch around if the given klass is in fact an array (either obj or prim).
3844 // If obj_array/non_array==false/true:
3845 // Branch around if the given klass is not an array klass of any kind.
3846 // If obj_array/non_array==true/true:
3847 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3848 // If obj_array/non_array==true/false:
3849 // Branch around if the kls is an oop array (Object[] or subtype)
3850 //
3851 // Like generate_guard, adds a new path onto the region.
3852 jint layout_con = 0;
3853 Node* layout_val = get_layout_helper(kls, layout_con);
3854 if (layout_val == NULL) {
3855 bool query = (obj_array
3856 ? Klass::layout_helper_is_objArray(layout_con)
3857 : Klass::layout_helper_is_array(layout_con));
3858 if (query == not_array) {
3859 return NULL; // never a branch
3860 } else { // always a branch
3861 Node* always_branch = control();
3862 if (region != NULL)
3863 region->add_req(always_branch);
3864 set_control(top());
3865 return always_branch;
3866 }
3867 }
3868 // Now test the correct condition.
3869 jint nval = (obj_array
3870 ? (jint)(Klass::_lh_array_tag_type_value
3871 << Klass::_lh_array_tag_shift)
3872 : Klass::_lh_neutral_value);
3873 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3874 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3875 // invert the test if we are looking for a non-array
3876 if (not_array) btest = BoolTest(btest).negate();
3877 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3878 return generate_fair_guard(bol, region);
3879 }
3880
3881
3882 //-----------------------inline_native_newArray--------------------------
3883 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3884 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3885 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3886 Node* mirror;
3887 Node* count_val;
3888 if (uninitialized) {
3889 mirror = argument(1);
3890 count_val = argument(2);
3891 } else {
3892 mirror = argument(0);
3893 count_val = argument(1);
3894 }
3895
3896 mirror = null_check(mirror);
3897 // If mirror or obj is dead, only null-path is taken.
3898 if (stopped()) return true;
3899
3900 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3901 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3902 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3903 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3904 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3905
3906 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3907 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3908 result_reg, _slow_path);
3909 Node* normal_ctl = control();
3910 Node* no_array_ctl = result_reg->in(_slow_path);
3911
3912 // Generate code for the slow case. We make a call to newArray().
3913 set_control(no_array_ctl);
3914 if (!stopped()) {
3915 // Either the input type is void.class, or else the
3916 // array klass has not yet been cached. Either the
3917 // ensuing call will throw an exception, or else it
3918 // will cache the array klass for next time.
3919 PreserveJVMState pjvms(this);
3920 CallJavaNode* slow_call = NULL;
3921 if (uninitialized) {
3922 // Generate optimized virtual call (holder class 'Unsafe' is final)
3923 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3924 } else {
3925 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3926 }
3927 Node* slow_result = set_results_for_java_call(slow_call);
3928 // this->control() comes from set_results_for_java_call
3929 result_reg->set_req(_slow_path, control());
3930 result_val->set_req(_slow_path, slow_result);
3931 result_io ->set_req(_slow_path, i_o());
3932 result_mem->set_req(_slow_path, reset_memory());
3933 }
3934
3935 set_control(normal_ctl);
3936 if (!stopped()) {
3937 // Normal case: The array type has been cached in the java.lang.Class.
3938 // The following call works fine even if the array type is polymorphic.
3939 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3940 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3941 result_reg->init_req(_normal_path, control());
3942 result_val->init_req(_normal_path, obj);
3943 result_io ->init_req(_normal_path, i_o());
3944 result_mem->init_req(_normal_path, reset_memory());
3945
3946 if (uninitialized) {
3947 // Mark the allocation so that zeroing is skipped
3948 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3949 alloc->maybe_set_complete(&_gvn);
3950 }
3951 }
3952
3953 // Return the combined state.
3954 set_i_o( _gvn.transform(result_io) );
3955 set_all_memory( _gvn.transform(result_mem));
3956
3957 C->set_has_split_ifs(true); // Has chance for split-if optimization
3958 set_result(result_reg, result_val);
3959 return true;
3960 }
3961
3962 //----------------------inline_native_getLength--------------------------
3963 // public static native int java.lang.reflect.Array.getLength(Object array);
3964 bool LibraryCallKit::inline_native_getLength() {
3965 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3966
3967 Node* array = null_check(argument(0));
3968 // If array is dead, only null-path is taken.
3969 if (stopped()) return true;
3970
3971 // Deoptimize if it is a non-array.
3972 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3973
3974 if (non_array != NULL) {
3975 PreserveJVMState pjvms(this);
3976 set_control(non_array);
3977 uncommon_trap(Deoptimization::Reason_intrinsic,
3978 Deoptimization::Action_maybe_recompile);
3979 }
3980
3981 // If control is dead, only non-array-path is taken.
3982 if (stopped()) return true;
3983
3984 // The works fine even if the array type is polymorphic.
3985 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3986 Node* result = load_array_length(array);
3987
3988 C->set_has_split_ifs(true); // Has chance for split-if optimization
3989 set_result(result);
3990 return true;
3991 }
3992
3993 //------------------------inline_array_copyOf----------------------------
3994 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3995 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3996 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3997 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3998
3999 // Get the arguments.
4000 Node* original = argument(0);
4001 Node* start = is_copyOfRange? argument(1): intcon(0);
4002 Node* end = is_copyOfRange? argument(2): argument(1);
4003 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4004
4005 Node* newcopy = NULL;
4006
4007 // Set the original stack and the reexecute bit for the interpreter to reexecute
4008 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4009 { PreserveReexecuteState preexecs(this);
4010 jvms()->set_should_reexecute(true);
4011
4012 array_type_mirror = null_check(array_type_mirror);
4013 original = null_check(original);
4014
4015 // Check if a null path was taken unconditionally.
4016 if (stopped()) return true;
4017
4018 Node* orig_length = load_array_length(original);
4019
4020 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4021 klass_node = null_check(klass_node);
4022
4023 RegionNode* bailout = new RegionNode(1);
4024 record_for_igvn(bailout);
4025
4026 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4027 // Bail out if that is so.
4028 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4029 if (not_objArray != NULL) {
4030 // Improve the klass node's type from the new optimistic assumption:
4031 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4032 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4033 Node* cast = new CastPPNode(klass_node, akls);
4034 cast->init_req(0, control());
4035 klass_node = _gvn.transform(cast);
4036 }
4037
4038 // Bail out if either start or end is negative.
4039 generate_negative_guard(start, bailout, &start);
4040 generate_negative_guard(end, bailout, &end);
4041
4042 Node* length = end;
4043 if (_gvn.type(start) != TypeInt::ZERO) {
4044 length = _gvn.transform(new SubINode(end, start));
4045 }
4046
4047 // Bail out if length is negative.
4048 // Without this the new_array would throw
4049 // NegativeArraySizeException but IllegalArgumentException is what
4050 // should be thrown
4051 generate_negative_guard(length, bailout, &length);
4052
4053 if (bailout->req() > 1) {
4054 PreserveJVMState pjvms(this);
4055 set_control(_gvn.transform(bailout));
4056 uncommon_trap(Deoptimization::Reason_intrinsic,
4057 Deoptimization::Action_maybe_recompile);
4058 }
4059
4060 if (!stopped()) {
4061 // How many elements will we copy from the original?
4062 // The answer is MinI(orig_length - start, length).
4063 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4064 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4065
4066 // Generate a direct call to the right arraycopy function(s).
4067 // We know the copy is disjoint but we might not know if the
4068 // oop stores need checking.
4069 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4070 // This will fail a store-check if x contains any non-nulls.
4071
4072 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4073 // loads/stores but it is legal only if we're sure the
4074 // Arrays.copyOf would succeed. So we need all input arguments
4075 // to the copyOf to be validated, including that the copy to the
4076 // new array won't trigger an ArrayStoreException. That subtype
4077 // check can be optimized if we know something on the type of
4078 // the input array from type speculation.
4079 if (_gvn.type(klass_node)->singleton()) {
4080 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4081 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4082
4083 int test = C->static_subtype_check(superk, subk);
4084 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4085 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4086 if (t_original->speculative_type() != NULL) {
4087 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4088 }
4089 }
4090 }
4091
4092 bool validated = false;
4093 // Reason_class_check rather than Reason_intrinsic because we
4094 // want to intrinsify even if this traps.
4095 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4096 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4097
4098 if (not_subtype_ctrl != top()) {
4099 PreserveJVMState pjvms(this);
4100 set_control(not_subtype_ctrl);
4101 uncommon_trap(Deoptimization::Reason_class_check,
4102 Deoptimization::Action_make_not_entrant);
4103 assert(stopped(), "Should be stopped");
4104 }
4105 validated = true;
4106 }
4107
4108 if (!stopped()) {
4109 newcopy = new_array(klass_node, length, 0); // no arguments to push
4110
4111 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4112 load_object_klass(original), klass_node);
4113 if (!is_copyOfRange) {
4114 ac->set_copyof(validated);
4115 } else {
4116 ac->set_copyofrange(validated);
4117 }
4118 Node* n = _gvn.transform(ac);
4119 if (n == ac) {
4120 ac->connect_outputs(this);
4121 } else {
4122 assert(validated, "shouldn't transform if all arguments not validated");
4123 set_all_memory(n);
4124 }
4125 }
4126 }
4127 } // original reexecute is set back here
4128
4129 C->set_has_split_ifs(true); // Has chance for split-if optimization
4130 if (!stopped()) {
4131 set_result(newcopy);
4132 }
4133 return true;
4134 }
4135
4136
4137 //----------------------generate_virtual_guard---------------------------
4138 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4139 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4140 RegionNode* slow_region) {
4141 ciMethod* method = callee();
4142 int vtable_index = method->vtable_index();
4143 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4144 "bad index %d", vtable_index);
4145 // Get the Method* out of the appropriate vtable entry.
4146 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4147 vtable_index*vtableEntry::size_in_bytes() +
4148 vtableEntry::method_offset_in_bytes();
4149 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4150 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4151
4152 // Compare the target method with the expected method (e.g., Object.hashCode).
4153 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4154
4155 Node* native_call = makecon(native_call_addr);
4156 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4157 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4158
4159 return generate_slow_guard(test_native, slow_region);
4160 }
4161
4162 //-----------------------generate_method_call----------------------------
4163 // Use generate_method_call to make a slow-call to the real
4164 // method if the fast path fails. An alternative would be to
4165 // use a stub like OptoRuntime::slow_arraycopy_Java.
4166 // This only works for expanding the current library call,
4167 // not another intrinsic. (E.g., don't use this for making an
4168 // arraycopy call inside of the copyOf intrinsic.)
4169 CallJavaNode*
4170 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4171 // When compiling the intrinsic method itself, do not use this technique.
4172 guarantee(callee() != C->method(), "cannot make slow-call to self");
4173
4174 ciMethod* method = callee();
4175 // ensure the JVMS we have will be correct for this call
4176 guarantee(method_id == method->intrinsic_id(), "must match");
4177
4178 const TypeFunc* tf = TypeFunc::make(method);
4179 CallJavaNode* slow_call;
4180 if (is_static) {
4181 assert(!is_virtual, "");
4182 slow_call = new CallStaticJavaNode(C, tf,
4183 SharedRuntime::get_resolve_static_call_stub(), method);
4184 } else if (is_virtual) {
4185 null_check_receiver();
4186 int vtable_index = Method::invalid_vtable_index;
4187 if (UseInlineCaches) {
4188 // Suppress the vtable call
4189 } else {
4190 // hashCode and clone are not a miranda methods,
4191 // so the vtable index is fixed.
4192 // No need to use the linkResolver to get it.
4193 vtable_index = method->vtable_index();
4194 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4195 "bad index %d", vtable_index);
4196 }
4197 slow_call = new CallDynamicJavaNode(tf,
4198 SharedRuntime::get_resolve_virtual_call_stub(),
4199 method, vtable_index);
4200 } else { // neither virtual nor static: opt_virtual
4201 null_check_receiver();
4202 slow_call = new CallStaticJavaNode(C, tf,
4203 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4204 slow_call->set_optimized_virtual(true);
4205 }
4206 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4207 // To be able to issue a direct call (optimized virtual or virtual)
4208 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4209 // about the method being invoked should be attached to the call site to
4210 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4211 slow_call->set_override_symbolic_info(true);
4212 }
4213 set_arguments_for_java_call(slow_call);
4214 set_edges_for_java_call(slow_call);
4215 return slow_call;
4216 }
4217
4218
4219 /**
4220 * Build special case code for calls to hashCode on an object. This call may
4221 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4222 * slightly different code.
4223 */
4224 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4225 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4226 assert(!(is_virtual && is_static), "either virtual, special, or static");
4227
4228 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4229
4230 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4231 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4232 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4233 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4234 Node* obj = NULL;
4235 if (!is_static) {
4236 // Check for hashing null object
4237 obj = null_check_receiver();
4238 if (stopped()) return true; // unconditionally null
4239 result_reg->init_req(_null_path, top());
4240 result_val->init_req(_null_path, top());
4241 } else {
4242 // Do a null check, and return zero if null.
4243 // System.identityHashCode(null) == 0
4244 obj = argument(0);
4245 Node* null_ctl = top();
4246 obj = null_check_oop(obj, &null_ctl);
4247 result_reg->init_req(_null_path, null_ctl);
4248 result_val->init_req(_null_path, _gvn.intcon(0));
4249 }
4250
4251 // Unconditionally null? Then return right away.
4252 if (stopped()) {
4253 set_control( result_reg->in(_null_path));
4254 if (!stopped())
4255 set_result(result_val->in(_null_path));
4256 return true;
4257 }
4258
4259 // We only go to the fast case code if we pass a number of guards. The
4260 // paths which do not pass are accumulated in the slow_region.
4261 RegionNode* slow_region = new RegionNode(1);
4262 record_for_igvn(slow_region);
4263
4264 // If this is a virtual call, we generate a funny guard. We pull out
4265 // the vtable entry corresponding to hashCode() from the target object.
4266 // If the target method which we are calling happens to be the native
4267 // Object hashCode() method, we pass the guard. We do not need this
4268 // guard for non-virtual calls -- the caller is known to be the native
4269 // Object hashCode().
4270 if (is_virtual) {
4271 // After null check, get the object's klass.
4272 Node* obj_klass = load_object_klass(obj);
4273 generate_virtual_guard(obj_klass, slow_region);
4274 }
4275
4276 // Get the header out of the object, use LoadMarkNode when available
4277 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4278 // The control of the load must be NULL. Otherwise, the load can move before
4279 // the null check after castPP removal.
4280 Node* no_ctrl = NULL;
4281 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4282
4283 // Test the header to see if it is unlocked.
4284 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4285 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4286 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4287 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4288 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4289
4290 generate_slow_guard(test_unlocked, slow_region);
4291
4292 // Get the hash value and check to see that it has been properly assigned.
4293 // We depend on hash_mask being at most 32 bits and avoid the use of
4294 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4295 // vm: see markWord.hpp.
4296 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4297 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4298 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4299 // This hack lets the hash bits live anywhere in the mark object now, as long
4300 // as the shift drops the relevant bits into the low 32 bits. Note that
4301 // Java spec says that HashCode is an int so there's no point in capturing
4302 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4303 hshifted_header = ConvX2I(hshifted_header);
4304 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4305
4306 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4307 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4308 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4309
4310 generate_slow_guard(test_assigned, slow_region);
4311
4312 Node* init_mem = reset_memory();
4313 // fill in the rest of the null path:
4314 result_io ->init_req(_null_path, i_o());
4315 result_mem->init_req(_null_path, init_mem);
4316
4317 result_val->init_req(_fast_path, hash_val);
4318 result_reg->init_req(_fast_path, control());
4319 result_io ->init_req(_fast_path, i_o());
4320 result_mem->init_req(_fast_path, init_mem);
4321
4322 // Generate code for the slow case. We make a call to hashCode().
