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.as_string();
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.as_string();
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: return inline_fp_conversions(intrinsic_id());
518
519 case vmIntrinsics::_floatIsInfinite:
520 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
521
522 case vmIntrinsics::_numberOfLeadingZeros_i:
523 case vmIntrinsics::_numberOfLeadingZeros_l:
524 case vmIntrinsics::_numberOfTrailingZeros_i:
525 case vmIntrinsics::_numberOfTrailingZeros_l:
526 case vmIntrinsics::_bitCount_i:
527 case vmIntrinsics::_bitCount_l:
528 case vmIntrinsics::_reverseBytes_i:
529 case vmIntrinsics::_reverseBytes_l:
530 case vmIntrinsics::_reverseBytes_s:
531 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
532
533 case vmIntrinsics::_compress_i:
534 case vmIntrinsics::_compress_l:
535 case vmIntrinsics::_expand_i:
536 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
537
538 case vmIntrinsics::_divideUnsigned_i:
539 case vmIntrinsics::_divideUnsigned_l:
540 case vmIntrinsics::_remainderUnsigned_i:
541 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
542
543 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
544
545 case vmIntrinsics::_Reference_get: return inline_reference_get();
546 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
547 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
548
549 case vmIntrinsics::_Class_cast: return inline_Class_cast();
550
551 case vmIntrinsics::_aescrypt_encryptBlock:
552 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
553
554 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
555 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
556 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
557
558 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
559 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
560 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
561
562 case vmIntrinsics::_counterMode_AESCrypt:
563 return inline_counterMode_AESCrypt(intrinsic_id());
564
565 case vmIntrinsics::_galoisCounterMode_AESCrypt:
566 return inline_galoisCounterMode_AESCrypt();
567
568 case vmIntrinsics::_md5_implCompress:
569 case vmIntrinsics::_sha_implCompress:
570 case vmIntrinsics::_sha2_implCompress:
571 case vmIntrinsics::_sha5_implCompress:
572 case vmIntrinsics::_sha3_implCompress:
573 return inline_digestBase_implCompress(intrinsic_id());
574
575 case vmIntrinsics::_digestBase_implCompressMB:
576 return inline_digestBase_implCompressMB(predicate);
577
578 case vmIntrinsics::_multiplyToLen:
579 return inline_multiplyToLen();
580
581 case vmIntrinsics::_squareToLen:
582 return inline_squareToLen();
583
584 case vmIntrinsics::_mulAdd:
585 return inline_mulAdd();
586
587 case vmIntrinsics::_montgomeryMultiply:
588 return inline_montgomeryMultiply();
589 case vmIntrinsics::_montgomerySquare:
590 return inline_montgomerySquare();
591
592 case vmIntrinsics::_bigIntegerRightShiftWorker:
593 return inline_bigIntegerShift(true);
594 case vmIntrinsics::_bigIntegerLeftShiftWorker:
595 return inline_bigIntegerShift(false);
596
597 case vmIntrinsics::_vectorizedMismatch:
598 return inline_vectorizedMismatch();
599
600 case vmIntrinsics::_ghash_processBlocks:
601 return inline_ghash_processBlocks();
602 case vmIntrinsics::_base64_encodeBlock:
603 return inline_base64_encodeBlock();
604 case vmIntrinsics::_base64_decodeBlock:
605 return inline_base64_decodeBlock();
606
607 case vmIntrinsics::_encodeISOArray:
608 case vmIntrinsics::_encodeByteISOArray:
609 return inline_encodeISOArray(false);
610 case vmIntrinsics::_encodeAsciiArray:
611 return inline_encodeISOArray(true);
612
613 case vmIntrinsics::_updateCRC32:
614 return inline_updateCRC32();
615 case vmIntrinsics::_updateBytesCRC32:
616 return inline_updateBytesCRC32();
617 case vmIntrinsics::_updateByteBufferCRC32:
618 return inline_updateByteBufferCRC32();
619
620 case vmIntrinsics::_updateBytesCRC32C:
621 return inline_updateBytesCRC32C();
622 case vmIntrinsics::_updateDirectByteBufferCRC32C:
623 return inline_updateDirectByteBufferCRC32C();
624
625 case vmIntrinsics::_updateBytesAdler32:
626 return inline_updateBytesAdler32();
627 case vmIntrinsics::_updateByteBufferAdler32:
628 return inline_updateByteBufferAdler32();
629
630 case vmIntrinsics::_profileBoolean:
631 return inline_profileBoolean();
632 case vmIntrinsics::_isCompileConstant:
633 return inline_isCompileConstant();
634
635 case vmIntrinsics::_countPositives:
636 return inline_countPositives();
637
638 case vmIntrinsics::_fmaD:
639 case vmIntrinsics::_fmaF:
640 return inline_fma(intrinsic_id());
641
642 case vmIntrinsics::_Continuation_doYield:
643 return inline_continuation_do_yield();
644
645 case vmIntrinsics::_isDigit:
646 case vmIntrinsics::_isLowerCase:
647 case vmIntrinsics::_isUpperCase:
648 case vmIntrinsics::_isWhitespace:
649 return inline_character_compare(intrinsic_id());
650
651 case vmIntrinsics::_min:
652 case vmIntrinsics::_max:
653 case vmIntrinsics::_min_strict:
654 case vmIntrinsics::_max_strict:
655 return inline_min_max(intrinsic_id());
656
657 case vmIntrinsics::_maxF:
658 case vmIntrinsics::_minF:
659 case vmIntrinsics::_maxD:
660 case vmIntrinsics::_minD:
661 case vmIntrinsics::_maxF_strict:
662 case vmIntrinsics::_minF_strict:
663 case vmIntrinsics::_maxD_strict:
664 case vmIntrinsics::_minD_strict:
665 return inline_fp_min_max(intrinsic_id());
666
667 case vmIntrinsics::_VectorUnaryOp:
668 return inline_vector_nary_operation(1);
669 case vmIntrinsics::_VectorBinaryOp:
670 return inline_vector_nary_operation(2);
671 case vmIntrinsics::_VectorTernaryOp:
672 return inline_vector_nary_operation(3);
673 case vmIntrinsics::_VectorFromBitsCoerced:
674 return inline_vector_frombits_coerced();
675 case vmIntrinsics::_VectorShuffleIota:
676 return inline_vector_shuffle_iota();
677 case vmIntrinsics::_VectorMaskOp:
678 return inline_vector_mask_operation();
679 case vmIntrinsics::_VectorShuffleToVector:
680 return inline_vector_shuffle_to_vector();
681 case vmIntrinsics::_VectorLoadOp:
682 return inline_vector_mem_operation(/*is_store=*/false);
683 case vmIntrinsics::_VectorLoadMaskedOp:
684 return inline_vector_mem_masked_operation(/*is_store*/false);
685 case vmIntrinsics::_VectorStoreOp:
686 return inline_vector_mem_operation(/*is_store=*/true);
687 case vmIntrinsics::_VectorStoreMaskedOp:
688 return inline_vector_mem_masked_operation(/*is_store=*/true);
689 case vmIntrinsics::_VectorGatherOp:
690 return inline_vector_gather_scatter(/*is_scatter*/ false);
691 case vmIntrinsics::_VectorScatterOp:
692 return inline_vector_gather_scatter(/*is_scatter*/ true);
693 case vmIntrinsics::_VectorReductionCoerced:
694 return inline_vector_reduction();
695 case vmIntrinsics::_VectorTest:
696 return inline_vector_test();
697 case vmIntrinsics::_VectorBlend:
698 return inline_vector_blend();
699 case vmIntrinsics::_VectorRearrange:
700 return inline_vector_rearrange();
701 case vmIntrinsics::_VectorCompare:
702 return inline_vector_compare();
703 case vmIntrinsics::_VectorBroadcastInt:
704 return inline_vector_broadcast_int();
705 case vmIntrinsics::_VectorConvert:
706 return inline_vector_convert();
707 case vmIntrinsics::_VectorInsert:
708 return inline_vector_insert();
709 case vmIntrinsics::_VectorExtract:
710 return inline_vector_extract();
711 case vmIntrinsics::_VectorCompressExpand:
712 return inline_vector_compress_expand();
713
714 case vmIntrinsics::_getObjectSize:
715 return inline_getObjectSize();
716
717 case vmIntrinsics::_blackhole:
718 return inline_blackhole();
719
720 default:
721 // If you get here, it may be that someone has added a new intrinsic
722 // to the list in vmIntrinsics.hpp without implementing it here.