4323 set_control(_gvn.transform(slow_region));
4324 if (!stopped()) {
4325 // No need for PreserveJVMState, because we're using up the present state.
4326 set_all_memory(init_mem);
4327 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4328 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4329 Node* slow_result = set_results_for_java_call(slow_call);
4330 // this->control() comes from set_results_for_java_call
4331 result_reg->init_req(_slow_path, control());
4332 result_val->init_req(_slow_path, slow_result);
4333 result_io ->set_req(_slow_path, i_o());
4334 result_mem ->set_req(_slow_path, reset_memory());
4335 }
4336
4337 // Return the combined state.
4338 set_i_o( _gvn.transform(result_io) );
4339 set_all_memory( _gvn.transform(result_mem));
4340
4341 set_result(result_reg, result_val);
4342 return true;
4343 }
4344
4345 //---------------------------inline_native_getClass----------------------------
4346 // public final native Class<?> java.lang.Object.getClass();
4347 //
4348 // Build special case code for calls to getClass on an object.
4349 bool LibraryCallKit::inline_native_getClass() {
4350 Node* obj = null_check_receiver();
4351 if (stopped()) return true;
4352 set_result(load_mirror_from_klass(load_object_klass(obj)));
4353 return true;
4354 }
4355
4356 //-----------------inline_native_Reflection_getCallerClass---------------------
4357 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4358 //
4359 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4360 //
4361 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4362 // in that it must skip particular security frames and checks for
4363 // caller sensitive methods.
4364 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4365 #ifndef PRODUCT
4366 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4367 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4368 }
4369 #endif
4370
4371 if (!jvms()->has_method()) {
4372 #ifndef PRODUCT
4373 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4374 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4375 }
4376 #endif
4377 return false;
4378 }
4379
4380 // Walk back up the JVM state to find the caller at the required
4381 // depth.
4382 JVMState* caller_jvms = jvms();
4383
4384 // Cf. JVM_GetCallerClass
4385 // NOTE: Start the loop at depth 1 because the current JVM state does
4386 // not include the Reflection.getCallerClass() frame.
4387 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4388 ciMethod* m = caller_jvms->method();
4389 switch (n) {
4390 case 0:
4391 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4392 break;
4393 case 1:
4394 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4395 if (!m->caller_sensitive()) {
4396 #ifndef PRODUCT
4397 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4398 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4399 }
4400 #endif
4401 return false; // bail-out; let JVM_GetCallerClass do the work
4402 }
4403 break;
4404 default:
4405 if (!m->is_ignored_by_security_stack_walk()) {
4406 // We have reached the desired frame; return the holder class.
4407 // Acquire method holder as java.lang.Class and push as constant.
4408 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4409 ciInstance* caller_mirror = caller_klass->java_mirror();
4410 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4411
4412 #ifndef PRODUCT
4413 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4414 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());
4415 tty->print_cr(" JVM state at this point:");
4416 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4417 ciMethod* m = jvms()->of_depth(i)->method();
4418 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4419 }
4420 }
4421 #endif
4422 return true;
4423 }
4424 break;
4425 }
4426 }
4427
4428 #ifndef PRODUCT
4429 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4430 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4431 tty->print_cr(" JVM state at this point:");
4432 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4433 ciMethod* m = jvms()->of_depth(i)->method();
4434 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4435 }
4436 }
4437 #endif
4438
4439 return false; // bail-out; let JVM_GetCallerClass do the work
4440 }
4441
4442 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4443 Node* arg = argument(0);
4444 Node* result = NULL;
4445
4446 switch (id) {
4447 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4448 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4449 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4450 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4451 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4452 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4453
4454 case vmIntrinsics::_doubleToLongBits: {
4455 // two paths (plus control) merge in a wood
4456 RegionNode *r = new RegionNode(3);
4457 Node *phi = new PhiNode(r, TypeLong::LONG);
4458
4459 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4460 // Build the boolean node
4461 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4462
4463 // Branch either way.
4464 // NaN case is less traveled, which makes all the difference.
4465 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4466 Node *opt_isnan = _gvn.transform(ifisnan);
4467 assert( opt_isnan->is_If(), "Expect an IfNode");
4468 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4469 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4470
4471 set_control(iftrue);
4472
4473 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4474 Node *slow_result = longcon(nan_bits); // return NaN
4475 phi->init_req(1, _gvn.transform( slow_result ));
4476 r->init_req(1, iftrue);
4477
4478 // Else fall through
4479 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4480 set_control(iffalse);
4481
4482 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4483 r->init_req(2, iffalse);
4484
4485 // Post merge
4486 set_control(_gvn.transform(r));
4487 record_for_igvn(r);
4488
4489 C->set_has_split_ifs(true); // Has chance for split-if optimization
4490 result = phi;
4491 assert(result->bottom_type()->isa_long(), "must be");
4492 break;
4493 }
4494
4495 case vmIntrinsics::_floatToIntBits: {
4496 // two paths (plus control) merge in a wood
4497 RegionNode *r = new RegionNode(3);
4498 Node *phi = new PhiNode(r, TypeInt::INT);
4499
4500 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4501 // Build the boolean node
4502 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4503
4504 // Branch either way.
4505 // NaN case is less traveled, which makes all the difference.
4506 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4507 Node *opt_isnan = _gvn.transform(ifisnan);
4508 assert( opt_isnan->is_If(), "Expect an IfNode");
4509 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4510 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4511
4512 set_control(iftrue);
4513
4514 static const jint nan_bits = 0x7fc00000;
4515 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4516 phi->init_req(1, _gvn.transform( slow_result ));
4517 r->init_req(1, iftrue);
4518
4519 // Else fall through
4520 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4521 set_control(iffalse);
4522
4523 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4524 r->init_req(2, iffalse);
4525
4526 // Post merge
4527 set_control(_gvn.transform(r));
4528 record_for_igvn(r);
4529
4530 C->set_has_split_ifs(true); // Has chance for split-if optimization
4531 result = phi;
4532 assert(result->bottom_type()->isa_int(), "must be");
4533 break;
4534 }
4535
4536 default:
4537 fatal_unexpected_iid(id);
4538 break;
4539 }
4540 set_result(_gvn.transform(result));
4541 return true;
4542 }
4543
4544 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4545 Node* arg = argument(0);
4546 Node* result = NULL;
4547
4548 switch (id) {
4549 case vmIntrinsics::_floatIsInfinite:
4550 result = new IsInfiniteFNode(arg);
4551 break;
4552 case vmIntrinsics::_floatIsFinite:
4553 result = new IsFiniteFNode(arg);
4554 break;
4555 case vmIntrinsics::_doubleIsInfinite:
4556 result = new IsInfiniteDNode(arg);
4557 break;
4558 case vmIntrinsics::_doubleIsFinite:
4559 result = new IsFiniteDNode(arg);
4560 break;
4561 default:
4562 fatal_unexpected_iid(id);
4563 break;
4564 }
4565 set_result(_gvn.transform(result));
4566 return true;
4567 }
4568
4569 //----------------------inline_unsafe_copyMemory-------------------------
4570 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4571
4572 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4573 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4574 const Type* base_t = gvn.type(base);
4575
4576 bool in_native = (base_t == TypePtr::NULL_PTR);
4577 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4578 bool is_mixed = !in_heap && !in_native;
4579
4580 if (is_mixed) {
4581 return true; // mixed accesses can touch both on-heap and off-heap memory
4582 }
4583 if (in_heap) {
4584 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4585 if (!is_prim_array) {
4586 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4587 // there's not enough type information available to determine proper memory slice for it.
4588 return true;
4589 }
4590 }
4591 return false;
4592 }
4593
4594 bool LibraryCallKit::inline_unsafe_copyMemory() {
4595 if (callee()->is_static()) return false; // caller must have the capability!
4596 null_check_receiver(); // null-check receiver
4597 if (stopped()) return true;
4598
4599 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4600
4601 Node* src_base = argument(1); // type: oop
4602 Node* src_off = ConvL2X(argument(2)); // type: long
4603 Node* dst_base = argument(4); // type: oop
4604 Node* dst_off = ConvL2X(argument(5)); // type: long
4605 Node* size = ConvL2X(argument(7)); // type: long
4606
4607 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4608 "fieldOffset must be byte-scaled");
4609
4610 Node* src_addr = make_unsafe_address(src_base, src_off);
4611 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4612
4613 Node* thread = _gvn.transform(new ThreadLocalNode());
4614 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4615 BasicType doing_unsafe_access_bt = T_BYTE;
4616 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4617
4618 // update volatile field
4619 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4620
4621 int flags = RC_LEAF | RC_NO_FP;
4622
4623 const TypePtr* dst_type = TypePtr::BOTTOM;
4624
4625 // Adjust memory effects of the runtime call based on input values.
4626 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4627 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4628 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4629
4630 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4631 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4632 flags |= RC_NARROW_MEM; // narrow in memory
4633 }
4634 }
4635
4636 // Call it. Note that the length argument is not scaled.
4637 make_runtime_call(flags,
4638 OptoRuntime::fast_arraycopy_Type(),
4639 StubRoutines::unsafe_arraycopy(),
4640 "unsafe_arraycopy",
4641 dst_type,
4642 src_addr, dst_addr, size XTOP);
4643
4644 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4645
4646 return true;
4647 }
4648
4649 #undef XTOP
4650
4651 //------------------------clone_coping-----------------------------------
4652 // Helper function for inline_native_clone.
4653 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4654 assert(obj_size != NULL, "");
4655 Node* raw_obj = alloc_obj->in(1);
4656 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4657
4658 AllocateNode* alloc = NULL;
4659 if (ReduceBulkZeroing) {
4660 // We will be completely responsible for initializing this object -
4661 // mark Initialize node as complete.
4662 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4663 // The object was just allocated - there should be no any stores!
4664 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4665 // Mark as complete_with_arraycopy so that on AllocateNode
4666 // expansion, we know this AllocateNode is initialized by an array
4667 // copy and a StoreStore barrier exists after the array copy.
4668 alloc->initialization()->set_complete_with_arraycopy();
4669 }
4670
4671 Node* size = _gvn.transform(obj_size);
4672 access_clone(obj, alloc_obj, size, is_array);
4673
4674 // Do not let reads from the cloned object float above the arraycopy.
4675 if (alloc != NULL) {
4676 // Do not let stores that initialize this object be reordered with
4677 // a subsequent store that would make this object accessible by
4678 // other threads.
4679 // Record what AllocateNode this StoreStore protects so that
4680 // escape analysis can go from the MemBarStoreStoreNode to the
4681 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4682 // based on the escape status of the AllocateNode.
4683 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4684 } else {
4685 insert_mem_bar(Op_MemBarCPUOrder);
4686 }
4687 }
4688
4689 //------------------------inline_native_clone----------------------------
4690 // protected native Object java.lang.Object.clone();
4691 //
4692 // Here are the simple edge cases:
4693 // null receiver => normal trap
4694 // virtual and clone was overridden => slow path to out-of-line clone
4695 // not cloneable or finalizer => slow path to out-of-line Object.clone
4696 //
4697 // The general case has two steps, allocation and copying.
4698 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4699 //
4700 // Copying also has two cases, oop arrays and everything else.
4701 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4702 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4703 //
4704 // These steps fold up nicely if and when the cloned object's klass
4705 // can be sharply typed as an object array, a type array, or an instance.
4706 //
4707 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4708 PhiNode* result_val;
4709
4710 // Set the reexecute bit for the interpreter to reexecute
4711 // the bytecode that invokes Object.clone if deoptimization happens.
4712 { PreserveReexecuteState preexecs(this);
4713 jvms()->set_should_reexecute(true);
4714
4715 Node* obj = null_check_receiver();
4716 if (stopped()) return true;
4717
4718 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4719
4720 // If we are going to clone an instance, we need its exact type to
4721 // know the number and types of fields to convert the clone to
4722 // loads/stores. Maybe a speculative type can help us.
4723 if (!obj_type->klass_is_exact() &&
4724 obj_type->speculative_type() != NULL &&
4725 obj_type->speculative_type()->is_instance_klass()) {
4726 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4727 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4728 !spec_ik->has_injected_fields()) {
4729 if (!obj_type->isa_instptr() ||
4730 obj_type->is_instptr()->instance_klass()->has_subklass()) {
4731 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4732 }
4733 }
4734 }
4735
4736 // Conservatively insert a memory barrier on all memory slices.
4737 // Do not let writes into the original float below the clone.
4738 insert_mem_bar(Op_MemBarCPUOrder);
4739
4740 // paths into result_reg:
4741 enum {
4742 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4743 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4744 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4745 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4746 PATH_LIMIT
4747 };
4748 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4749 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4750 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4751 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4752 record_for_igvn(result_reg);
4753
4754 Node* obj_klass = load_object_klass(obj);
4755 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4756 if (array_ctl != NULL) {
4757 // It's an array.
4758 PreserveJVMState pjvms(this);
4759 set_control(array_ctl);
4760 Node* obj_length = load_array_length(obj);
4761 Node* obj_size = NULL;
4762 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4763
4764 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4765 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4766 // If it is an oop array, it requires very special treatment,
4767 // because gc barriers are required when accessing the array.
4768 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4769 if (is_obja != NULL) {
4770 PreserveJVMState pjvms2(this);
4771 set_control(is_obja);
4772 // Generate a direct call to the right arraycopy function(s).
4773 // Clones are always tightly coupled.
4774 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4775 ac->set_clone_oop_array();
4776 Node* n = _gvn.transform(ac);
4777 assert(n == ac, "cannot disappear");
4778 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4779
4780 result_reg->init_req(_objArray_path, control());
4781 result_val->init_req(_objArray_path, alloc_obj);
4782 result_i_o ->set_req(_objArray_path, i_o());
4783 result_mem ->set_req(_objArray_path, reset_memory());
4784 }
4785 }
4786 // Otherwise, there are no barriers to worry about.
4787 // (We can dispense with card marks if we know the allocation
4788 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4789 // causes the non-eden paths to take compensating steps to
4790 // simulate a fresh allocation, so that no further
4791 // card marks are required in compiled code to initialize
4792 // the object.)
4793
4794 if (!stopped()) {
4795 copy_to_clone(obj, alloc_obj, obj_size, true);
4796
4797 // Present the results of the copy.
4798 result_reg->init_req(_array_path, control());
4799 result_val->init_req(_array_path, alloc_obj);
4800 result_i_o ->set_req(_array_path, i_o());
4801 result_mem ->set_req(_array_path, reset_memory());
4802 }
4803 }
4804
4805 // We only go to the instance fast case code if we pass a number of guards.
4806 // The paths which do not pass are accumulated in the slow_region.
4807 RegionNode* slow_region = new RegionNode(1);
4808 record_for_igvn(slow_region);
4809 if (!stopped()) {
4810 // It's an instance (we did array above). Make the slow-path tests.
4811 // If this is a virtual call, we generate a funny guard. We grab
4812 // the vtable entry corresponding to clone() from the target object.
4813 // If the target method which we are calling happens to be the
4814 // Object clone() method, we pass the guard. We do not need this
4815 // guard for non-virtual calls; the caller is known to be the native
4816 // Object clone().
4817 if (is_virtual) {
4818 generate_virtual_guard(obj_klass, slow_region);
4819 }
4820
4821 // The object must be easily cloneable and must not have a finalizer.
4822 // Both of these conditions may be checked in a single test.