723 #ifndef PRODUCT
724 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
725 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
726 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
727 }
728 #endif
729 return false;
730 }
731 }
732
733 Node* LibraryCallKit::try_to_predicate(int predicate) {
734 if (!jvms()->has_method()) {
735 // Root JVMState has a null method.
736 assert(map()->memory()->Opcode() == Op_Parm, "");
737 // Insert the memory aliasing node
738 set_all_memory(reset_memory());
739 }
740 assert(merged_memory(), "");
741
742 switch (intrinsic_id()) {
743 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
744 return inline_cipherBlockChaining_AESCrypt_predicate(false);
745 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
746 return inline_cipherBlockChaining_AESCrypt_predicate(true);
747 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
748 return inline_electronicCodeBook_AESCrypt_predicate(false);
749 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
750 return inline_electronicCodeBook_AESCrypt_predicate(true);
751 case vmIntrinsics::_counterMode_AESCrypt:
752 return inline_counterMode_AESCrypt_predicate();
753 case vmIntrinsics::_digestBase_implCompressMB:
754 return inline_digestBase_implCompressMB_predicate(predicate);
755 case vmIntrinsics::_galoisCounterMode_AESCrypt:
756 return inline_galoisCounterMode_AESCrypt_predicate();
757
758 default:
759 // If you get here, it may be that someone has added a new intrinsic
760 // to the list in vmIntrinsics.hpp without implementing it here.
761 #ifndef PRODUCT
762 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
763 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
764 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
765 }
766 #endif
767 Node* slow_ctl = control();
768 set_control(top()); // No fast path intrinsic
769 return slow_ctl;
770 }
771 }
772
773 //------------------------------set_result-------------------------------
774 // Helper function for finishing intrinsics.
775 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
776 record_for_igvn(region);
777 set_control(_gvn.transform(region));
778 set_result( _gvn.transform(value));
779 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
780 }
781
782 //------------------------------generate_guard---------------------------
783 // Helper function for generating guarded fast-slow graph structures.
784 // The given 'test', if true, guards a slow path. If the test fails
785 // then a fast path can be taken. (We generally hope it fails.)
786 // In all cases, GraphKit::control() is updated to the fast path.
787 // The returned value represents the control for the slow path.
788 // The return value is never 'top'; it is either a valid control
789 // or NULL if it is obvious that the slow path can never be taken.
790 // Also, if region and the slow control are not NULL, the slow edge
791 // is appended to the region.
792 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
793 if (stopped()) {
794 // Already short circuited.
795 return NULL;
796 }
797
798 // Build an if node and its projections.
799 // If test is true we take the slow path, which we assume is uncommon.
800 if (_gvn.type(test) == TypeInt::ZERO) {
801 // The slow branch is never taken. No need to build this guard.
802 return NULL;
803 }
804
805 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
806
807 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
808 if (if_slow == top()) {
809 // The slow branch is never taken. No need to build this guard.
810 return NULL;
811 }
812
813 if (region != NULL)
814 region->add_req(if_slow);
815
816 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
817 set_control(if_fast);
818
819 return if_slow;
820 }
821
822 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
823 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
824 }
825 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
826 return generate_guard(test, region, PROB_FAIR);
827 }
828
829 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
830 Node* *pos_index) {
831 if (stopped())
832 return NULL; // already stopped
833 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
834 return NULL; // index is already adequately typed
835 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
836 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
837 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
838 if (is_neg != NULL && pos_index != NULL) {
839 // Emulate effect of Parse::adjust_map_after_if.
840 Node* ccast = new CastIINode(index, TypeInt::POS);
841 ccast->set_req(0, control());
842 (*pos_index) = _gvn.transform(ccast);
843 }
844 return is_neg;
845 }
846
847 // Make sure that 'position' is a valid limit index, in [0..length].
848 // There are two equivalent plans for checking this:
849 // A. (offset + copyLength) unsigned<= arrayLength
850 // B. offset <= (arrayLength - copyLength)
851 // We require that all of the values above, except for the sum and
852 // difference, are already known to be non-negative.
853 // Plan A is robust in the face of overflow, if offset and copyLength
854 // are both hugely positive.
855 //
856 // Plan B is less direct and intuitive, but it does not overflow at
857 // all, since the difference of two non-negatives is always
858 // representable. Whenever Java methods must perform the equivalent
859 // check they generally use Plan B instead of Plan A.
860 // For the moment we use Plan A.