4823 // We could optimize the test further, but we don't care.
4824 generate_access_flags_guard(obj_klass,
4825 // Test both conditions:
4826 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4827 // Must be cloneable but not finalizer:
4828 JVM_ACC_IS_CLONEABLE_FAST,
4829 slow_region);
4830 }
4831
4832 if (!stopped()) {
4833 // It's an instance, and it passed the slow-path tests.
4834 PreserveJVMState pjvms(this);
4835 Node* obj_size = NULL;
4836 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4837 // is reexecuted if deoptimization occurs and there could be problems when merging
4838 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4839 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4840
4841 copy_to_clone(obj, alloc_obj, obj_size, false);
4842
4843 // Present the results of the slow call.
4844 result_reg->init_req(_instance_path, control());
4845 result_val->init_req(_instance_path, alloc_obj);
4846 result_i_o ->set_req(_instance_path, i_o());
4847 result_mem ->set_req(_instance_path, reset_memory());
4848 }
4849
4850 // Generate code for the slow case. We make a call to clone().
4851 set_control(_gvn.transform(slow_region));
4852 if (!stopped()) {
4853 PreserveJVMState pjvms(this);
4854 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4855 // We need to deoptimize on exception (see comment above)
4856 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4857 // this->control() comes from set_results_for_java_call
4858 result_reg->init_req(_slow_path, control());
4859 result_val->init_req(_slow_path, slow_result);
4860 result_i_o ->set_req(_slow_path, i_o());
4861 result_mem ->set_req(_slow_path, reset_memory());
4862 }
4863
4864 // Return the combined state.
4865 set_control( _gvn.transform(result_reg));
4866 set_i_o( _gvn.transform(result_i_o));
4867 set_all_memory( _gvn.transform(result_mem));
4868 } // original reexecute is set back here
4869
4870 set_result(_gvn.transform(result_val));
4871 return true;
4872 }
4873
4874 // If we have a tightly coupled allocation, the arraycopy may take care
4875 // of the array initialization. If one of the guards we insert between
4876 // the allocation and the arraycopy causes a deoptimization, an
4877 // uninitialized array will escape the compiled method. To prevent that
4878 // we set the JVM state for uncommon traps between the allocation and
4879 // the arraycopy to the state before the allocation so, in case of
4880 // deoptimization, we'll reexecute the allocation and the
4881 // initialization.
4882 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4883 if (alloc != NULL) {
4884 ciMethod* trap_method = alloc->jvms()->method();
4885 int trap_bci = alloc->jvms()->bci();
4886
4887 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4888 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4889 // Make sure there's no store between the allocation and the
4890 // arraycopy otherwise visible side effects could be rexecuted
4891 // in case of deoptimization and cause incorrect execution.
4892 bool no_interfering_store = true;
4893 Node* mem = alloc->in(TypeFunc::Memory);
4894 if (mem->is_MergeMem()) {
4895 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4896 Node* n = mms.memory();
4897 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4898 assert(n->is_Store(), "what else?");
4899 no_interfering_store = false;
4900 break;
4901 }
4902 }
4903 } else {
4904 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4905 Node* n = mms.memory();
4906 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4907 assert(n->is_Store(), "what else?");
4908 no_interfering_store = false;
4909 break;
4910 }
4911 }
4912 }
4913
4914 if (no_interfering_store) {
4915 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4916 uint size = alloc->req();
4917 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4918 old_jvms->set_map(sfpt);
4919 for (uint i = 0; i < size; i++) {
4920 sfpt->init_req(i, alloc->in(i));
4921 }
4922 // re-push array length for deoptimization
4923 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4924 old_jvms->set_sp(old_jvms->sp()+1);
4925 old_jvms->set_monoff(old_jvms->monoff()+1);
4926 old_jvms->set_scloff(old_jvms->scloff()+1);
4927 old_jvms->set_endoff(old_jvms->endoff()+1);
4928 old_jvms->set_should_reexecute(true);
4929
4930 sfpt->set_i_o(map()->i_o());
4931 sfpt->set_memory(map()->memory());
4932 sfpt->set_control(map()->control());
4933
4934 JVMState* saved_jvms = jvms();
4935 saved_reexecute_sp = _reexecute_sp;
4936
4937 set_jvms(sfpt->jvms());
4938 _reexecute_sp = jvms()->sp();
4939
4940 return saved_jvms;
4941 }
4942 }
4943 }
4944 return NULL;
4945 }
4946
4947 // In case of a deoptimization, we restart execution at the
4948 // allocation, allocating a new array. We would leave an uninitialized
4949 // array in the heap that GCs wouldn't expect. Move the allocation
4950 // after the traps so we don't allocate the array if we
4951 // deoptimize. This is possible because tightly_coupled_allocation()
4952 // guarantees there's no observer of the allocated array at this point
4953 // and the control flow is simple enough.
4954 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4955 int saved_reexecute_sp, uint new_idx) {
4956 if (saved_jvms != NULL && !stopped()) {
4957 assert(alloc != NULL, "only with a tightly coupled allocation");
4958 // restore JVM state to the state at the arraycopy
4959 saved_jvms->map()->set_control(map()->control());
4960 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4961 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4962 // If we've improved the types of some nodes (null check) while
4963 // emitting the guards, propagate them to the current state
4964 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4965 set_jvms(saved_jvms);
4966 _reexecute_sp = saved_reexecute_sp;
4967
4968 // Remove the allocation from above the guards
4969 CallProjections callprojs;
4970 alloc->extract_projections(&callprojs, true);
4971 InitializeNode* init = alloc->initialization();
4972 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4973 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4974 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4975
4976 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4977 // the allocation (i.e. is only valid if the allocation succeeds):
4978 // 1) replace CastIINode with AllocateArrayNode's length here
4979 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4980 //
4981 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4982 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4983 Node* init_control = init->proj_out(TypeFunc::Control);
4984 Node* alloc_length = alloc->Ideal_length();
4985 #ifdef ASSERT
4986 Node* prev_cast = NULL;
4987 #endif
4988 for (uint i = 0; i < init_control->outcnt(); i++) {
4989 Node* init_out = init_control->raw_out(i);
4990 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4991 #ifdef ASSERT
4992 if (prev_cast == NULL) {
4993 prev_cast = init_out;
4994 } else {
4995 if (prev_cast->cmp(*init_out) == false) {
4996 prev_cast->dump();
4997 init_out->dump();
4998 assert(false, "not equal CastIINode");
4999 }
5000 }
5001 #endif
5002 C->gvn_replace_by(init_out, alloc_length);
5003 }
5004 }
5005 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5006
5007 // move the allocation here (after the guards)
5008 _gvn.hash_delete(alloc);
5009 alloc->set_req(TypeFunc::Control, control());
5010 alloc->set_req(TypeFunc::I_O, i_o());
5011 Node *mem = reset_memory();
5012 set_all_memory(mem);
5013 alloc->set_req(TypeFunc::Memory, mem);
5014 set_control(init->proj_out_or_null(TypeFunc::Control));
5015 set_i_o(callprojs.fallthrough_ioproj);
5016
5017 // Update memory as done in GraphKit::set_output_for_allocation()
5018 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5019 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5020 if (ary_type->isa_aryptr() && length_type != NULL) {
5021 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5022 }
5023 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5024 int elemidx = C->get_alias_index(telemref);
5025 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5026 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5027
5028 Node* allocx = _gvn.transform(alloc);
5029 assert(allocx == alloc, "where has the allocation gone?");
5030 assert(dest->is_CheckCastPP(), "not an allocation result?");
5031
5032 _gvn.hash_delete(dest);
5033 dest->set_req(0, control());
5034 Node* destx = _gvn.transform(dest);
5035 assert(destx == dest, "where has the allocation result gone?");
5036
5037 array_ideal_length(alloc, ary_type, true);
5038 }
5039 }
5040
5041
5042 //------------------------------inline_arraycopy-----------------------
5043 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5044 // Object dest, int destPos,
5045 // int length);
5046 bool LibraryCallKit::inline_arraycopy() {
5047 // Get the arguments.
5048 Node* src = argument(0); // type: oop
5049 Node* src_offset = argument(1); // type: int
5050 Node* dest = argument(2); // type: oop
5051 Node* dest_offset = argument(3); // type: int
5052 Node* length = argument(4); // type: int
5053
5054 uint new_idx = C->unique();
5055
5056 // Check for allocation before we add nodes that would confuse
5057 // tightly_coupled_allocation()
5058 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5059
5060 int saved_reexecute_sp = -1;
5061 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5062 // See arraycopy_restore_alloc_state() comment
5063 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5064 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5065 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5066 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5067
5068 // The following tests must be performed
5069 // (1) src and dest are arrays.
5070 // (2) src and dest arrays must have elements of the same BasicType
5071 // (3) src and dest must not be null.
5072 // (4) src_offset must not be negative.
5073 // (5) dest_offset must not be negative.
5074 // (6) length must not be negative.
5075 // (7) src_offset + length must not exceed length of src.
5076 // (8) dest_offset + length must not exceed length of dest.
5077 // (9) each element of an oop array must be assignable
5078
5079 // (3) src and dest must not be null.
5080 // always do this here because we need the JVM state for uncommon traps
5081 Node* null_ctl = top();
5082 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5083 assert(null_ctl->is_top(), "no null control here");
5084 dest = null_check(dest, T_ARRAY);
5085
5086 if (!can_emit_guards) {
5087 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5088 // guards but the arraycopy node could still take advantage of a
5089 // tightly allocated allocation. tightly_coupled_allocation() is
5090 // called again to make sure it takes the null check above into
5091 // account: the null check is mandatory and if it caused an
5092 // uncommon trap to be emitted then the allocation can't be
5093 // considered tightly coupled in this context.
5094 alloc = tightly_coupled_allocation(dest);
5095 }
5096
5097 bool validated = false;
5098
5099 const Type* src_type = _gvn.type(src);
5100 const Type* dest_type = _gvn.type(dest);
5101 const TypeAryPtr* top_src = src_type->isa_aryptr();
5102 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5103
5104 // Do we have the type of src?
5105 bool has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5106 // Do we have the type of dest?
5107 bool has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5108 // Is the type for src from speculation?
5109 bool src_spec = false;
5110 // Is the type for dest from speculation?
5111 bool dest_spec = false;
5112
5113 if ((!has_src || !has_dest) && can_emit_guards) {
5114 // We don't have sufficient type information, let's see if
5115 // speculative types can help. We need to have types for both src
5116 // and dest so that it pays off.
5117
5118 // Do we already have or could we have type information for src
5119 bool could_have_src = has_src;
5120 // Do we already have or could we have type information for dest
5121 bool could_have_dest = has_dest;
5122
5123 ciKlass* src_k = NULL;
5124 if (!has_src) {
5125 src_k = src_type->speculative_type_not_null();
5126 if (src_k != NULL && src_k->is_array_klass()) {
5127 could_have_src = true;
5128 }
5129 }
5130
5131 ciKlass* dest_k = NULL;
5132 if (!has_dest) {
5133 dest_k = dest_type->speculative_type_not_null();
5134 if (dest_k != NULL && dest_k->is_array_klass()) {
5135 could_have_dest = true;
5136 }
5137 }
5138
5139 if (could_have_src && could_have_dest) {
5140 // This is going to pay off so emit the required guards
5141 if (!has_src) {
5142 src = maybe_cast_profiled_obj(src, src_k, true);
5143 src_type = _gvn.type(src);
5144 top_src = src_type->isa_aryptr();
5145 has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5146 src_spec = true;
5147 }
5148 if (!has_dest) {
5149 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5150 dest_type = _gvn.type(dest);
5151 top_dest = dest_type->isa_aryptr();
5152 has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5153 dest_spec = true;
5154 }
5155 }
5156 }
5157
5158 if (has_src && has_dest && can_emit_guards) {
5159 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5160 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5161 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5162 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5163
5164 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5165 // If both arrays are object arrays then having the exact types
5166 // for both will remove the need for a subtype check at runtime
5167 // before the call and may make it possible to pick a faster copy
5168 // routine (without a subtype check on every element)
5169 // Do we have the exact type of src?
5170 bool could_have_src = src_spec;
5171 // Do we have the exact type of dest?
5172 bool could_have_dest = dest_spec;
5173 ciKlass* src_k = NULL;
5174 ciKlass* dest_k = NULL;
5175 if (!src_spec) {
5176 src_k = src_type->speculative_type_not_null();
5177 if (src_k != NULL && src_k->is_array_klass()) {
5178 could_have_src = true;
5179 }
5180 }
5181 if (!dest_spec) {
5182 dest_k = dest_type->speculative_type_not_null();
5183 if (dest_k != NULL && dest_k->is_array_klass()) {
5184 could_have_dest = true;
5185 }
5186 }
5187 if (could_have_src && could_have_dest) {
5188 // If we can have both exact types, emit the missing guards
5189 if (could_have_src && !src_spec) {
5190 src = maybe_cast_profiled_obj(src, src_k, true);
5191 }
5192 if (could_have_dest && !dest_spec) {
5193 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5194 }
5195 }
5196 }
5197 }
5198
5199 ciMethod* trap_method = method();
5200 int trap_bci = bci();
5201 if (saved_jvms != NULL) {
5202 trap_method = alloc->jvms()->method();
5203 trap_bci = alloc->jvms()->bci();
5204 }
5205
5206 bool negative_length_guard_generated = false;
5207
5208 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5209 can_emit_guards &&
5210 !src->is_top() && !dest->is_top()) {
5211 // validate arguments: enables transformation the ArrayCopyNode
5212 validated = true;
5213
5214 RegionNode* slow_region = new RegionNode(1);
5215 record_for_igvn(slow_region);
5216
5217 // (1) src and dest are arrays.
5218 generate_non_array_guard(load_object_klass(src), slow_region);
5219 generate_non_array_guard(load_object_klass(dest), slow_region);
5220
5221 // (2) src and dest arrays must have elements of the same BasicType
5222 // done at macro expansion or at Ideal transformation time
5223
5224 // (4) src_offset must not be negative.
5225 generate_negative_guard(src_offset, slow_region);
5226
5227 // (5) dest_offset must not be negative.
5228 generate_negative_guard(dest_offset, slow_region);
5229
5230 // (7) src_offset + length must not exceed length of src.
5231 generate_limit_guard(src_offset, length,
5232 load_array_length(src),
5233 slow_region);
5234
5235 // (8) dest_offset + length must not exceed length of dest.
5236 generate_limit_guard(dest_offset, length,
5237 load_array_length(dest),
5238 slow_region);
5239
5240 // (6) length must not be negative.
5241 // This is also checked in generate_arraycopy() during macro expansion, but
5242 // we also have to check it here for the case where the ArrayCopyNode will
5243 // be eliminated by Escape Analysis.
5244 if (EliminateAllocations) {
5245 generate_negative_guard(length, slow_region);
5246 negative_length_guard_generated = true;
5247 }
5248
5249 // (9) each element of an oop array must be assignable
5250 Node* dest_klass = load_object_klass(dest);
5251 if (src != dest) {
5252 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5253
5254 if (not_subtype_ctrl != top()) {
5255 PreserveJVMState pjvms(this);
5256 set_control(not_subtype_ctrl);
5257 uncommon_trap(Deoptimization::Reason_intrinsic,
5258 Deoptimization::Action_make_not_entrant);
5259 assert(stopped(), "Should be stopped");
5260 }
5261 }
5262 {
5263 PreserveJVMState pjvms(this);
5264 set_control(_gvn.transform(slow_region));
5265 uncommon_trap(Deoptimization::Reason_intrinsic,
5266 Deoptimization::Action_make_not_entrant);
5267 assert(stopped(), "Should be stopped");
5268 }
5269
5270 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5271 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5272 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5273 }
5274
5275 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5276
5277 if (stopped()) {
5278 return true;
5279 }
5280
5281 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5282 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5283 // so the compiler has a chance to eliminate them: during macro expansion,
5284 // we have to set their control (CastPP nodes are eliminated).
5285 load_object_klass(src), load_object_klass(dest),
5286 load_array_length(src), load_array_length(dest));
5287
5288 ac->set_arraycopy(validated);
5289
5290 Node* n = _gvn.transform(ac);
5291 if (n == ac) {
5292 ac->connect_outputs(this);
5293 } else {
5294 assert(validated, "shouldn't transform if all arguments not validated");
5295 set_all_memory(n);
5296 }
5297 clear_upper_avx();
5298
5299
5300 return true;
5301 }
5302
5303
5304 // Helper function which determines if an arraycopy immediately follows
5305 // an allocation, with no intervening tests or other escapes for the object.
5306 AllocateArrayNode*
5307 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5308 if (stopped()) return NULL; // no fast path
5309 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5310
5311 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5312 if (alloc == NULL) return NULL;
5313
5314 Node* rawmem = memory(Compile::AliasIdxRaw);
5315 // Is the allocation's memory state untouched?
5316 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5317 // Bail out if there have been raw-memory effects since the allocation.
5318 // (Example: There might have been a call or safepoint.)
5319 return NULL;
5320 }
5321 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5322 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5323 return NULL;
5324 }
5325
5326 // There must be no unexpected observers of this allocation.
5327 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5328 Node* obs = ptr->fast_out(i);
5329 if (obs != this->map()) {
5330 return NULL;
5331 }
5332 }
5333
5334 // This arraycopy must unconditionally follow the allocation of the ptr.
5335 Node* alloc_ctl = ptr->in(0);
5336 Node* ctl = control();
5337 while (ctl != alloc_ctl) {
5338 // There may be guards which feed into the slow_region.
5339 // Any other control flow means that we might not get a chance
5340 // to finish initializing the allocated object.
5341 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5342 IfNode* iff = ctl->in(0)->as_If();
5343 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5344 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5345 // One more try: Various low-level checks bottom out in
5346 // uncommon traps. If the debug-info of the trap omits
5347 // any reference to the allocation, as we've already
5348 // observed, then there can be no objection to the trap.
5349 bool found_trap = false;
5350 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5351 Node* obs = not_ctl->fast_out(j);
5352 if (obs->in(0) == not_ctl && obs->is_Call() &&
5353 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5354 found_trap = true; break;
5355 }
5356 }
5357 if (found_trap) {
5358 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5359 continue;
5360 }
5361 }
5362 return NULL;
5363 }
5364
5365 // If we get this far, we have an allocation which immediately
5366 // precedes the arraycopy, and we can take over zeroing the new object.
5367 // The arraycopy will finish the initialization, and provide
5368 // a new control state to which we will anchor the destination pointer.
5369
5370 return alloc;
5371 }
5372
5373 //-------------inline_encodeISOArray-----------------------------------
5374 // encode char[] to byte[] in ISO_8859_1 or ASCII
5375 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5376 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5377 // no receiver since it is static method
5378 Node *src = argument(0);
5379 Node *src_offset = argument(1);
5380 Node *dst = argument(2);
5381 Node *dst_offset = argument(3);
5382 Node *length = argument(4);
5383
5384 src = must_be_not_null(src, true);
5385 dst = must_be_not_null(dst, true);
5386
5387 const Type* src_type = src->Value(&_gvn);
5388 const Type* dst_type = dst->Value(&_gvn);
5389 const TypeAryPtr* top_src = src_type->isa_aryptr();
5390 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5391 if (top_src == NULL || top_src->elem() == Type::BOTTOM ||
5392 top_dest == NULL || top_dest->elem() == Type::BOTTOM) {
5393 // failed array check
5394 return false;
5395 }
5396
5397 // Figure out the size and type of the elements we will be copying.
5398 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5399 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
5400 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5401 return false;
5402 }
5403
5404 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5405 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5406 // 'src_start' points to src array + scaled offset
5407 // 'dst_start' points to dst array + scaled offset
5408
5409 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5410 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5411 enc = _gvn.transform(enc);
5412 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5413 set_memory(res_mem, mtype);
5414 set_result(enc);
5415 clear_upper_avx();
5416
5417 return true;
5418 }
5419
5420 //-------------inline_multiplyToLen-----------------------------------
5421 bool LibraryCallKit::inline_multiplyToLen() {
5422 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5423
5424 address stubAddr = StubRoutines::multiplyToLen();
5425 if (stubAddr == NULL) {
5426 return false; // Intrinsic's stub is not implemented on this platform
5427 }
5428 const char* stubName = "multiplyToLen";
5429
5430 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5431
5432 // no receiver because it is a static method
5433 Node* x = argument(0);
5434 Node* xlen = argument(1);
5435 Node* y = argument(2);
5436 Node* ylen = argument(3);
5437 Node* z = argument(4);
5438
5439 x = must_be_not_null(x, true);
5440 y = must_be_not_null(y, true);
5441
5442 const Type* x_type = x->Value(&_gvn);
5443 const Type* y_type = y->Value(&_gvn);
5444 const TypeAryPtr* top_x = x_type->isa_aryptr();
5445 const TypeAryPtr* top_y = y_type->isa_aryptr();
5446 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5447 top_y == NULL || top_y->elem() == Type::BOTTOM) {
5448 // failed array check
5449 return false;
5450 }
5451
5452 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5453 BasicType y_elem = y_type->isa_aryptr()->elem()->array_element_basic_type();
5454 if (x_elem != T_INT || y_elem != T_INT) {
5455 return false;
5456 }
5457
5458 // Set the original stack and the reexecute bit for the interpreter to reexecute
5459 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5460 // on the return from z array allocation in runtime.
5461 { PreserveReexecuteState preexecs(this);
5462 jvms()->set_should_reexecute(true);
5463
5464 Node* x_start = array_element_address(x, intcon(0), x_elem);
5465 Node* y_start = array_element_address(y, intcon(0), y_elem);
5466 // 'x_start' points to x array + scaled xlen
5467 // 'y_start' points to y array + scaled ylen
5468
5469 // Allocate the result array
5470 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5471 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5472 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5473
5474 IdealKit ideal(this);
5475
5476 #define __ ideal.
5477 Node* one = __ ConI(1);
5478 Node* zero = __ ConI(0);
5479 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5480 __ set(need_alloc, zero);
5481 __ set(z_alloc, z);
5482 __ if_then(z, BoolTest::eq, null()); {
5483 __ increment (need_alloc, one);
5484 } __ else_(); {
5485 // Update graphKit memory and control from IdealKit.
5486 sync_kit(ideal);
5487 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5488 cast->init_req(0, control());
5489 _gvn.set_type(cast, cast->bottom_type());
5490 C->record_for_igvn(cast);
5491
5492 Node* zlen_arg = load_array_length(cast);
5493 // Update IdealKit memory and control from graphKit.
5494 __ sync_kit(this);
5495 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5496 __ increment (need_alloc, one);
5497 } __ end_if();
5498 } __ end_if();
5499
5500 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5501 // Update graphKit memory and control from IdealKit.
5502 sync_kit(ideal);
5503 Node * narr = new_array(klass_node, zlen, 1);
5504 // Update IdealKit memory and control from graphKit.
5505 __ sync_kit(this);
5506 __ set(z_alloc, narr);
5507 } __ end_if();
5508
5509 sync_kit(ideal);
5510 z = __ value(z_alloc);
5511 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5512 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5513 // Final sync IdealKit and GraphKit.
5514 final_sync(ideal);
5515 #undef __
5516
5517 Node* z_start = array_element_address(z, intcon(0), T_INT);
5518
5519 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5520 OptoRuntime::multiplyToLen_Type(),
5521 stubAddr, stubName, TypePtr::BOTTOM,
5522 x_start, xlen, y_start, ylen, z_start, zlen);
5523 } // original reexecute is set back here
5524
5525 C->set_has_split_ifs(true); // Has chance for split-if optimization
5526 set_result(z);
5527 return true;
5528 }
5529
5530 //-------------inline_squareToLen------------------------------------
5531 bool LibraryCallKit::inline_squareToLen() {
5532 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5533
5534 address stubAddr = StubRoutines::squareToLen();
5535 if (stubAddr == NULL) {
5536 return false; // Intrinsic's stub is not implemented on this platform
5537 }
5538 const char* stubName = "squareToLen";
5539
5540 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5541
5542 Node* x = argument(0);
5543 Node* len = argument(1);
5544 Node* z = argument(2);
5545 Node* zlen = argument(3);
5546
5547 x = must_be_not_null(x, true);
5548 z = must_be_not_null(z, true);
5549
5550 const Type* x_type = x->Value(&_gvn);
5551 const Type* z_type = z->Value(&_gvn);
5552 const TypeAryPtr* top_x = x_type->isa_aryptr();
5553 const TypeAryPtr* top_z = z_type->isa_aryptr();
5554 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5555 top_z == NULL || top_z->elem() == Type::BOTTOM) {
5556 // failed array check
5557 return false;
5558 }
5559
5560 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5561 BasicType z_elem = z_type->isa_aryptr()->elem()->array_element_basic_type();
5562 if (x_elem != T_INT || z_elem != T_INT) {
5563 return false;
5564 }
5565
5566
5567 Node* x_start = array_element_address(x, intcon(0), x_elem);
5568 Node* z_start = array_element_address(z, intcon(0), z_elem);
5569
5570 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5571 OptoRuntime::squareToLen_Type(),
5572 stubAddr, stubName, TypePtr::BOTTOM,
5573 x_start, len, z_start, zlen);
5574
5575 set_result(z);
5576 return true;
5577 }
5578
5579 //-------------inline_mulAdd------------------------------------------
5580 bool LibraryCallKit::inline_mulAdd() {
5581 assert(UseMulAddIntrinsic, "not implemented on this platform");
5582
5583 address stubAddr = StubRoutines::mulAdd();
5584 if (stubAddr == NULL) {
5585 return false; // Intrinsic's stub is not implemented on this platform
5586 }
5587 const char* stubName = "mulAdd";
5588
5589 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5590
5591 Node* out = argument(0);
5592 Node* in = argument(1);
5593 Node* offset = argument(2);
5594 Node* len = argument(3);
5595 Node* k = argument(4);
5596
5597 out = must_be_not_null(out, true);
5598
5599 const Type* out_type = out->Value(&_gvn);
5600 const Type* in_type = in->Value(&_gvn);
5601 const TypeAryPtr* top_out = out_type->isa_aryptr();
5602 const TypeAryPtr* top_in = in_type->isa_aryptr();
5603 if (top_out == NULL || top_out->elem() == Type::BOTTOM ||
5604 top_in == NULL || top_in->elem() == Type::BOTTOM) {
5605 // failed array check
5606 return false;
5607 }
5608
5609 BasicType out_elem = out_type->isa_aryptr()->elem()->array_element_basic_type();
5610 BasicType in_elem = in_type->isa_aryptr()->elem()->array_element_basic_type();
5611 if (out_elem != T_INT || in_elem != T_INT) {
5612 return false;
5613 }
5614
5615 Node* outlen = load_array_length(out);
5616 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5617 Node* out_start = array_element_address(out, intcon(0), out_elem);
5618 Node* in_start = array_element_address(in, intcon(0), in_elem);
5619
5620 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5621 OptoRuntime::mulAdd_Type(),
5622 stubAddr, stubName, TypePtr::BOTTOM,
5623 out_start,in_start, new_offset, len, k);
5624 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5625 set_result(result);
5626 return true;
5627 }
5628
5629 //-------------inline_montgomeryMultiply-----------------------------------
5630 bool LibraryCallKit::inline_montgomeryMultiply() {
5631 address stubAddr = StubRoutines::montgomeryMultiply();
5632 if (stubAddr == NULL) {
5633 return false; // Intrinsic's stub is not implemented on this platform
5634 }
5635
5636 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5637 const char* stubName = "montgomery_multiply";
5638
5639 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5640
5641 Node* a = argument(0);
5642 Node* b = argument(1);
5643 Node* n = argument(2);
5644 Node* len = argument(3);
5645 Node* inv = argument(4);
5646 Node* m = argument(6);
5647
5648 const Type* a_type = a->Value(&_gvn);
5649 const TypeAryPtr* top_a = a_type->isa_aryptr();
5650 const Type* b_type = b->Value(&_gvn);
5651 const TypeAryPtr* top_b = b_type->isa_aryptr();
5652 const Type* n_type = a->Value(&_gvn);
5653 const TypeAryPtr* top_n = n_type->isa_aryptr();
5654 const Type* m_type = a->Value(&_gvn);
5655 const TypeAryPtr* top_m = m_type->isa_aryptr();
5656 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5657 top_b == NULL || top_b->elem() == Type::BOTTOM ||
5658 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5659 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5660 // failed array check
5661 return false;
5662 }
5663
5664 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5665 BasicType b_elem = b_type->isa_aryptr()->elem()->array_element_basic_type();
5666 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5667 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5668 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5669 return false;
5670 }
5671
5672 // Make the call
5673 {
5674 Node* a_start = array_element_address(a, intcon(0), a_elem);
5675 Node* b_start = array_element_address(b, intcon(0), b_elem);
5676 Node* n_start = array_element_address(n, intcon(0), n_elem);
5677 Node* m_start = array_element_address(m, intcon(0), m_elem);
5678
5679 Node* call = make_runtime_call(RC_LEAF,
5680 OptoRuntime::montgomeryMultiply_Type(),
5681 stubAddr, stubName, TypePtr::BOTTOM,
5682 a_start, b_start, n_start, len, inv, top(),
5683 m_start);
5684 set_result(m);
5685 }
5686
5687 return true;
5688 }
5689
5690 bool LibraryCallKit::inline_montgomerySquare() {
5691 address stubAddr = StubRoutines::montgomerySquare();
5692 if (stubAddr == NULL) {
5693 return false; // Intrinsic's stub is not implemented on this platform
5694 }
5695
5696 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5697 const char* stubName = "montgomery_square";
5698
5699 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5700
5701 Node* a = argument(0);
5702 Node* n = argument(1);
5703 Node* len = argument(2);
5704 Node* inv = argument(3);
5705 Node* m = argument(5);
5706
5707 const Type* a_type = a->Value(&_gvn);
5708 const TypeAryPtr* top_a = a_type->isa_aryptr();
5709 const Type* n_type = a->Value(&_gvn);
5710 const TypeAryPtr* top_n = n_type->isa_aryptr();
5711 const Type* m_type = a->Value(&_gvn);
5712 const TypeAryPtr* top_m = m_type->isa_aryptr();
5713 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5714 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5715 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5716 // failed array check
5717 return false;
5718 }
5719
5720 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5721 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5722 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5723 if (a_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* n_start = array_element_address(n, intcon(0), n_elem);
5731 Node* m_start = array_element_address(m, intcon(0), m_elem);
5732
5733 Node* call = make_runtime_call(RC_LEAF,
5734 OptoRuntime::montgomerySquare_Type(),
5735 stubAddr, stubName, TypePtr::BOTTOM,
5736 a_start, n_start, len, inv, top(),
5737 m_start);
5738 set_result(m);
5739 }
5740
5741 return true;
5742 }
5743
5744 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5745 address stubAddr = NULL;
5746 const char* stubName = NULL;
5747
5748 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5749 if (stubAddr == NULL) {
5750 return false; // Intrinsic's stub is not implemented on this platform
5751 }
5752
5753 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5754
5755 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5756
5757 Node* newArr = argument(0);
5758 Node* oldArr = argument(1);
5759 Node* newIdx = argument(2);
5760 Node* shiftCount = argument(3);
5761 Node* numIter = argument(4);
5762
5763 const Type* newArr_type = newArr->Value(&_gvn);
5764 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5765 const Type* oldArr_type = oldArr->Value(&_gvn);
5766 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5767 if (top_newArr == NULL || top_newArr->elem() == Type::BOTTOM || top_oldArr == NULL
5768 || top_oldArr->elem() == Type::BOTTOM) {
5769 return false;
5770 }
5771
5772 BasicType newArr_elem = newArr_type->isa_aryptr()->elem()->array_element_basic_type();
5773 BasicType oldArr_elem = oldArr_type->isa_aryptr()->elem()->array_element_basic_type();
5774 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5775 return false;
5776 }
5777
5778 // Make the call
5779 {