861 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
862 Node* subseq_length,
863 Node* array_length,
864 RegionNode* region) {
865 if (stopped())
866 return NULL; // already stopped
867 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
868 if (zero_offset && subseq_length->eqv_uncast(array_length))
869 return NULL; // common case of whole-array copy
870 Node* last = subseq_length;
871 if (!zero_offset) // last += offset
872 last = _gvn.transform(new AddINode(last, offset));
873 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
874 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
875 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
876 return is_over;
877 }
878
879 // Emit range checks for the given String.value byte array
880 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
881 if (stopped()) {
882 return; // already stopped
883 }
884 RegionNode* bailout = new RegionNode(1);
885 record_for_igvn(bailout);
886 if (char_count) {
887 // Convert char count to byte count
888 count = _gvn.transform(new LShiftINode(count, intcon(1)));
889 }
890
891 // Offset and count must not be negative
892 generate_negative_guard(offset, bailout);
893 generate_negative_guard(count, bailout);
894 // Offset + count must not exceed length of array
895 generate_limit_guard(offset, count, load_array_length(array), bailout);
896
897 if (bailout->req() > 1) {
898 PreserveJVMState pjvms(this);
899 set_control(_gvn.transform(bailout));
900 uncommon_trap(Deoptimization::Reason_intrinsic,
901 Deoptimization::Action_maybe_recompile);
902 }
903 }
904
905 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
906 bool is_immutable) {
907 ciKlass* thread_klass = env()->Thread_klass();
908 const Type* thread_type
909 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
910
911 Node* thread = _gvn.transform(new ThreadLocalNode());
912 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
913 tls_output = thread;
914
915 Node* thread_obj_handle
916 = (is_immutable
917 ? LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
918 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
919 : make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
920 thread_obj_handle = _gvn.transform(thread_obj_handle);
921
922 DecoratorSet decorators = IN_NATIVE;
923 if (is_immutable) {
924 decorators |= C2_IMMUTABLE_MEMORY;
925 }
926 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
927 }
928
929 //--------------------------generate_current_thread--------------------
930 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
931 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
932 /*is_immutable*/false);
933 }
934
935 //--------------------------generate_virtual_thread--------------------
936 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
937 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
938 !C->method()->changes_current_thread());
939 }
940
941 //------------------------------make_string_method_node------------------------
942 // Helper method for String intrinsic functions. This version is called with
943 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
944 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
945 // containing the lengths of str1 and str2.
946 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
947 Node* result = NULL;
948 switch (opcode) {
949 case Op_StrIndexOf:
950 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
951 str1_start, cnt1, str2_start, cnt2, ae);
952 break;
953 case Op_StrComp:
954 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
955 str1_start, cnt1, str2_start, cnt2, ae);
956 break;
957 case Op_StrEquals:
958 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
959 // Use the constant length if there is one because optimized match rule may exist.
960 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
961 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
962 break;
963 default:
964 ShouldNotReachHere();
965 return NULL;
966 }
967
968 // All these intrinsics have checks.
969 C->set_has_split_ifs(true); // Has chance for split-if optimization
970 clear_upper_avx();
971
972 return _gvn.transform(result);
973 }
974
975 //------------------------------inline_string_compareTo------------------------
976 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
977 Node* arg1 = argument(0);
978 Node* arg2 = argument(1);
979
980 arg1 = must_be_not_null(arg1, true);
981 arg2 = must_be_not_null(arg2, true);
982
983 // Get start addr and length of first argument
984 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
985 Node* arg1_cnt = load_array_length(arg1);
986
987 // Get start addr and length of second argument
988 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
989 Node* arg2_cnt = load_array_length(arg2);
990
991 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
992 set_result(result);
993 return true;
994 }
995
996 //------------------------------inline_string_equals------------------------
997 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
998 Node* arg1 = argument(0);
999 Node* arg2 = argument(1);
1000
1001 // paths (plus control) merge
1002 RegionNode* region = new RegionNode(3);
1003 Node* phi = new PhiNode(region, TypeInt::BOOL);
1004
1005 if (!stopped()) {
1006
1007 arg1 = must_be_not_null(arg1, true);
1008 arg2 = must_be_not_null(arg2, true);
1009
1010 // Get start addr and length of first argument
1011 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1012 Node* arg1_cnt = load_array_length(arg1);
1013
1014 // Get start addr and length of second argument
1015 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1016 Node* arg2_cnt = load_array_length(arg2);
1017
1018 // Check for arg1_cnt != arg2_cnt
1019 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1020 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1021 Node* if_ne = generate_slow_guard(bol, NULL);
1022 if (if_ne != NULL) {
1023 phi->init_req(2, intcon(0));
1024 region->init_req(2, if_ne);
1025 }
1026
1027 // Check for count == 0 is done by assembler code for StrEquals.
1028
1029 if (!stopped()) {
1030 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1031 phi->init_req(1, equals);
1032 region->init_req(1, control());
1033 }
1034 }
1035
1036 // post merge
1037 set_control(_gvn.transform(region));
1038 record_for_igvn(region);
1039
1040 set_result(_gvn.transform(phi));
1041 return true;
1042 }
1043
1044 //------------------------------inline_array_equals----------------------------
1045 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1046 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1047 Node* arg1 = argument(0);
1048 Node* arg2 = argument(1);
1049
1050 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1051 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1052 clear_upper_avx();
1053
1054 return true;
1055 }
1056
1057 //------------------------------inline_countPositives------------------------------
1058 bool LibraryCallKit::inline_countPositives() {
1059 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1060 return false;
1061 }
1062
1063 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1064 // no receiver since it is static method
1065 Node* ba = argument(0);
1066 Node* offset = argument(1);
1067 Node* len = argument(2);
1068
1069 ba = must_be_not_null(ba, true);
1070
1071 // Range checks
1072 generate_string_range_check(ba, offset, len, false);
1073 if (stopped()) {
1074 return true;
1075 }
1076 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1077 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1078 set_result(_gvn.transform(result));
1079 return true;
1080 }
1081
1082 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1083 Node* index = argument(0);
1084 Node* length = bt == T_INT ? argument(1) : argument(2);
1085 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1086 return false;
1087 }
1088
1089 // check that length is positive
1090 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1091 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1092
1093 {
1094 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1095 uncommon_trap(Deoptimization::Reason_intrinsic,
1096 Deoptimization::Action_make_not_entrant);
1097 }
1098
1099 if (stopped()) {
1100 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1101 return true;
1102 }
1103
1104 // length is now known positive, add a cast node to make this explicit
1105 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1106 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1107 casted_length = _gvn.transform(casted_length);
1108 replace_in_map(length, casted_length);
1109 length = casted_length;
1110
1111 // Use an unsigned comparison for the range check itself
1112 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1113 BoolTest::mask btest = BoolTest::lt;
1114 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1115 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1116 _gvn.set_type(rc, rc->Value(&_gvn));
1117 if (!rc_bool->is_Con()) {
1118 record_for_igvn(rc);
1119 }
1120 set_control(_gvn.transform(new IfTrueNode(rc)));
1121 {
1122 PreserveJVMState pjvms(this);
1123 set_control(_gvn.transform(new IfFalseNode(rc)));
1124 uncommon_trap(Deoptimization::Reason_range_check,
1125 Deoptimization::Action_make_not_entrant);
1126 }
1127
1128 if (stopped()) {
1129 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1130 return true;
1131 }
1132
1133 // index is now known to be >= 0 and < length, cast it
1134 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1135 result = _gvn.transform(result);
1136 set_result(result);
1137 replace_in_map(index, result);
1138 return true;
1139 }
1140
1141 //------------------------------inline_string_indexOf------------------------
1142 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1143 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1144 return false;
1145 }
1146 Node* src = argument(0);
1147 Node* tgt = argument(1);
1148
1149 // Make the merge point
1150 RegionNode* result_rgn = new RegionNode(4);
1151 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1152
1153 src = must_be_not_null(src, true);
1154 tgt = must_be_not_null(tgt, true);
1155
1156 // Get start addr and length of source string
1157 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1158 Node* src_count = load_array_length(src);
1159
1160 // Get start addr and length of substring
1161 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1162 Node* tgt_count = load_array_length(tgt);
1163
1164 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1165 // Divide src size by 2 if String is UTF16 encoded
1166 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1167 }
1168 if (ae == StrIntrinsicNode::UU) {
1169 // Divide substring size by 2 if String is UTF16 encoded
1170 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1171 }
1172
1173 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1174 if (result != NULL) {
1175 result_phi->init_req(3, result);
1176 result_rgn->init_req(3, control());
1177 }
1178 set_control(_gvn.transform(result_rgn));
1179 record_for_igvn(result_rgn);
1180 set_result(_gvn.transform(result_phi));
1181
1182 return true;
1183 }
1184
1185 //-----------------------------inline_string_indexOf-----------------------
1186 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1187 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1188 return false;
1189 }
1190 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1191 return false;
1192 }
1193 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1194 Node* src = argument(0); // byte[]
1195 Node* src_count = argument(1); // char count
1196 Node* tgt = argument(2); // byte[]
1197 Node* tgt_count = argument(3); // char count
1198 Node* from_index = argument(4); // char index
1199
1200 src = must_be_not_null(src, true);
1201 tgt = must_be_not_null(tgt, true);
1202
1203 // Multiply byte array index by 2 if String is UTF16 encoded
1204 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1205 src_count = _gvn.transform(new SubINode(src_count, from_index));
1206 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1207 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1208
1209 // Range checks
1210 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1211 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1212 if (stopped()) {
1213 return true;
1214 }
1215
1216 RegionNode* region = new RegionNode(5);
1217 Node* phi = new PhiNode(region, TypeInt::INT);
1218
1219 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1220 if (result != NULL) {
1221 // The result is index relative to from_index if substring was found, -1 otherwise.
1222 // Generate code which will fold into cmove.
1223 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1224 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1225
1226 Node* if_lt = generate_slow_guard(bol, NULL);
1227 if (if_lt != NULL) {
1228 // result == -1
1229 phi->init_req(3, result);
1230 region->init_req(3, if_lt);
1231 }
1232 if (!stopped()) {
1233 result = _gvn.transform(new AddINode(result, from_index));
1234 phi->init_req(4, result);
1235 region->init_req(4, control());
1236 }
1237 }
1238
1239 set_control(_gvn.transform(region));
1240 record_for_igvn(region);
1241 set_result(_gvn.transform(phi));
1242 clear_upper_avx();
1243
1244 return true;
1245 }
1246
1247 // Create StrIndexOfNode with fast path checks
1248 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1249 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1250 // Check for substr count > string count
1251 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1252 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1253 Node* if_gt = generate_slow_guard(bol, NULL);
1254 if (if_gt != NULL) {
1255 phi->init_req(1, intcon(-1));
1256 region->init_req(1, if_gt);
1257 }
1258 if (!stopped()) {
1259 // Check for substr count == 0
1260 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1261 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1262 Node* if_zero = generate_slow_guard(bol, NULL);
1263 if (if_zero != NULL) {
1264 phi->init_req(2, intcon(0));
1265 region->init_req(2, if_zero);
1266 }
1267 }
1268 if (!stopped()) {
1269 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1270 }
1271 return NULL;
1272 }
1273
1274 //-----------------------------inline_string_indexOfChar-----------------------
1275 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1276 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1277 return false;
1278 }
1279 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1280 return false;
1281 }
1282 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1283 Node* src = argument(0); // byte[]
1284 Node* tgt = argument(1); // tgt is int ch
1285 Node* from_index = argument(2);
1286 Node* max = argument(3);
1287
1288 src = must_be_not_null(src, true);
1289
1290 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1291 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1292 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1293
1294 // Range checks
1295 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1296 if (stopped()) {
1297 return true;
1298 }
1299
1300 RegionNode* region = new RegionNode(3);
1301 Node* phi = new PhiNode(region, TypeInt::INT);
1302
1303 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1304 C->set_has_split_ifs(true); // Has chance for split-if optimization
1305 _gvn.transform(result);
1306
1307 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1308 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1309
1310 Node* if_lt = generate_slow_guard(bol, NULL);
1311 if (if_lt != NULL) {
1312 // result == -1
1313 phi->init_req(2, result);
1314 region->init_req(2, if_lt);
1315 }
1316 if (!stopped()) {
1317 result = _gvn.transform(new AddINode(result, from_index));
1318 phi->init_req(1, result);
1319 region->init_req(1, control());
1320 }
1321 set_control(_gvn.transform(region));
1322 record_for_igvn(region);
1323 set_result(_gvn.transform(phi));
1324
1325 return true;
1326 }
1327 //---------------------------inline_string_copy---------------------
1328 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1329 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1330 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1331 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1332 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1333 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1334 bool LibraryCallKit::inline_string_copy(bool compress) {
1335 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1336 return false;
1337 }
1338 int nargs = 5; // 2 oops, 3 ints
1339 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1340
1341 Node* src = argument(0);
1342 Node* src_offset = argument(1);
1343 Node* dst = argument(2);
1344 Node* dst_offset = argument(3);
1345 Node* length = argument(4);
1346
1347 // Check for allocation before we add nodes that would confuse
1348 // tightly_coupled_allocation()
1349 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1350
1351 // Figure out the size and type of the elements we will be copying.
1352 const Type* src_type = src->Value(&_gvn);
1353 const Type* dst_type = dst->Value(&_gvn);
1354 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
1355 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
1356 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1357 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1358 "Unsupported array types for inline_string_copy");
1359
1360 src = must_be_not_null(src, true);
1361 dst = must_be_not_null(dst, true);
1362
1363 // Convert char[] offsets to byte[] offsets
1364 bool convert_src = (compress && src_elem == T_BYTE);
1365 bool convert_dst = (!compress && dst_elem == T_BYTE);
1366 if (convert_src) {
1367 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1368 } else if (convert_dst) {
1369 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1370 }
1371
1372 // Range checks
1373 generate_string_range_check(src, src_offset, length, convert_src);
1374 generate_string_range_check(dst, dst_offset, length, convert_dst);
1375 if (stopped()) {
1376 return true;
1377 }
1378
1379 Node* src_start = array_element_address(src, src_offset, src_elem);
1380 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1381 // 'src_start' points to src array + scaled offset
1382 // 'dst_start' points to dst array + scaled offset
1383 Node* count = NULL;
1384 if (compress) {
1385 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1386 } else {
1387 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1388 }
1389
1390 if (alloc != NULL) {
1391 if (alloc->maybe_set_complete(&_gvn)) {
1392 // "You break it, you buy it."