5780 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5781 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5782
5783 Node* call = make_runtime_call(RC_LEAF,
5784 OptoRuntime::bigIntegerShift_Type(),
5785 stubAddr,
5786 stubName,
5787 TypePtr::BOTTOM,
5788 newArr_start,
5789 oldArr_start,
5790 newIdx,
5791 shiftCount,
5792 numIter);
5793 }
5794
5795 return true;
5796 }
5797
5798 //-------------inline_vectorizedMismatch------------------------------
5799 bool LibraryCallKit::inline_vectorizedMismatch() {
5800 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5801
5802 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5803 Node* obja = argument(0); // Object
5804 Node* aoffset = argument(1); // long
5805 Node* objb = argument(3); // Object
5806 Node* boffset = argument(4); // long
5807 Node* length = argument(6); // int
5808 Node* scale = argument(7); // int
5809
5810 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5811 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5812 if (obja_t == NULL || obja_t->elem() == Type::BOTTOM ||
5813 objb_t == NULL || objb_t->elem() == Type::BOTTOM ||
5814 scale == top()) {
5815 return false; // failed input validation
5816 }
5817
5818 Node* obja_adr = make_unsafe_address(obja, aoffset);
5819 Node* objb_adr = make_unsafe_address(objb, boffset);
5820
5821 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5822 //
5823 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5824 // if (length <= inline_limit) {
5825 // inline_path:
5826 // vmask = VectorMaskGen length
5827 // vload1 = LoadVectorMasked obja, vmask
5828 // vload2 = LoadVectorMasked objb, vmask
5829 // result1 = VectorCmpMasked vload1, vload2, vmask
5830 // } else {
5831 // call_stub_path:
5832 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5833 // }
5834 // exit_block:
5835 // return Phi(result1, result2);
5836 //
5837 enum { inline_path = 1, // input is small enough to process it all at once
5838 stub_path = 2, // input is too large; call into the VM
5839 PATH_LIMIT = 3
5840 };
5841
5842 Node* exit_block = new RegionNode(PATH_LIMIT);
5843 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5844 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5845
5846 Node* call_stub_path = control();
5847
5848 BasicType elem_bt = T_ILLEGAL;
5849
5850 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5851 if (scale_t->is_con()) {
5852 switch (scale_t->get_con()) {
5853 case 0: elem_bt = T_BYTE; break;
5854 case 1: elem_bt = T_SHORT; break;
5855 case 2: elem_bt = T_INT; break;
5856 case 3: elem_bt = T_LONG; break;
5857
5858 default: elem_bt = T_ILLEGAL; break; // not supported
5859 }
5860 }
5861
5862 int inline_limit = 0;
5863 bool do_partial_inline = false;
5864
5865 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5866 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5867 do_partial_inline = inline_limit >= 16;
5868 }
5869
5870 if (do_partial_inline) {
5871 assert(elem_bt != T_ILLEGAL, "sanity");
5872
5873 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5874 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5875 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5876
5877 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5878 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5879 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5880
5881 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5882
5883 if (!stopped()) {
5884 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5885
5886 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5887 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5888 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5889 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5890
5891 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
5892 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5893 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5894 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5895
5896 exit_block->init_req(inline_path, control());
5897 memory_phi->init_req(inline_path, map()->memory());
5898 result_phi->init_req(inline_path, result);
5899
5900 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5901 clear_upper_avx();
5902 }
5903 }
5904 }
5905
5906 if (call_stub_path != NULL) {
5907 set_control(call_stub_path);
5908
5909 Node* call = make_runtime_call(RC_LEAF,
5910 OptoRuntime::vectorizedMismatch_Type(),
5911 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5912 obja_adr, objb_adr, length, scale);
5913
5914 exit_block->init_req(stub_path, control());
5915 memory_phi->init_req(stub_path, map()->memory());
5916 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5917 }
5918
5919 exit_block = _gvn.transform(exit_block);
5920 memory_phi = _gvn.transform(memory_phi);
5921 result_phi = _gvn.transform(result_phi);
5922
5923 set_control(exit_block);
5924 set_all_memory(memory_phi);
5925 set_result(result_phi);
5926
5927 return true;
5928 }
5929
5930 /**
5931 * Calculate CRC32 for byte.
5932 * int java.util.zip.CRC32.update(int crc, int b)
5933 */
5934 bool LibraryCallKit::inline_updateCRC32() {
5935 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5936 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5937 // no receiver since it is static method
5938 Node* crc = argument(0); // type: int
5939 Node* b = argument(1); // type: int
5940
5941 /*
5942 * int c = ~ crc;
5943 * b = timesXtoThe32[(b ^ c) & 0xFF];
5944 * b = b ^ (c >>> 8);
5945 * crc = ~b;
5946 */
5947
5948 Node* M1 = intcon(-1);
5949 crc = _gvn.transform(new XorINode(crc, M1));
5950 Node* result = _gvn.transform(new XorINode(crc, b));
5951 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5952
5953 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5954 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5955 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5956 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5957
5958 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5959 result = _gvn.transform(new XorINode(crc, result));
5960 result = _gvn.transform(new XorINode(result, M1));
5961 set_result(result);
5962 return true;
5963 }
5964
5965 /**
5966 * Calculate CRC32 for byte[] array.
5967 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5968 */
5969 bool LibraryCallKit::inline_updateBytesCRC32() {
5970 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5971 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5972 // no receiver since it is static method
5973 Node* crc = argument(0); // type: int
5974 Node* src = argument(1); // type: oop
5975 Node* offset = argument(2); // type: int
5976 Node* length = argument(3); // type: int
5977
5978 const Type* src_type = src->Value(&_gvn);
5979 const TypeAryPtr* top_src = src_type->isa_aryptr();
5980 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
5981 // failed array check
5982 return false;
5983 }
5984
5985 // Figure out the size and type of the elements we will be copying.
5986 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5987 if (src_elem != T_BYTE) {
5988 return false;
5989 }
5990
5991 // 'src_start' points to src array + scaled offset
5992 src = must_be_not_null(src, true);
5993 Node* src_start = array_element_address(src, offset, src_elem);
5994
5995 // We assume that range check is done by caller.
5996 // TODO: generate range check (offset+length < src.length) in debug VM.
5997
5998 // Call the stub.
5999 address stubAddr = StubRoutines::updateBytesCRC32();
6000 const char *stubName = "updateBytesCRC32";
6001
6002 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6003 stubAddr, stubName, TypePtr::BOTTOM,
6004 crc, src_start, length);
6005 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6006 set_result(result);
6007 return true;
6008 }
6009
6010 /**
6011 * Calculate CRC32 for ByteBuffer.
6012 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6013 */
6014 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6015 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6016 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6017 // no receiver since it is static method
6018 Node* crc = argument(0); // type: int
6019 Node* src = argument(1); // type: long
6020 Node* offset = argument(3); // type: int
6021 Node* length = argument(4); // type: int
6022
6023 src = ConvL2X(src); // adjust Java long to machine word
6024 Node* base = _gvn.transform(new CastX2PNode(src));
6025 offset = ConvI2X(offset);
6026
6027 // 'src_start' points to src array + scaled offset
6028 Node* src_start = basic_plus_adr(top(), base, offset);
6029
6030 // Call the stub.
6031 address stubAddr = StubRoutines::updateBytesCRC32();
6032 const char *stubName = "updateBytesCRC32";
6033
6034 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6035 stubAddr, stubName, TypePtr::BOTTOM,
6036 crc, src_start, length);
6037 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6038 set_result(result);
6039 return true;
6040 }
6041
6042 //------------------------------get_table_from_crc32c_class-----------------------
6043 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6044 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6045 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6046
6047 return table;
6048 }
6049
6050 //------------------------------inline_updateBytesCRC32C-----------------------
6051 //
6052 // Calculate CRC32C for byte[] array.
6053 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6054 //
6055 bool LibraryCallKit::inline_updateBytesCRC32C() {
6056 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6057 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6058 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6059 // no receiver since it is a static method
6060 Node* crc = argument(0); // type: int
6061 Node* src = argument(1); // type: oop
6062 Node* offset = argument(2); // type: int
6063 Node* end = argument(3); // type: int
6064
6065 Node* length = _gvn.transform(new SubINode(end, offset));
6066
6067 const Type* src_type = src->Value(&_gvn);
6068 const TypeAryPtr* top_src = src_type->isa_aryptr();
6069 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6070 // failed array check
6071 return false;
6072 }
6073
6074 // Figure out the size and type of the elements we will be copying.
6075 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6076 if (src_elem != T_BYTE) {
6077 return false;
6078 }
6079
6080 // 'src_start' points to src array + scaled offset
6081 src = must_be_not_null(src, true);
6082 Node* src_start = array_element_address(src, offset, src_elem);
6083
6084 // static final int[] byteTable in class CRC32C
6085 Node* table = get_table_from_crc32c_class(callee()->holder());
6086 table = must_be_not_null(table, true);
6087 Node* table_start = array_element_address(table, intcon(0), T_INT);
6088
6089 // We assume that range check is done by caller.
6090 // TODO: generate range check (offset+length < src.length) in debug VM.
6091
6092 // Call the stub.
6093 address stubAddr = StubRoutines::updateBytesCRC32C();
6094 const char *stubName = "updateBytesCRC32C";
6095
6096 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6097 stubAddr, stubName, TypePtr::BOTTOM,
6098 crc, src_start, length, table_start);
6099 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6100 set_result(result);
6101 return true;
6102 }
6103
6104 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6105 //
6106 // Calculate CRC32C for DirectByteBuffer.
6107 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6108 //
6109 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6110 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6111 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6112 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6113 // no receiver since it is a static method
6114 Node* crc = argument(0); // type: int
6115 Node* src = argument(1); // type: long
6116 Node* offset = argument(3); // type: int
6117 Node* end = argument(4); // type: int
6118
6119 Node* length = _gvn.transform(new SubINode(end, offset));
6120
6121 src = ConvL2X(src); // adjust Java long to machine word
6122 Node* base = _gvn.transform(new CastX2PNode(src));
6123 offset = ConvI2X(offset);
6124
6125 // 'src_start' points to src array + scaled offset
6126 Node* src_start = basic_plus_adr(top(), base, offset);
6127
6128 // static final int[] byteTable in class CRC32C
6129 Node* table = get_table_from_crc32c_class(callee()->holder());
6130 table = must_be_not_null(table, true);
6131 Node* table_start = array_element_address(table, intcon(0), T_INT);
6132
6133 // Call the stub.
6134 address stubAddr = StubRoutines::updateBytesCRC32C();
6135 const char *stubName = "updateBytesCRC32C";
6136
6137 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6138 stubAddr, stubName, TypePtr::BOTTOM,
6139 crc, src_start, length, table_start);
6140 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6141 set_result(result);
6142 return true;
6143 }
6144
6145 //------------------------------inline_updateBytesAdler32----------------------
6146 //
6147 // Calculate Adler32 checksum for byte[] array.
6148 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6149 //
6150 bool LibraryCallKit::inline_updateBytesAdler32() {
6151 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6152 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6153 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6154 // no receiver since it is static method
6155 Node* crc = argument(0); // type: int
6156 Node* src = argument(1); // type: oop
6157 Node* offset = argument(2); // type: int
6158 Node* length = argument(3); // type: int
6159
6160 const Type* src_type = src->Value(&_gvn);
6161 const TypeAryPtr* top_src = src_type->isa_aryptr();
6162 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6163 // failed array check
6164 return false;
6165 }
6166
6167 // Figure out the size and type of the elements we will be copying.
6168 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6169 if (src_elem != T_BYTE) {
6170 return false;
6171 }
6172
6173 // 'src_start' points to src array + scaled offset
6174 Node* src_start = array_element_address(src, offset, src_elem);
6175
6176 // We assume that range check is done by caller.
6177 // TODO: generate range check (offset+length < src.length) in debug VM.
6178
6179 // Call the stub.
6180 address stubAddr = StubRoutines::updateBytesAdler32();
6181 const char *stubName = "updateBytesAdler32";
6182
6183 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6184 stubAddr, stubName, TypePtr::BOTTOM,
6185 crc, src_start, length);
6186 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6187 set_result(result);
6188 return true;
6189 }
6190
6191 //------------------------------inline_updateByteBufferAdler32---------------
6192 //
6193 // Calculate Adler32 checksum for DirectByteBuffer.
6194 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6195 //
6196 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6197 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6198 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6199 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6200 // no receiver since it is static method
6201 Node* crc = argument(0); // type: int
6202 Node* src = argument(1); // type: long
6203 Node* offset = argument(3); // type: int
6204 Node* length = argument(4); // type: int
6205
6206 src = ConvL2X(src); // adjust Java long to machine word
6207 Node* base = _gvn.transform(new CastX2PNode(src));
6208 offset = ConvI2X(offset);
6209
6210 // 'src_start' points to src array + scaled offset
6211 Node* src_start = basic_plus_adr(top(), base, offset);
6212
6213 // Call the stub.
6214 address stubAddr = StubRoutines::updateBytesAdler32();
6215 const char *stubName = "updateBytesAdler32";
6216
6217 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6218 stubAddr, stubName, TypePtr::BOTTOM,
6219 crc, src_start, length);
6220
6221 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6222 set_result(result);
6223 return true;
6224 }
6225
6226 //----------------------------inline_reference_get----------------------------
6227 // public T java.lang.ref.Reference.get();
6228 bool LibraryCallKit::inline_reference_get() {
6229 const int referent_offset = java_lang_ref_Reference::referent_offset();
6230
6231 // Get the argument:
6232 Node* reference_obj = null_check_receiver();
6233 if (stopped()) return true;
6234
6235 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6236 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6237 decorators, /*is_static*/ false, NULL);
6238 if (result == NULL) return false;
6239
6240 // Add memory barrier to prevent commoning reads from this field
6241 // across safepoint since GC can change its value.
6242 insert_mem_bar(Op_MemBarCPUOrder);
6243
6244 set_result(result);
6245 return true;
6246 }
6247
6248 //----------------------------inline_reference_refersTo0----------------------------
6249 // bool java.lang.ref.Reference.refersTo0();
6250 // bool java.lang.ref.PhantomReference.refersTo0();
6251 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6252 // Get arguments:
6253 Node* reference_obj = null_check_receiver();
6254 Node* other_obj = argument(1);
6255 if (stopped()) return true;
6256
6257 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6258 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6259 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6260 decorators, /*is_static*/ false, NULL);
6261 if (referent == NULL) return false;
6262
6263 // Add memory barrier to prevent commoning reads from this field
6264 // across safepoint since GC can change its value.