1393 InitializeNode* init = alloc->initialization();
1394 assert(init->is_complete(), "we just did this");
1395 init->set_complete_with_arraycopy();
1396 assert(dst->is_CheckCastPP(), "sanity");
1397 assert(dst->in(0)->in(0) == init, "dest pinned");
1398 }
1399 // Do not let stores that initialize this object be reordered with
1400 // a subsequent store that would make this object accessible by
1401 // other threads.
1402 // Record what AllocateNode this StoreStore protects so that
1403 // escape analysis can go from the MemBarStoreStoreNode to the
1404 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1405 // based on the escape status of the AllocateNode.
1406 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1407 }
1408 if (compress) {
1409 set_result(_gvn.transform(count));
1410 }
1411 clear_upper_avx();
1412
1413 return true;
1414 }
1415
1416 #ifdef _LP64
1417 #define XTOP ,top() /*additional argument*/
1418 #else //_LP64
1419 #define XTOP /*no additional argument*/
1420 #endif //_LP64
1421
1422 //------------------------inline_string_toBytesU--------------------------
1423 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1424 bool LibraryCallKit::inline_string_toBytesU() {
1425 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1426 return false;
1427 }
1428 // Get the arguments.
1429 Node* value = argument(0);
1430 Node* offset = argument(1);
1431 Node* length = argument(2);
1432
1433 Node* newcopy = NULL;
1434
1435 // Set the original stack and the reexecute bit for the interpreter to reexecute
1436 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1437 { PreserveReexecuteState preexecs(this);
1438 jvms()->set_should_reexecute(true);
1439
1440 // Check if a null path was taken unconditionally.
1441 value = null_check(value);
1442
1443 RegionNode* bailout = new RegionNode(1);
1444 record_for_igvn(bailout);
1445
1446 // Range checks
1447 generate_negative_guard(offset, bailout);
1448 generate_negative_guard(length, bailout);
1449 generate_limit_guard(offset, length, load_array_length(value), bailout);
1450 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1451 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1452
1453 if (bailout->req() > 1) {
1454 PreserveJVMState pjvms(this);
1455 set_control(_gvn.transform(bailout));
1456 uncommon_trap(Deoptimization::Reason_intrinsic,
1457 Deoptimization::Action_maybe_recompile);
1458 }
1459 if (stopped()) {
1460 return true;
1461 }
1462
1463 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1464 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1465 newcopy = new_array(klass_node, size, 0); // no arguments to push
1466 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1467 guarantee(alloc != NULL, "created above");
1468
1469 // Calculate starting addresses.
1470 Node* src_start = array_element_address(value, offset, T_CHAR);
1471 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1472
1473 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1474 const TypeInt* toffset = gvn().type(offset)->is_int();
1475 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1476
1477 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1478 const char* copyfunc_name = "arraycopy";
1479 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1480 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1481 OptoRuntime::fast_arraycopy_Type(),
1482 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1483 src_start, dst_start, ConvI2X(length) XTOP);
1484 // Do not let reads from the cloned object float above the arraycopy.
1485 if (alloc->maybe_set_complete(&_gvn)) {
1486 // "You break it, you buy it."
1487 InitializeNode* init = alloc->initialization();
1488 assert(init->is_complete(), "we just did this");
1489 init->set_complete_with_arraycopy();
1490 assert(newcopy->is_CheckCastPP(), "sanity");
1491 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1492 }
1493 // Do not let stores that initialize this object be reordered with
1494 // a subsequent store that would make this object accessible by
1495 // other threads.
1496 // Record what AllocateNode this StoreStore protects so that
1497 // escape analysis can go from the MemBarStoreStoreNode to the
1498 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1499 // based on the escape status of the AllocateNode.
1500 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1501 } // original reexecute is set back here
1502
1503 C->set_has_split_ifs(true); // Has chance for split-if optimization
1504 if (!stopped()) {
1505 set_result(newcopy);
1506 }
1507 clear_upper_avx();
1508
1509 return true;
1510 }
1511
1512 //------------------------inline_string_getCharsU--------------------------
1513 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1514 bool LibraryCallKit::inline_string_getCharsU() {
1515 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1516 return false;
1517 }
1518
1519 // Get the arguments.
1520 Node* src = argument(0);
1521 Node* src_begin = argument(1);
1522 Node* src_end = argument(2); // exclusive offset (i < src_end)
1523 Node* dst = argument(3);
1524 Node* dst_begin = argument(4);
1525
1526 // Check for allocation before we add nodes that would confuse
1527 // tightly_coupled_allocation()
1528 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1529
1530 // Check if a null path was taken unconditionally.
1531 src = null_check(src);
1532 dst = null_check(dst);
1533 if (stopped()) {
1534 return true;
1535 }
1536
1537 // Get length and convert char[] offset to byte[] offset
1538 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1539 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1540
1541 // Range checks
1542 generate_string_range_check(src, src_begin, length, true);
1543 generate_string_range_check(dst, dst_begin, length, false);
1544 if (stopped()) {
1545 return true;
1546 }
1547
1548 if (!stopped()) {
1549 // Calculate starting addresses.
1550 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1551 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1552
1553 // Check if array addresses are aligned to HeapWordSize
1554 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1555 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1556 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1557 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1558
1559 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1560 const char* copyfunc_name = "arraycopy";
1561 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1562 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1563 OptoRuntime::fast_arraycopy_Type(),
1564 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1565 src_start, dst_start, ConvI2X(length) XTOP);
1566 // Do not let reads from the cloned object float above the arraycopy.
1567 if (alloc != NULL) {
1568 if (alloc->maybe_set_complete(&_gvn)) {
1569 // "You break it, you buy it."
1570 InitializeNode* init = alloc->initialization();
1571 assert(init->is_complete(), "we just did this");
1572 init->set_complete_with_arraycopy();
1573 assert(dst->is_CheckCastPP(), "sanity");
1574 assert(dst->in(0)->in(0) == init, "dest pinned");
1575 }
1576 // Do not let stores that initialize this object be reordered with
1577 // a subsequent store that would make this object accessible by
1578 // other threads.
1579 // Record what AllocateNode this StoreStore protects so that
1580 // escape analysis can go from the MemBarStoreStoreNode to the
1581 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1582 // based on the escape status of the AllocateNode.
1583 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1584 } else {
1585 insert_mem_bar(Op_MemBarCPUOrder);
1586 }
1587 }
1588
1589 C->set_has_split_ifs(true); // Has chance for split-if optimization
1590 return true;
1591 }
1592
1593 //----------------------inline_string_char_access----------------------------
1594 // Store/Load char to/from byte[] array.