6265 insert_mem_bar(Op_MemBarCPUOrder);
6266
6267 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6268 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6269 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6270
6271 RegionNode* region = new RegionNode(3);
6272 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6273
6274 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6275 region->init_req(1, if_true);
6276 phi->init_req(1, intcon(1));
6277
6278 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6279 region->init_req(2, if_false);
6280 phi->init_req(2, intcon(0));
6281
6282 set_control(_gvn.transform(region));
6283 record_for_igvn(region);
6284 set_result(_gvn.transform(phi));
6285 return true;
6286 }
6287
6288
6289 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6290 DecoratorSet decorators, bool is_static,
6291 ciInstanceKlass* fromKls) {
6292 if (fromKls == NULL) {
6293 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6294 assert(tinst != NULL, "obj is null");
6295 assert(tinst->is_loaded(), "obj is not loaded");
6296 fromKls = tinst->instance_klass();
6297 } else {
6298 assert(is_static, "only for static field access");
6299 }
6300 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6301 ciSymbol::make(fieldTypeString),
6302 is_static);
6303
6304 assert(field != NULL, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6305 if (field == NULL) return (Node *) NULL;
6306
6307 if (is_static) {
6308 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6309 fromObj = makecon(tip);
6310 }
6311
6312 // Next code copied from Parse::do_get_xxx():
6313
6314 // Compute address and memory type.
6315 int offset = field->offset_in_bytes();
6316 bool is_vol = field->is_volatile();
6317 ciType* field_klass = field->type();
6318 assert(field_klass->is_loaded(), "should be loaded");
6319 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6320 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6321 BasicType bt = field->layout_type();
6322
6323 // Build the resultant type of the load
6324 const Type *type;
6325 if (bt == T_OBJECT) {
6326 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6327 } else {
6328 type = Type::get_const_basic_type(bt);
6329 }
6330
6331 if (is_vol) {
6332 decorators |= MO_SEQ_CST;
6333 }
6334
6335 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6336 }
6337
6338 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6339 bool is_exact /* true */, bool is_static /* false */,
6340 ciInstanceKlass * fromKls /* NULL */) {
6341 if (fromKls == NULL) {
6342 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6343 assert(tinst != NULL, "obj is null");
6344 assert(tinst->is_loaded(), "obj is not loaded");
6345 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6346 fromKls = tinst->instance_klass();
6347 }
6348 else {
6349 assert(is_static, "only for static field access");
6350 }
6351 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6352 ciSymbol::make(fieldTypeString),
6353 is_static);
6354
6355 assert(field != NULL, "undefined field");
6356 assert(!field->is_volatile(), "not defined for volatile fields");
6357
6358 if (is_static) {
6359 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6360 fromObj = makecon(tip);
6361 }
6362
6363 // Next code copied from Parse::do_get_xxx():
6364
6365 // Compute address and memory type.
6366 int offset = field->offset_in_bytes();
6367 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6368
6369 return adr;
6370 }
6371
6372 //------------------------------inline_aescrypt_Block-----------------------
6373 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6374 address stubAddr = NULL;
6375 const char *stubName;
6376 assert(UseAES, "need AES instruction support");
6377
6378 switch(id) {
6379 case vmIntrinsics::_aescrypt_encryptBlock:
6380 stubAddr = StubRoutines::aescrypt_encryptBlock();
6381 stubName = "aescrypt_encryptBlock";
6382 break;
6383 case vmIntrinsics::_aescrypt_decryptBlock:
6384 stubAddr = StubRoutines::aescrypt_decryptBlock();
6385 stubName = "aescrypt_decryptBlock";
6386 break;
6387 default:
6388 break;
6389 }
6390 if (stubAddr == NULL) return false;
6391
6392 Node* aescrypt_object = argument(0);
6393 Node* src = argument(1);
6394 Node* src_offset = argument(2);
6395 Node* dest = argument(3);
6396 Node* dest_offset = argument(4);
6397
6398 src = must_be_not_null(src, true);
6399 dest = must_be_not_null(dest, true);
6400
6401 // (1) src and dest are arrays.
6402 const Type* src_type = src->Value(&_gvn);
6403 const Type* dest_type = dest->Value(&_gvn);
6404 const TypeAryPtr* top_src = src_type->isa_aryptr();
6405 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6406 assert (top_src != NULL && top_src->elem() != Type::BOTTOM && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6407
6408 // for the quick and dirty code we will skip all the checks.
6409 // we are just trying to get the call to be generated.
6410 Node* src_start = src;
6411 Node* dest_start = dest;
6412 if (src_offset != NULL || dest_offset != NULL) {
6413 assert(src_offset != NULL && dest_offset != NULL, "");
6414 src_start = array_element_address(src, src_offset, T_BYTE);
6415 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6416 }
6417
6418 // now need to get the start of its expanded key array
6419 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6420 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6421 if (k_start == NULL) return false;
6422
6423 // Call the stub.
6424 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6425 stubAddr, stubName, TypePtr::BOTTOM,
6426 src_start, dest_start, k_start);
6427
6428 return true;
6429 }
6430
6431 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6432 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6433 address stubAddr = NULL;
6434 const char *stubName = NULL;
6435
6436 assert(UseAES, "need AES instruction support");
6437
6438 switch(id) {
6439 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6440 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6441 stubName = "cipherBlockChaining_encryptAESCrypt";
6442 break;
6443 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6444 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6445 stubName = "cipherBlockChaining_decryptAESCrypt";
6446 break;
6447 default:
6448 break;
6449 }
6450 if (stubAddr == NULL) return false;
6451
6452 Node* cipherBlockChaining_object = argument(0);
6453 Node* src = argument(1);
6454 Node* src_offset = argument(2);
6455 Node* len = argument(3);
6456 Node* dest = argument(4);
6457 Node* dest_offset = argument(5);
6458
6459 src = must_be_not_null(src, false);
6460 dest = must_be_not_null(dest, false);
6461
6462 // (1) src and dest are arrays.
6463 const Type* src_type = src->Value(&_gvn);
6464 const Type* dest_type = dest->Value(&_gvn);
6465 const TypeAryPtr* top_src = src_type->isa_aryptr();
6466 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6467 assert (top_src != NULL && top_src->elem() != Type::BOTTOM
6468 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6469
6470 // checks are the responsibility of the caller
6471 Node* src_start = src;
6472 Node* dest_start = dest;
6473 if (src_offset != NULL || dest_offset != NULL) {
6474 assert(src_offset != NULL && dest_offset != NULL, "");
6475 src_start = array_element_address(src, src_offset, T_BYTE);
6476 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6477 }
6478
6479 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6480 // (because of the predicated logic executed earlier).
6481 // so we cast it here safely.
6482 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6483
6484 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6485 if (embeddedCipherObj == NULL) return false;
6486
6487 // cast it to what we know it will be at runtime
6488 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6489 assert(tinst != NULL, "CBC obj is null");
6490 assert(tinst->is_loaded(), "CBC obj is not loaded");
6491 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6492 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6493
6494 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6495 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6496 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6497 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6498 aescrypt_object = _gvn.transform(aescrypt_object);
6499
6500 // we need to get the start of the aescrypt_object's expanded key array
6501 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6502 if (k_start == NULL) return false;
6503
6504 // similarly, get the start address of the r vector
6505 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6506 if (objRvec == NULL) return false;
6507 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6508
6509 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6510 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6511 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6512 stubAddr, stubName, TypePtr::BOTTOM,
6513 src_start, dest_start, k_start, r_start, len);
6514
6515 // return cipher length (int)
6516 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6517 set_result(retvalue);
6518 return true;
6519 }
6520
6521 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6522 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6523 address stubAddr = NULL;
6524 const char *stubName = NULL;
6525
6526 assert(UseAES, "need AES instruction support");
6527
6528 switch (id) {
6529 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6530 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6531 stubName = "electronicCodeBook_encryptAESCrypt";
6532 break;
6533 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6534 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6535 stubName = "electronicCodeBook_decryptAESCrypt";
6536 break;
6537 default:
6538 break;
6539 }
6540
6541 if (stubAddr == NULL) return false;
6542
6543 Node* electronicCodeBook_object = argument(0);
6544 Node* src = argument(1);
6545 Node* src_offset = argument(2);
6546 Node* len = argument(3);
6547 Node* dest = argument(4);
6548 Node* dest_offset = argument(5);
6549
6550 // (1) src and dest are arrays.
6551 const Type* src_type = src->Value(&_gvn);
6552 const Type* dest_type = dest->Value(&_gvn);
6553 const TypeAryPtr* top_src = src_type->isa_aryptr();
6554 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6555 assert(top_src != NULL && top_src->elem() != Type::BOTTOM
6556 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6557
6558 // checks are the responsibility of the caller
6559 Node* src_start = src;
6560 Node* dest_start = dest;
6561 if (src_offset != NULL || dest_offset != NULL) {
6562 assert(src_offset != NULL && dest_offset != NULL, "");
6563 src_start = array_element_address(src, src_offset, T_BYTE);
6564 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6565 }
6566
6567 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6568 // (because of the predicated logic executed earlier).
6569 // so we cast it here safely.
6570 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6571
6572 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6573 if (embeddedCipherObj == NULL) return false;
6574
6575 // cast it to what we know it will be at runtime
6576 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6577 assert(tinst != NULL, "ECB obj is null");
6578 assert(tinst->is_loaded(), "ECB obj is not loaded");
6579 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6580 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6581
6582 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6583 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6584 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6585 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6586 aescrypt_object = _gvn.transform(aescrypt_object);
6587
6588 // we need to get the start of the aescrypt_object's expanded key array
6589 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6590 if (k_start == NULL) return false;
6591
6592 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6593 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6594 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6595 stubAddr, stubName, TypePtr::BOTTOM,
6596 src_start, dest_start, k_start, len);
6597
6598 // return cipher length (int)
6599 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6600 set_result(retvalue);
6601 return true;
6602 }
6603
6604 //------------------------------inline_counterMode_AESCrypt-----------------------
6605 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6606 assert(UseAES, "need AES instruction support");
6607 if (!UseAESCTRIntrinsics) return false;
6608
6609 address stubAddr = NULL;
6610 const char *stubName = NULL;
6611 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6612 stubAddr = StubRoutines::counterMode_AESCrypt();
6613 stubName = "counterMode_AESCrypt";
6614 }
6615 if (stubAddr == NULL) return false;
6616
6617 Node* counterMode_object = argument(0);
6618 Node* src = argument(1);
6619 Node* src_offset = argument(2);
6620 Node* len = argument(3);
6621 Node* dest = argument(4);
6622 Node* dest_offset = argument(5);
6623
6624 // (1) src and dest are arrays.
6625 const Type* src_type = src->Value(&_gvn);
6626 const Type* dest_type = dest->Value(&_gvn);
6627 const TypeAryPtr* top_src = src_type->isa_aryptr();
6628 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6629 assert(top_src != NULL && top_src->elem() != Type::BOTTOM &&
6630 top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6631
6632 // checks are the responsibility of the caller
6633 Node* src_start = src;
6634 Node* dest_start = dest;
6635 if (src_offset != NULL || dest_offset != NULL) {
6636 assert(src_offset != NULL && dest_offset != NULL, "");
6637 src_start = array_element_address(src, src_offset, T_BYTE);
6638 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6639 }
6640
6641 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6642 // (because of the predicated logic executed earlier).
6643 // so we cast it here safely.
6644 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6645 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6646 if (embeddedCipherObj == NULL) return false;
6647 // cast it to what we know it will be at runtime
6648 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6649 assert(tinst != NULL, "CTR obj is null");
6650 assert(tinst->is_loaded(), "CTR obj is not loaded");
6651 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6652 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6653 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6654 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6655 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6656 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6657 aescrypt_object = _gvn.transform(aescrypt_object);
6658 // we need to get the start of the aescrypt_object's expanded key array
6659 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6660 if (k_start == NULL) return false;
6661 // similarly, get the start address of the r vector
6662 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6663 if (obj_counter == NULL) return false;
6664 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6665
6666 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6667 if (saved_encCounter == NULL) return false;
6668 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6669 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6670
6671 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6672 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6673 OptoRuntime::counterMode_aescrypt_Type(),
6674 stubAddr, stubName, TypePtr::BOTTOM,
6675 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6676
6677 // return cipher length (int)
6678 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6679 set_result(retvalue);
6680 return true;
6681 }
6682
6683 //------------------------------get_key_start_from_aescrypt_object-----------------------
6684 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6685 #if defined(PPC64) || defined(S390)
6686 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6687 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6688 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6689 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6690 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6691 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6692 if (objSessionK == NULL) {
6693 return (Node *) NULL;
6694 }
6695 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6696 #else
6697 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6698 #endif // PPC64
6699 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6700 if (objAESCryptKey == NULL) return (Node *) NULL;
6701
6702 // now have the array, need to get the start address of the K array
6703 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6704 return k_start;
6705 }
6706
6707 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6708 // Return node representing slow path of predicate check.
6709 // the pseudo code we want to emulate with this predicate is:
6710 // for encryption:
6711 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6712 // for decryption:
6713 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6714 // note cipher==plain is more conservative than the original java code but that's OK
6715 //
6716 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6717 // The receiver was checked for NULL already.
6718 Node* objCBC = argument(0);
6719
6720 Node* src = argument(1);
6721 Node* dest = argument(4);
6722
6723 // Load embeddedCipher field of CipherBlockChaining object.
6724 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6725
6726 // get AESCrypt klass for instanceOf check
6727 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6728 // will have same classloader as CipherBlockChaining object
6729 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6730 assert(tinst != NULL, "CBCobj is null");
6731 assert(tinst->is_loaded(), "CBCobj is not loaded");
6732
6733 // we want to do an instanceof comparison against the AESCrypt class
6734 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6735 if (!klass_AESCrypt->is_loaded()) {
6736 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6737 Node* ctrl = control();
6738 set_control(top()); // no regular fast path
6739 return ctrl;
6740 }
6741
6742 src = must_be_not_null(src, true);
6743 dest = must_be_not_null(dest, true);
6744
6745 // Resolve oops to stable for CmpP below.
6746 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6747
6748 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6749 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6750 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6751
6752 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6753
6754 // for encryption, we are done
6755 if (!decrypting)
6756 return instof_false; // even if it is NULL
6757
6758 // for decryption, we need to add a further check to avoid
6759 // taking the intrinsic path when cipher and plain are the same
6760 // see the original java code for why.
6761 RegionNode* region = new RegionNode(3);
6762 region->init_req(1, instof_false);
6763
6764 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6765 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6766 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6767 region->init_req(2, src_dest_conjoint);
6768
6769 record_for_igvn(region);
6770 return _gvn.transform(region);
6771 }
6772
6773 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6774 // Return node representing slow path of predicate check.
6775 // the pseudo code we want to emulate with this predicate is:
6776 // for encryption:
6777 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6778 // for decryption:
6779 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6780 // note cipher==plain is more conservative than the original java code but that's OK
6781 //
6782 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6783 // The receiver was checked for NULL already.
6784 Node* objECB = argument(0);
6785
6786 // Load embeddedCipher field of ElectronicCodeBook object.
6787 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6788
6789 // get AESCrypt klass for instanceOf check
6790 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6791 // will have same classloader as ElectronicCodeBook object
6792 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6793 assert(tinst != NULL, "ECBobj is null");
6794 assert(tinst->is_loaded(), "ECBobj is not loaded");
6795
6796 // we want to do an instanceof comparison against the AESCrypt class
6797 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6798 if (!klass_AESCrypt->is_loaded()) {
6799 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6800 Node* ctrl = control();
6801 set_control(top()); // no regular fast path
6802 return ctrl;
6803 }
6804 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6805
6806 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6807 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6808 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6809
6810 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6811
6812 // for encryption, we are done
6813 if (!decrypting)
6814 return instof_false; // even if it is NULL
6815
6816 // for decryption, we need to add a further check to avoid
6817 // taking the intrinsic path when cipher and plain are the same
6818 // see the original java code for why.