1595 // static void StringUTF16.putChar(byte[] val, int index, int c)
1596 // static char StringUTF16.getChar(byte[] val, int index)
1597 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1598 Node* value = argument(0);
1599 Node* index = argument(1);
1600 Node* ch = is_store ? argument(2) : NULL;
1601
1602 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1603 // correctly requires matched array shapes.
1604 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1605 "sanity: byte[] and char[] bases agree");
1606 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1607 "sanity: byte[] and char[] scales agree");
1608
1609 // Bail when getChar over constants is requested: constant folding would
1610 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1611 // Java method would constant fold nicely instead.
1612 if (!is_store && value->is_Con() && index->is_Con()) {
1613 return false;
1614 }
1615
1616 value = must_be_not_null(value, true);
1617
1618 Node* adr = array_element_address(value, index, T_CHAR);
1619 if (adr->is_top()) {
1620 return false;
1621 }
1622 if (is_store) {
1623 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1624 } else {
1625 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);
1626 set_result(ch);
1627 }
1628 return true;
1629 }
1630
1631 //--------------------------round_double_node--------------------------------
1632 // Round a double node if necessary.
1633 Node* LibraryCallKit::round_double_node(Node* n) {
1634 if (Matcher::strict_fp_requires_explicit_rounding) {
1635 #ifdef IA32
1636 if (UseSSE < 2) {
1637 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1638 }
1639 #else
1640 Unimplemented();
1641 #endif // IA32
1642 }
1643 return n;
1644 }
1645
1646 //------------------------------inline_math-----------------------------------
1647 // public static double Math.abs(double)
1648 // public static double Math.sqrt(double)
1649 // public static double Math.log(double)
1650 // public static double Math.log10(double)
1651 // public static double Math.round(double)
1652 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1653 Node* arg = round_double_node(argument(0));
1654 Node* n = NULL;
1655 switch (id) {
1656 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1657 case vmIntrinsics::_dsqrt:
1658 case vmIntrinsics::_dsqrt_strict:
1659 n = new SqrtDNode(C, control(), arg); break;
1660 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1661 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1662 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1663 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1664 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1665 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1666 default: fatal_unexpected_iid(id); break;
1667 }
1668 set_result(_gvn.transform(n));
1669 return true;
1670 }
1671
1672 //------------------------------inline_math-----------------------------------
1673 // public static float Math.abs(float)
1674 // public static int Math.abs(int)
1675 // public static long Math.abs(long)
1676 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1677 Node* arg = argument(0);
1678 Node* n = NULL;
1679 switch (id) {
1680 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1681 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1682 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1683 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1684 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1685 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1686 default: fatal_unexpected_iid(id); break;
1687 }
1688 set_result(_gvn.transform(n));
1689 return true;
1690 }
1691
1692 //------------------------------runtime_math-----------------------------
1693 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1694 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1695 "must be (DD)D or (D)D type");
1696
1697 // Inputs
1698 Node* a = round_double_node(argument(0));
1699 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1700
1701 const TypePtr* no_memory_effects = NULL;
1702 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1703 no_memory_effects,
1704 a, top(), b, b ? top() : NULL);
1705 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1706 #ifdef ASSERT
1707 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1708 assert(value_top == top(), "second value must be top");
1709 #endif
1710
1711 set_result(value);
1712 return true;
1713 }
1714
1715 //------------------------------inline_math_pow-----------------------------
1716 bool LibraryCallKit::inline_math_pow() {
1717 Node* exp = round_double_node(argument(2));
1718 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1719 if (d != NULL) {
1720 if (d->getd() == 2.0) {
1721 // Special case: pow(x, 2.0) => x * x
1722 Node* base = round_double_node(argument(0));
1723 set_result(_gvn.transform(new MulDNode(base, base)));
1724 return true;
1725 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1726 // Special case: pow(x, 0.5) => sqrt(x)
1727 Node* base = round_double_node(argument(0));
1728 Node* zero = _gvn.zerocon(T_DOUBLE);
1729
1730 RegionNode* region = new RegionNode(3);
1731 Node* phi = new PhiNode(region, Type::DOUBLE);
1732
1733 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1734 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1735 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1736 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1737 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1738
1739 Node* if_pow = generate_slow_guard(test, NULL);
1740 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1741 phi->init_req(1, value_sqrt);
1742 region->init_req(1, control());
1743
1744 if (if_pow != NULL) {
1745 set_control(if_pow);
1746 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1747 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1748 const TypePtr* no_memory_effects = NULL;
1749 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1750 no_memory_effects, base, top(), exp, top());
1751 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1752 #ifdef ASSERT
1753 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1754 assert(value_top == top(), "second value must be top");
1755 #endif
1756 phi->init_req(2, value_pow);
1757 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1758 }
1759
1760 C->set_has_split_ifs(true); // Has chance for split-if optimization
1761 set_control(_gvn.transform(region));
1762 record_for_igvn(region);
1763 set_result(_gvn.transform(phi));
1764
1765 return true;
1766 }
1767 }
1768
1769 return StubRoutines::dpow() != NULL ?
1770 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1771 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1772 }
1773
1774 //------------------------------inline_math_native-----------------------------
1775 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1776 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1777 switch (id) {
1778 // These intrinsics are not properly supported on all hardware
1779 case vmIntrinsics::_dsin:
1780 return StubRoutines::dsin() != NULL ?
1781 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1782 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1783 case vmIntrinsics::_dcos:
1784 return StubRoutines::dcos() != NULL ?
1785 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1786 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1787 case vmIntrinsics::_dtan:
1788 return StubRoutines::dtan() != NULL ?
1789 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1790 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1791 case vmIntrinsics::_dlog:
1792 return StubRoutines::dlog() != NULL ?
1793 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1794 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1795 case vmIntrinsics::_dlog10:
1796 return StubRoutines::dlog10() != NULL ?
1797 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1798 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1799
1800 // These intrinsics are supported on all hardware
1801 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1802 case vmIntrinsics::_ceil:
1803 case vmIntrinsics::_floor:
1804 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1805
1806 case vmIntrinsics::_dsqrt:
1807 case vmIntrinsics::_dsqrt_strict:
1808 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1809 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1810 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1811 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1812 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1813
1814 case vmIntrinsics::_dexp:
1815 return StubRoutines::dexp() != NULL ?
1816 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1817 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1818 #undef FN_PTR
1819
1820 case vmIntrinsics::_dpow: return inline_math_pow();
1821 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1822 case vmIntrinsics::_fcopySign: return inline_math(id);
1823 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1824 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1825 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1826
1827 // These intrinsics are not yet correctly implemented
1828 case vmIntrinsics::_datan2:
1829 return false;
1830
1831 default:
1832 fatal_unexpected_iid(id);
1833 return false;
1834 }
1835 }
1836
1837 static bool is_simple_name(Node* n) {
1838 return (n->req() == 1 // constant
1839 || (n->is_Type() && n->as_Type()->type()->singleton())
1840 || n->is_Proj() // parameter or return value
1841 || n->is_Phi() // local of some sort
1842 );
1843 }
1844
1845 //----------------------------inline_notify-----------------------------------*
1846 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1847 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1848 address func;
1849 if (id == vmIntrinsics::_notify) {
1850 func = OptoRuntime::monitor_notify_Java();
1851 } else {
1852 func = OptoRuntime::monitor_notifyAll_Java();
1853 }
1854 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1855 make_slow_call_ex(call, env()->Throwable_klass(), false);
1856 return true;
1857 }
1858
1859
1860 //----------------------------inline_min_max-----------------------------------
1861 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1862 set_result(generate_min_max(id, argument(0), argument(1)));
1863 return true;
1864 }
1865
1866 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1867 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1868 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1869 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1870 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1871
1872 {
1873 PreserveJVMState pjvms(this);
1874 PreserveReexecuteState preexecs(this);
1875 jvms()->set_should_reexecute(true);
1876
1877 set_control(slow_path);
1878 set_i_o(i_o());
1879
1880 uncommon_trap(Deoptimization::Reason_intrinsic,
1881 Deoptimization::Action_none);
1882 }
1883
1884 set_control(fast_path);
1885 set_result(math);
1886 }
1887
1888 template <typename OverflowOp>
1889 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1890 typedef typename OverflowOp::MathOp MathOp;
1891
1892 MathOp* mathOp = new MathOp(arg1, arg2);
1893 Node* operation = _gvn.transform( mathOp );
1894 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1895 inline_math_mathExact(operation, ofcheck);
1896 return true;
1897 }
1898
1899 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1900 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1901 }
1902
1903 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1904 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1905 }
1906
1907 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1908 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1909 }
1910
1911 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1912 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1913 }
1914
1915 bool LibraryCallKit::inline_math_negateExactI() {
1916 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1917 }
1918
1919 bool LibraryCallKit::inline_math_negateExactL() {
1920 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1921 }
1922
1923 bool LibraryCallKit::inline_math_multiplyExactI() {
1924 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1925 }
1926
1927 bool LibraryCallKit::inline_math_multiplyExactL() {
1928 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1929 }
1930
1931 bool LibraryCallKit::inline_math_multiplyHigh() {
1932 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1933 return true;
1934 }
1935
1936 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1937 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1938 return true;
1939 }
1940
1941 Node*
1942 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1943 // These are the candidate return value:
1944 Node* xvalue = x0;
1945 Node* yvalue = y0;
1946
1947 if (xvalue == yvalue) {
1948 return xvalue;
1949 }
1950
1951 bool want_max = (id == vmIntrinsics::_max || id == vmIntrinsics::_max_strict);
1952
1953 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1954 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1955 if (txvalue == NULL || tyvalue == NULL) return top();
1956 // This is not really necessary, but it is consistent with a
1957 // hypothetical MaxINode::Value method:
1958 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1959
1960 // %%% This folding logic should (ideally) be in a different place.
1961 // Some should be inside IfNode, and there to be a more reliable
1962 // transformation of ?: style patterns into cmoves. We also want
1963 // more powerful optimizations around cmove and min/max.
1964
1965 // Try to find a dominating comparison of these guys.
1966 // It can simplify the index computation for Arrays.copyOf
1967 // and similar uses of System.arraycopy.
1968 // First, compute the normalized version of CmpI(x, y).
1969 int cmp_op = Op_CmpI;
1970 Node* xkey = xvalue;
1971 Node* ykey = yvalue;
1972 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1973 if (ideal_cmpxy->is_Cmp()) {
1974 // E.g., if we have CmpI(length - offset, count),
1975 // it might idealize to CmpI(length, count + offset)
1976 cmp_op = ideal_cmpxy->Opcode();
1977 xkey = ideal_cmpxy->in(1);
1978 ykey = ideal_cmpxy->in(2);
1979 }
1980
1981 // Start by locating any relevant comparisons.
1982 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1983 Node* cmpxy = NULL;
1984 Node* cmpyx = NULL;
1985 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1986 Node* cmp = start_from->fast_out(k);
1987 if (cmp->outcnt() > 0 && // must have prior uses
1988 cmp->in(0) == NULL && // must be context-independent
1989 cmp->Opcode() == cmp_op) { // right kind of compare
1990 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1991 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1992 }
1993 }
1994
1995 const int NCMPS = 2;
1996 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1997 int cmpn;
1998 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1999 if (cmps[cmpn] != NULL) break; // find a result
2000 }
2001 if (cmpn < NCMPS) {
2002 // Look for a dominating test that tells us the min and max.
2003 int depth = 0; // Limit search depth for speed
2004 Node* dom = control();
2005 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2006 if (++depth >= 100) break;
2007 Node* ifproj = dom;
2008 if (!ifproj->is_Proj()) continue;
2009 Node* iff = ifproj->in(0);
2010 if (!iff->is_If()) continue;
2011 Node* bol = iff->in(1);
2012 if (!bol->is_Bool()) continue;
2013 Node* cmp = bol->in(1);
2014 if (cmp == NULL) continue;
2015 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2016 if (cmps[cmpn] == cmp) break;
2017 if (cmpn == NCMPS) continue;
2018 BoolTest::mask btest = bol->as_Bool()->_test._test;
2019 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2020 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2021 // At this point, we know that 'x btest y' is true.
2022 switch (btest) {
2023 case BoolTest::eq:
2024 // They are proven equal, so we can collapse the min/max.
2025 // Either value is the answer. Choose the simpler.
2026 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2027 return yvalue;
2028 return xvalue;
2029 case BoolTest::lt: // x < y
2030 case BoolTest::le: // x <= y
2031 return (want_max ? yvalue : xvalue);
2032 case BoolTest::gt: // x > y
2033 case BoolTest::ge: // x >= y
2034 return (want_max ? xvalue : yvalue);
2035 default:
2036 break;
2037 }
2038 }
2039 }
2040
2041 // We failed to find a dominating test.
2042 // Let's pick a test that might GVN with prior tests.
2043 Node* best_bol = NULL;
2044 BoolTest::mask best_btest = BoolTest::illegal;
2045 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2046 Node* cmp = cmps[cmpn];
2047 if (cmp == NULL) continue;
2048 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2049 Node* bol = cmp->fast_out(j);
2050 if (!bol->is_Bool()) continue;
2051 BoolTest::mask btest = bol->as_Bool()->_test._test;
2052 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2053 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2054 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2055 best_bol = bol->as_Bool();
2056 best_btest = btest;
2057 }
2058 }
2059 }
2060
2061 Node* answer_if_true = NULL;
2062 Node* answer_if_false = NULL;
2063 switch (best_btest) {
2064 default:
2065 if (cmpxy == NULL)
2066 cmpxy = ideal_cmpxy;
2067 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2068 // and fall through:
2069 case BoolTest::lt: // x < y
2070 case BoolTest::le: // x <= y
2071 answer_if_true = (want_max ? yvalue : xvalue);
2072 answer_if_false = (want_max ? xvalue : yvalue);
2073 break;
2074 case BoolTest::gt: // x > y
2075 case BoolTest::ge: // x >= y
2076 answer_if_true = (want_max ? xvalue : yvalue);
2077 answer_if_false = (want_max ? yvalue : xvalue);
2078 break;
2079 }
2080
2081 jint hi, lo;
2082 if (want_max) {
2083 // We can sharpen the minimum.