6819 RegionNode* region = new RegionNode(3);
6820 region->init_req(1, instof_false);
6821 Node* src = argument(1);
6822 Node* dest = argument(4);
6823 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6824 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6825 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6826 region->init_req(2, src_dest_conjoint);
6827
6828 record_for_igvn(region);
6829 return _gvn.transform(region);
6830 }
6831
6832 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6833 // Return node representing slow path of predicate check.
6834 // the pseudo code we want to emulate with this predicate is:
6835 // for encryption:
6836 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6837 // for decryption:
6838 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6839 // note cipher==plain is more conservative than the original java code but that's OK
6840 //
6841
6842 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6843 // The receiver was checked for NULL already.
6844 Node* objCTR = argument(0);
6845
6846 // Load embeddedCipher field of CipherBlockChaining object.
6847 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6848
6849 // get AESCrypt klass for instanceOf check
6850 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6851 // will have same classloader as CipherBlockChaining object
6852 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6853 assert(tinst != NULL, "CTRobj is null");
6854 assert(tinst->is_loaded(), "CTRobj is not loaded");
6855
6856 // we want to do an instanceof comparison against the AESCrypt class
6857 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6858 if (!klass_AESCrypt->is_loaded()) {
6859 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6860 Node* ctrl = control();
6861 set_control(top()); // no regular fast path
6862 return ctrl;
6863 }
6864
6865 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6866 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6867 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6868 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6869 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6870
6871 return instof_false; // even if it is NULL
6872 }
6873
6874 //------------------------------inline_ghash_processBlocks
6875 bool LibraryCallKit::inline_ghash_processBlocks() {
6876 address stubAddr;
6877 const char *stubName;
6878 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6879
6880 stubAddr = StubRoutines::ghash_processBlocks();
6881 stubName = "ghash_processBlocks";
6882
6883 Node* data = argument(0);
6884 Node* offset = argument(1);
6885 Node* len = argument(2);
6886 Node* state = argument(3);
6887 Node* subkeyH = argument(4);
6888
6889 state = must_be_not_null(state, true);
6890 subkeyH = must_be_not_null(subkeyH, true);
6891 data = must_be_not_null(data, true);
6892
6893 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6894 assert(state_start, "state is NULL");
6895 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6896 assert(subkeyH_start, "subkeyH is NULL");
6897 Node* data_start = array_element_address(data, offset, T_BYTE);
6898 assert(data_start, "data is NULL");
6899
6900 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6901 OptoRuntime::ghash_processBlocks_Type(),
6902 stubAddr, stubName, TypePtr::BOTTOM,
6903 state_start, subkeyH_start, data_start, len);
6904 return true;
6905 }
6906
6907 bool LibraryCallKit::inline_base64_encodeBlock() {
6908 address stubAddr;
6909 const char *stubName;
6910 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6911 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6912 stubAddr = StubRoutines::base64_encodeBlock();
6913 stubName = "encodeBlock";
6914
6915 if (!stubAddr) return false;
6916 Node* base64obj = argument(0);
6917 Node* src = argument(1);
6918 Node* offset = argument(2);
6919 Node* len = argument(3);
6920 Node* dest = argument(4);
6921 Node* dp = argument(5);
6922 Node* isURL = argument(6);
6923
6924 src = must_be_not_null(src, true);
6925 dest = must_be_not_null(dest, true);
6926
6927 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6928 assert(src_start, "source array is NULL");
6929 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6930 assert(dest_start, "destination array is NULL");
6931
6932 Node* base64 = make_runtime_call(RC_LEAF,
6933 OptoRuntime::base64_encodeBlock_Type(),
6934 stubAddr, stubName, TypePtr::BOTTOM,
6935 src_start, offset, len, dest_start, dp, isURL);
6936 return true;
6937 }
6938
6939 bool LibraryCallKit::inline_base64_decodeBlock() {
6940 address stubAddr;
6941 const char *stubName;
6942 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6943 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6944 stubAddr = StubRoutines::base64_decodeBlock();
6945 stubName = "decodeBlock";
6946
6947 if (!stubAddr) return false;
6948 Node* base64obj = argument(0);
6949 Node* src = argument(1);
6950 Node* src_offset = argument(2);
6951 Node* len = argument(3);
6952 Node* dest = argument(4);
6953 Node* dest_offset = argument(5);
6954 Node* isURL = argument(6);
6955 Node* isMIME = argument(7);
6956
6957 src = must_be_not_null(src, true);
6958 dest = must_be_not_null(dest, true);
6959
6960 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6961 assert(src_start, "source array is NULL");
6962 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6963 assert(dest_start, "destination array is NULL");
6964
6965 Node* call = make_runtime_call(RC_LEAF,
6966 OptoRuntime::base64_decodeBlock_Type(),
6967 stubAddr, stubName, TypePtr::BOTTOM,
6968 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6969 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6970 set_result(result);
6971 return true;
6972 }
6973
6974 //------------------------------inline_digestBase_implCompress-----------------------
6975 //
6976 // Calculate MD5 for single-block byte[] array.
6977 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6978 //
6979 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6980 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6981 //
6982 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6983 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6984 //
6985 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6986 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6987 //
6988 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6989 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6990 //
6991 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6992 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6993
6994 Node* digestBase_obj = argument(0);
6995 Node* src = argument(1); // type oop
6996 Node* ofs = argument(2); // type int
6997
6998 const Type* src_type = src->Value(&_gvn);
6999 const TypeAryPtr* top_src = src_type->isa_aryptr();
7000 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7001 // failed array check
7002 return false;
7003 }
7004 // Figure out the size and type of the elements we will be copying.
7005 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7006 if (src_elem != T_BYTE) {
7007 return false;
7008 }
7009 // 'src_start' points to src array + offset
7010 src = must_be_not_null(src, true);
7011 Node* src_start = array_element_address(src, ofs, src_elem);
7012 Node* state = NULL;
7013 Node* digest_length = NULL;
7014 address stubAddr;
7015 const char *stubName;
7016
7017 switch(id) {
7018 case vmIntrinsics::_md5_implCompress:
7019 assert(UseMD5Intrinsics, "need MD5 instruction support");
7020 state = get_state_from_digest_object(digestBase_obj, T_INT);
7021 stubAddr = StubRoutines::md5_implCompress();
7022 stubName = "md5_implCompress";
7023 break;
7024 case vmIntrinsics::_sha_implCompress:
7025 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7026 state = get_state_from_digest_object(digestBase_obj, T_INT);
7027 stubAddr = StubRoutines::sha1_implCompress();
7028 stubName = "sha1_implCompress";
7029 break;
7030 case vmIntrinsics::_sha2_implCompress:
7031 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7032 state = get_state_from_digest_object(digestBase_obj, T_INT);
7033 stubAddr = StubRoutines::sha256_implCompress();
7034 stubName = "sha256_implCompress";
7035 break;
7036 case vmIntrinsics::_sha5_implCompress:
7037 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7038 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7039 stubAddr = StubRoutines::sha512_implCompress();
7040 stubName = "sha512_implCompress";
7041 break;
7042 case vmIntrinsics::_sha3_implCompress:
7043 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7044 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7045 stubAddr = StubRoutines::sha3_implCompress();
7046 stubName = "sha3_implCompress";
7047 digest_length = get_digest_length_from_digest_object(digestBase_obj);
7048 if (digest_length == NULL) return false;
7049 break;
7050 default:
7051 fatal_unexpected_iid(id);
7052 return false;
7053 }
7054 if (state == NULL) return false;
7055
7056 assert(stubAddr != NULL, "Stub is generated");
7057 if (stubAddr == NULL) return false;
7058
7059 // Call the stub.
7060 Node* call;
7061 if (digest_length == NULL) {
7062 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7063 stubAddr, stubName, TypePtr::BOTTOM,
7064 src_start, state);
7065 } else {
7066 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7067 stubAddr, stubName, TypePtr::BOTTOM,
7068 src_start, state, digest_length);
7069 }
7070
7071 return true;
7072 }
7073
7074 //------------------------------inline_digestBase_implCompressMB-----------------------
7075 //
7076 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7077 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7078 //
7079 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7080 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7081 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7082 assert((uint)predicate < 5, "sanity");
7083 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7084
7085 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7086 Node* src = argument(1); // byte[] array
7087 Node* ofs = argument(2); // type int
7088 Node* limit = argument(3); // type int
7089
7090 const Type* src_type = src->Value(&_gvn);
7091 const TypeAryPtr* top_src = src_type->isa_aryptr();
7092 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7093 // failed array check
7094 return false;
7095 }
7096 // Figure out the size and type of the elements we will be copying.
7097 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7098 if (src_elem != T_BYTE) {
7099 return false;
7100 }
7101 // 'src_start' points to src array + offset
7102 src = must_be_not_null(src, false);
7103 Node* src_start = array_element_address(src, ofs, src_elem);
7104
7105 const char* klass_digestBase_name = NULL;
7106 const char* stub_name = NULL;
7107 address stub_addr = NULL;
7108 BasicType elem_type = T_INT;
7109
7110 switch (predicate) {
7111 case 0:
7112 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7113 klass_digestBase_name = "sun/security/provider/MD5";
7114 stub_name = "md5_implCompressMB";
7115 stub_addr = StubRoutines::md5_implCompressMB();
7116 }
7117 break;
7118 case 1:
7119 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7120 klass_digestBase_name = "sun/security/provider/SHA";
7121 stub_name = "sha1_implCompressMB";
7122 stub_addr = StubRoutines::sha1_implCompressMB();
7123 }
7124 break;
7125 case 2:
7126 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7127 klass_digestBase_name = "sun/security/provider/SHA2";
7128 stub_name = "sha256_implCompressMB";
7129 stub_addr = StubRoutines::sha256_implCompressMB();
7130 }
7131 break;
7132 case 3:
7133 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7134 klass_digestBase_name = "sun/security/provider/SHA5";
7135 stub_name = "sha512_implCompressMB";
7136 stub_addr = StubRoutines::sha512_implCompressMB();
7137 elem_type = T_LONG;
7138 }
7139 break;
7140 case 4:
7141 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7142 klass_digestBase_name = "sun/security/provider/SHA3";
7143 stub_name = "sha3_implCompressMB";
7144 stub_addr = StubRoutines::sha3_implCompressMB();
7145 elem_type = T_BYTE;
7146 }
7147 break;
7148 default:
7149 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7150 }
7151 if (klass_digestBase_name != NULL) {
7152 assert(stub_addr != NULL, "Stub is generated");
7153 if (stub_addr == NULL) return false;
7154
7155 // get DigestBase klass to lookup for SHA klass
7156 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7157 assert(tinst != NULL, "digestBase_obj is not instance???");
7158 assert(tinst->is_loaded(), "DigestBase is not loaded");
7159
7160 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7161 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7162 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7163 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7164 }
7165 return false;
7166 }
7167
7168 //------------------------------inline_digestBase_implCompressMB-----------------------
7169 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7170 BasicType elem_type, address stubAddr, const char *stubName,
7171 Node* src_start, Node* ofs, Node* limit) {
7172 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7173 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7174 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7175 digest_obj = _gvn.transform(digest_obj);
7176
7177 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7178 if (state == NULL) return false;
7179
7180 Node* digest_length = NULL;
7181 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7182 digest_length = get_digest_length_from_digest_object(digest_obj);
7183 if (digest_length == NULL) return false;
7184 }
7185
7186 // Call the stub.
7187 Node* call;
7188 if (digest_length == NULL) {
7189 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7190 OptoRuntime::digestBase_implCompressMB_Type(false),
7191 stubAddr, stubName, TypePtr::BOTTOM,
7192 src_start, state, ofs, limit);
7193 } else {
7194 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7195 OptoRuntime::digestBase_implCompressMB_Type(true),
7196 stubAddr, stubName, TypePtr::BOTTOM,
7197 src_start, state, digest_length, ofs, limit);
7198 }
7199
7200 // return ofs (int)
7201 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7202 set_result(result);
7203
7204 return true;
7205 }
7206
7207 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7208 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7209 assert(UseAES, "need AES instruction support");
7210 address stubAddr = NULL;
7211 const char *stubName = NULL;
7212 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7213 stubName = "galoisCounterMode_AESCrypt";
7214
7215 if (stubAddr == NULL) return false;
7216
7217 Node* in = argument(0);
7218 Node* inOfs = argument(1);
7219 Node* len = argument(2);
7220 Node* ct = argument(3);
7221 Node* ctOfs = argument(4);
7222 Node* out = argument(5);
7223 Node* outOfs = argument(6);
7224 Node* gctr_object = argument(7);
7225 Node* ghash_object = argument(8);
7226
7227 // (1) in, ct and out are arrays.
7228 const Type* in_type = in->Value(&_gvn);
7229 const Type* ct_type = ct->Value(&_gvn);
7230 const Type* out_type = out->Value(&_gvn);
7231 const TypeAryPtr* top_in = in_type->isa_aryptr();
7232 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7233 const TypeAryPtr* top_out = out_type->isa_aryptr();
7234 assert(top_in != NULL && top_in->elem() != Type::BOTTOM &&
7235 top_ct != NULL && top_ct->elem() != Type::BOTTOM &&
7236 top_out != NULL && top_out->elem() != Type::BOTTOM, "args are strange");
7237
7238 // checks are the responsibility of the caller
7239 Node* in_start = in;
7240 Node* ct_start = ct;
7241 Node* out_start = out;
7242 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7243 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7244 in_start = array_element_address(in, inOfs, T_BYTE);
7245 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7246 out_start = array_element_address(out, outOfs, T_BYTE);
7247 }
7248
7249 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7250 // (because of the predicated logic executed earlier).
7251 // so we cast it here safely.
7252 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7253 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7254 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7255 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7256 Node* state = load_field_from_object(ghash_object, "state", "[J");
7257
7258 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7259 return false;
7260 }
7261 // cast it to what we know it will be at runtime
7262 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7263 assert(tinst != NULL, "GCTR obj is null");
7264 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7265 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7266 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7267 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7268 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7269 const TypeOopPtr* xtype = aklass->as_instance_type();
7270 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7271 aescrypt_object = _gvn.transform(aescrypt_object);
7272 // we need to get the start of the aescrypt_object's expanded key array
7273 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7274 if (k_start == NULL) return false;
7275 // similarly, get the start address of the r vector
7276 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7277 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7278 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7279
7280
7281 // Call the stub, passing params
7282 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7283 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7284 stubAddr, stubName, TypePtr::BOTTOM,
7285 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7286
7287 // return cipher length (int)
7288 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7289 set_result(retvalue);
7290
7291 return true;
7292 }
7293
7294 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7295 // Return node representing slow path of predicate check.
7296 // the pseudo code we want to emulate with this predicate is:
7297 // for encryption:
7298 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7299 // for decryption:
7300 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7301 // note cipher==plain is more conservative than the original java code but that's OK
7302 //
7303
7304 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7305 // The receiver was checked for NULL already.
7306 Node* objGCTR = argument(7);
7307 // Load embeddedCipher field of GCTR object.
7308 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7309 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7310
7311 // get AESCrypt klass for instanceOf check
7312 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7313 // will have same classloader as CipherBlockChaining object
7314 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7315 assert(tinst != NULL, "GCTR obj is null");
7316 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7317
7318 // we want to do an instanceof comparison against the AESCrypt class
7319 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7320 if (!klass_AESCrypt->is_loaded()) {
7321 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7322 Node* ctrl = control();
7323 set_control(top()); // no regular fast path
7324 return ctrl;
7325 }
7326
7327 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7328 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7329 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7330 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7331 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7332
7333 return instof_false; // even if it is NULL
7334 }
7335
7336 //------------------------------get_state_from_digest_object-----------------------
7337 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7338 const char* state_type;
7339 switch (elem_type) {
7340 case T_BYTE: state_type = "[B"; break;
7341 case T_INT: state_type = "[I"; break;
7342 case T_LONG: state_type = "[J"; break;
7343 default: ShouldNotReachHere();
7344 }
7345 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7346 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7347 if (digest_state == NULL) return (Node *) NULL;
7348
7349 // now have the array, need to get the start address of the state array
7350 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7351 return state;
7352 }
7353
7354 //------------------------------get_digest_length_from_sha3_object----------------------------------
7355 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7356 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7357 assert (digest_length != NULL, "sanity");
7358 return digest_length;
7359 }
7360
7361 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7362 // Return node representing slow path of predicate check.
7363 // the pseudo code we want to emulate with this predicate is:
7364 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7365 //
7366 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7367 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7368 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7369 assert((uint)predicate < 5, "sanity");
7370
7371 // The receiver was checked for NULL already.
7372 Node* digestBaseObj = argument(0);
7373
7374 // get DigestBase klass for instanceOf check
7375 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7376 assert(tinst != NULL, "digestBaseObj is null");
7377 assert(tinst->is_loaded(), "DigestBase is not loaded");
7378
7379 const char* klass_name = NULL;
7380 switch (predicate) {
7381 case 0:
7382 if (UseMD5Intrinsics) {
7383 // we want to do an instanceof comparison against the MD5 class
7384 klass_name = "sun/security/provider/MD5";
7385 }
7386 break;
7387 case 1:
7388 if (UseSHA1Intrinsics) {
7389 // we want to do an instanceof comparison against the SHA class
7390 klass_name = "sun/security/provider/SHA";
7391 }
7392 break;
7393 case 2:
7394 if (UseSHA256Intrinsics) {
7395 // we want to do an instanceof comparison against the SHA2 class
7396 klass_name = "sun/security/provider/SHA2";
7397 }
7398 break;
7399 case 3:
7400 if (UseSHA512Intrinsics) {
7401 // we want to do an instanceof comparison against the SHA5 class
7402 klass_name = "sun/security/provider/SHA5";
7403 }
7404 break;
7405 case 4:
7406 if (UseSHA3Intrinsics) {
7407 // we want to do an instanceof comparison against the SHA3 class
7408 klass_name = "sun/security/provider/SHA3";
7409 }
7410 break;
7411 default:
7412 fatal("unknown SHA intrinsic predicate: %d", predicate);
7413 }
7414
7415 ciKlass* klass = NULL;
7416 if (klass_name != NULL) {
7417 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7418 }
7419 if ((klass == NULL) || !klass->is_loaded()) {
7420 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7421 Node* ctrl = control();
7422 set_control(top()); // no intrinsic path
7423 return ctrl;
7424 }
7425 ciInstanceKlass* instklass = klass->as_instance_klass();
7426
7427 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7428 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7429 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7430 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7431
7432 return instof_false; // even if it is NULL
7433 }
7434
7435 //-------------inline_fma-----------------------------------
7436 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7437 Node *a = NULL;
7438 Node *b = NULL;
7439 Node *c = NULL;
7440 Node* result = NULL;
7441 switch (id) {
7442 case vmIntrinsics::_fmaD:
7443 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7444 // no receiver since it is static method
7445 a = round_double_node(argument(0));
7446 b = round_double_node(argument(2));
7447 c = round_double_node(argument(4));
7448 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7449 break;
7450 case vmIntrinsics::_fmaF:
7451 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7452 a = argument(0);
7453 b = argument(1);
7454 c = argument(2);
7455 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7456 break;
7457 default:
7458 fatal_unexpected_iid(id); break;
7459 }
7460 set_result(result);
7461 return true;
7462 }
7463
7464 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7465 // argument(0) is receiver
7466 Node* codePoint = argument(1);
7467 Node* n = NULL;
7468
7469 switch (id) {
7470 case vmIntrinsics::_isDigit :
7471 n = new DigitNode(control(), codePoint);
7472 break;
7473 case vmIntrinsics::_isLowerCase :
7474 n = new LowerCaseNode(control(), codePoint);
7475 break;
7476 case vmIntrinsics::_isUpperCase :
7477 n = new UpperCaseNode(control(), codePoint);
7478 break;
7479 case vmIntrinsics::_isWhitespace :
7480 n = new WhitespaceNode(control(), codePoint);
7481 break;
7482 default:
7483 fatal_unexpected_iid(id);
7484 }
7485
7486 set_result(_gvn.transform(n));
7487 return true;
7488 }
7489
7490 //------------------------------inline_fp_min_max------------------------------
7491 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7492 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7493
7494 // The intrinsic should be used only when the API branches aren't predictable,
7495 // the last one performing the most important comparison. The following heuristic
7496 // uses the branch statistics to eventually bail out if necessary.
7497
7498 ciMethodData *md = callee()->method_data();
7499
7500 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7501 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7502
7503 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7504 // Bail out if the call-site didn't contribute enough to the statistics.
7505 return false;
7506 }
7507
7508 uint taken = 0, not_taken = 0;
7509
7510 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7511 if (p->is_BranchData()) {
7512 taken = ((ciBranchData*)p)->taken();
7513 not_taken = ((ciBranchData*)p)->not_taken();
7514 }
7515 }
7516
7517 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7518 balance = balance < 0 ? -balance : balance;
7519 if ( balance > 0.2 ) {
7520 // Bail out if the most important branch is predictable enough.
7521 return false;
7522 }
7523 }
7524 */
7525
7526 Node *a = NULL;
7527 Node *b = NULL;
7528 Node *n = NULL;
7529 switch (id) {
7530 case vmIntrinsics::_maxF:
7531 case vmIntrinsics::_minF:
7532 case vmIntrinsics::_maxF_strict:
7533 case vmIntrinsics::_minF_strict:
7534 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7535 a = argument(0);
7536 b = argument(1);
7537 break;
7538 case vmIntrinsics::_maxD:
7539 case vmIntrinsics::_minD:
7540 case vmIntrinsics::_maxD_strict:
7541 case vmIntrinsics::_minD_strict:
7542 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7543 a = round_double_node(argument(0));
7544 b = round_double_node(argument(2));
7545 break;
7546 default:
7547 fatal_unexpected_iid(id);
7548 break;
7549 }
7550 switch (id) {
7551 case vmIntrinsics::_maxF:
7552 case vmIntrinsics::_maxF_strict:
7553 n = new MaxFNode(a, b);
7554 break;
7555 case vmIntrinsics::_minF:
7556 case vmIntrinsics::_minF_strict:
7557 n = new MinFNode(a, b);
7558 break;
7559 case vmIntrinsics::_maxD:
7560 case vmIntrinsics::_maxD_strict:
7561 n = new MaxDNode(a, b);
7562 break;
7563 case vmIntrinsics::_minD:
7564 case vmIntrinsics::_minD_strict:
7565 n = new MinDNode(a, b);
7566 break;
7567 default:
7568 fatal_unexpected_iid(id);
7569 break;
7570 }
7571 set_result(_gvn.transform(n));
7572 return true;
7573 }
7574
7575 bool LibraryCallKit::inline_profileBoolean() {
7576 Node* counts = argument(1);
7577 const TypeAryPtr* ary = NULL;
7578 ciArray* aobj = NULL;
7579 if (counts->is_Con()
7580 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7581 && (aobj = ary->const_oop()->as_array()) != NULL
7582 && (aobj->length() == 2)) {
7583 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7584 jint false_cnt = aobj->element_value(0).as_int();
7585 jint true_cnt = aobj->element_value(1).as_int();
7586
7587 if (C->log() != NULL) {
7588 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7589 false_cnt, true_cnt);
7590 }
7591
7592 if (false_cnt + true_cnt == 0) {
7593 // According to profile, never executed.
7594 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7595 Deoptimization::Action_reinterpret);
7596 return true;
7597 }
7598
7599 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7600 // is a number of each value occurrences.
7601 Node* result = argument(0);
7602 if (false_cnt == 0 || true_cnt == 0) {
7603 // According to profile, one value has been never seen.
7604 int expected_val = (false_cnt == 0) ? 1 : 0;
7605
7606 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7607 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7608
7609 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7610 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7611 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7612
7613 { // Slow path: uncommon trap for never seen value and then reexecute
7614 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7615 // the value has been seen at least once.
7616 PreserveJVMState pjvms(this);
7617 PreserveReexecuteState preexecs(this);
7618 jvms()->set_should_reexecute(true);
7619
7620 set_control(slow_path);
7621 set_i_o(i_o());
7622
7623 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7624 Deoptimization::Action_reinterpret);
7625 }
7626 // The guard for never seen value enables sharpening of the result and
7627 // returning a constant. It allows to eliminate branches on the same value
7628 // later on.
7629 set_control(fast_path);
7630 result = intcon(expected_val);
7631 }
7632 // Stop profiling.
7633 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7634 // By replacing method body with profile data (represented as ProfileBooleanNode
7635 // on IR level) we effectively disable profiling.
7636 // It enables full speed execution once optimized code is generated.
7637 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7638 C->record_for_igvn(profile);
7639 set_result(profile);
7640 return true;
7641 } else {
7642 // Continue profiling.
7643 // Profile data isn't available at the moment. So, execute method's bytecode version.
7644 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7645 // is compiled and counters aren't available since corresponding MethodHandle
7646 // isn't a compile-time constant.
7647 return false;
7648 }
7649 }
7650
7651 bool LibraryCallKit::inline_isCompileConstant() {
7652 Node* n = argument(0);
7653 set_result(n->is_Con() ? intcon(1) : intcon(0));
7654 return true;
7655 }
7656
7657 //------------------------------- inline_getObjectSize/sizeOf --------------------------------------
7658 //
7659 // Calculate the runtime size of the object/array.
7660 //
7661 // long java.lang.Runtime.sizeOf(Object obj)
7662 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7663 //
7664 bool LibraryCallKit::inline_sizeOf() {
7665 Node* obj = argument(0);
7666 return inline_sizeOf_impl(obj);
7667 }
7668
7669 bool LibraryCallKit::inline_getObjectSize() {
7670 Node* obj = argument(3);
7671 return inline_sizeOf_impl(obj);
7672 }
7673
7674 bool LibraryCallKit::inline_sizeOf_impl(Node* obj) {
7675 Node* klass_node = load_object_klass(obj);
7676
7677 jint layout_con = Klass::_lh_neutral_value;
7678 Node* layout_val = get_layout_helper(klass_node, layout_con);
7679 int layout_is_con = (layout_val == NULL);
7680
7681 if (layout_is_con) {
7682 // Layout helper is constant, can figure out things at compile time.
7683
7684 if (Klass::layout_helper_is_instance(layout_con)) {
7685 // Instance case: layout_con contains the size itself.
7686 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7687 set_result(size);
7688 } else {
7689 // Array case: size is round(header + element_size*arraylength).
7690 // Since arraylength is different for every array instance, we have to
7691 // compute the whole thing at runtime.
7692
7693 Node* arr_length = load_array_length(obj);
7694
7695 int round_mask = MinObjAlignmentInBytes - 1;
7696 int hsize = Klass::layout_helper_header_size(layout_con);
7697 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7698
7699 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7700 round_mask = 0; // strength-reduce it if it goes away completely
7701 }
7702 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7703 Node* header_size = intcon(hsize + round_mask);
7704
7705 Node* lengthx = ConvI2X(arr_length);
7706 Node* headerx = ConvI2X(header_size);
7707
7708 Node* abody = lengthx;
7709 if (eshift != 0) {
7710 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7711 }
7712 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7713 if (round_mask != 0) {
7714 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7715 }
7716 size = ConvX2L(size);
7717 set_result(size);
7718 }
7719 } else {
7720 // Layout helper is not constant, need to test for array-ness at runtime.
7721
7722 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7723 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7724 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7725 record_for_igvn(result_reg);
7726
7727 Node* array_ctl = generate_array_guard(klass_node, NULL);
7728 if (array_ctl != NULL) {
7729 // Array case: size is round(header + element_size*arraylength).
7730 // Since arraylength is different for every array instance, we have to
7731 // compute the whole thing at runtime.
7732
7733 PreserveJVMState pjvms(this);
7734 set_control(array_ctl);
7735 Node* arr_length = load_array_length(obj);
7736
7737 int round_mask = MinObjAlignmentInBytes - 1;
7738 Node* mask = intcon(round_mask);
7739
7740 Node* hss = intcon(Klass::_lh_header_size_shift);
7741 Node* hsm = intcon(Klass::_lh_header_size_mask);
7742 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7743 header_size = _gvn.transform(new AndINode(header_size, hsm));
7744 header_size = _gvn.transform(new AddINode(header_size, mask));
7745
7746 // There is no need to mask or shift this value.
7747 // The semantics of LShiftINode include an implicit mask to 0x1F.
7748 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7749 Node* elem_shift = layout_val;
7750
7751 Node* lengthx = ConvI2X(arr_length);
7752 Node* headerx = ConvI2X(header_size);
7753
7754 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7755 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7756 if (round_mask != 0) {
7757 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7758 }
7759 size = ConvX2L(size);
7760
7761 result_reg->init_req(_array_path, control());
7762 result_val->init_req(_array_path, size);
7763 }
7764
7765 if (!stopped()) {
7766 // Instance case: the layout helper gives us instance size almost directly,
7767 // but we need to mask out the _lh_instance_slow_path_bit.
7768 Node* size = ConvI2X(layout_val);
7769 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7770 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7771 size = _gvn.transform(new AndXNode(size, mask));
7772 size = ConvX2L(size);
7773
7774 result_reg->init_req(_instance_path, control());
7775 result_val->init_req(_instance_path, size);
7776 }
7777
7778 set_result(result_reg, result_val);
7779 }
7780
7781 return true;
7782 }
7783
7784 //------------------------------- inline_blackhole --------------------------------------
7785 //
7786 // Make sure all arguments to this node are alive.
7787 // This matches methods that were requested to be blackholed through compile commands.
7788 //
7789 bool LibraryCallKit::inline_blackhole() {
7790 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7791 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7792 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7793
7794 // Blackhole node pinches only the control, not memory. This allows
7795 // the blackhole to be pinned in the loop that computes blackholed
7796 // values, but have no other side effects, like breaking the optimizations
7797 // across the blackhole.
7798
7799 Node* bh = _gvn.transform(new BlackholeNode(control()));
7800 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
7801
7802 // Bind call arguments as blackhole arguments to keep them alive
7803 uint nargs = callee()->arg_size();
7804 for (uint i = 0; i < nargs; i++) {
7805 bh->add_req(argument(i));
7806 }
7807
7808 return true;
7809 }
7810
7811 bool LibraryCallKit::inline_timestamp(bool serial) {
7812 insert_mem_bar(Op_MemBarCPUOrder);
7813 Node* node = serial ?
7814 _gvn.transform(new TimestampSerialNode(control())) :
7815 _gvn.transform(new TimestampNode(control()));
7816 set_result(node);
7817 insert_mem_bar(Op_MemBarCPUOrder);
7818 return true;
7819 }
7820
7821 bool LibraryCallKit::inline_addressOf() {
7822 Node* obj = argument(0);
7823 Node* raw_val = _gvn.transform(new CastP2XNode(NULL, obj));
7824 Node* long_val = ConvX2L(raw_val);
7825 set_result(long_val);
7826 return true;
7827 }