2084 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2085 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2086 } else {
2087 // We can sharpen the maximum.
2088 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2089 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2090 }
2091
2092 // Use a flow-free graph structure, to avoid creating excess control edges
2093 // which could hinder other optimizations.
2094 // Since Math.min/max is often used with arraycopy, we want
2095 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2096 Node* cmov = CMoveNode::make(NULL, best_bol,
2097 answer_if_false, answer_if_true,
2098 TypeInt::make(lo, hi, widen));
2099
2100 return _gvn.transform(cmov);
2101
2102 /*
2103 // This is not as desirable as it may seem, since Min and Max
2104 // nodes do not have a full set of optimizations.
2105 // And they would interfere, anyway, with 'if' optimizations
2106 // and with CMoveI canonical forms.
2107 switch (id) {
2108 case vmIntrinsics::_min:
2109 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2110 case vmIntrinsics::_max:
2111 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2112 default:
2113 ShouldNotReachHere();
2114 }
2115 */
2116 }
2117
2118 inline int
2119 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2120 const TypePtr* base_type = TypePtr::NULL_PTR;
2121 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2122 if (base_type == NULL) {
2123 // Unknown type.
2124 return Type::AnyPtr;
2125 } else if (base_type == TypePtr::NULL_PTR) {
2126 // Since this is a NULL+long form, we have to switch to a rawptr.
2127 base = _gvn.transform(new CastX2PNode(offset));
2128 offset = MakeConX(0);
2129 return Type::RawPtr;
2130 } else if (base_type->base() == Type::RawPtr) {
2131 return Type::RawPtr;
2132 } else if (base_type->isa_oopptr()) {
2133 // Base is never null => always a heap address.
2134 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2135 return Type::OopPtr;
2136 }
2137 // Offset is small => always a heap address.
2138 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2139 if (offset_type != NULL &&
2140 base_type->offset() == 0 && // (should always be?)
2141 offset_type->_lo >= 0 &&
2142 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2143 return Type::OopPtr;
2144 } else if (type == T_OBJECT) {
2145 // off heap access to an oop doesn't make any sense. Has to be on
2146 // heap.
2147 return Type::OopPtr;
2148 }
2149 // Otherwise, it might either be oop+off or NULL+addr.
2150 return Type::AnyPtr;
2151 } else {
2152 // No information:
2153 return Type::AnyPtr;
2154 }
2155 }
2156
2157 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2158 Node* uncasted_base = base;
2159 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2160 if (kind == Type::RawPtr) {
2161 return basic_plus_adr(top(), uncasted_base, offset);
2162 } else if (kind == Type::AnyPtr) {
2163 assert(base == uncasted_base, "unexpected base change");
2164 if (can_cast) {
2165 if (!_gvn.type(base)->speculative_maybe_null() &&
2166 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2167 // According to profiling, this access is always on
2168 // heap. Casting the base to not null and thus avoiding membars
2169 // around the access should allow better optimizations
2170 Node* null_ctl = top();
2171 base = null_check_oop(base, &null_ctl, true, true, true);
2172 assert(null_ctl->is_top(), "no null control here");
2173 return basic_plus_adr(base, offset);
2174 } else if (_gvn.type(base)->speculative_always_null() &&
2175 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2176 // According to profiling, this access is always off
2177 // heap.
2178 base = null_assert(base);
2179 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2180 offset = MakeConX(0);
2181 return basic_plus_adr(top(), raw_base, offset);
2182 }
2183 }
2184 // We don't know if it's an on heap or off heap access. Fall back
2185 // to raw memory access.
2186 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2187 return basic_plus_adr(top(), raw, offset);
2188 } else {
2189 assert(base == uncasted_base, "unexpected base change");
2190 // We know it's an on heap access so base can't be null
2191 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2192 base = must_be_not_null(base, true);
2193 }
2194 return basic_plus_adr(base, offset);
2195 }
2196 }
2197
2198 //--------------------------inline_number_methods-----------------------------
2199 // inline int Integer.numberOfLeadingZeros(int)
2200 // inline int Long.numberOfLeadingZeros(long)
2201 //
2202 // inline int Integer.numberOfTrailingZeros(int)
2203 // inline int Long.numberOfTrailingZeros(long)
2204 //
2205 // inline int Integer.bitCount(int)
2206 // inline int Long.bitCount(long)
2207 //
2208 // inline char Character.reverseBytes(char)
2209 // inline short Short.reverseBytes(short)
2210 // inline int Integer.reverseBytes(int)
2211 // inline long Long.reverseBytes(long)
2212 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2213 Node* arg = argument(0);
2214 Node* n = NULL;
2215 switch (id) {
2216 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2217 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2218 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2219 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2220 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2221 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2222 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2223 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2224 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2225 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2226 default: fatal_unexpected_iid(id); break;
2227 }
2228 set_result(_gvn.transform(n));
2229 return true;
2230 }
2231
2232 //--------------------------inline_bitshuffle_methods-----------------------------
2233 // inline int Integer.compress(int, int)
2234 // inline int Integer.expand(int, int)
2235 // inline long Long.compress(long, long)
2236 // inline long Long.expand(long, long)
2237 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2238 Node* n = NULL;
2239 switch (id) {
2240 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2241 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2242 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2243 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2244 default: fatal_unexpected_iid(id); break;
2245 }
2246 set_result(_gvn.transform(n));
2247 return true;
2248 }
2249
2250 //--------------------------inline_unsigned_divmod_methods-----------------------------
2251 // inline int Integer.divideUnsigned(int, int)
2252 // inline int Integer.remainderUnsigned(int, int)
2253 // inline long Long.divideUnsigned(long, long)
2254 // inline long Long.remainderUnsigned(long, long)
2255 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2256 Node* n = NULL;
2257 switch (id) {
2258 case vmIntrinsics::_divideUnsigned_i: {
2259 zero_check_int(argument(1));
2260 // Compile-time detect of null-exception
2261 if (stopped()) {
2262 return true; // keep the graph constructed so far
2263 }
2264 n = new UDivINode(control(), argument(0), argument(1));
2265 break;
2266 }
2267 case vmIntrinsics::_divideUnsigned_l: {
2268 zero_check_long(argument(2));
2269 // Compile-time detect of null-exception
2270 if (stopped()) {
2271 return true; // keep the graph constructed so far
2272 }
2273 n = new UDivLNode(control(), argument(0), argument(2));
2274 break;
2275 }
2276 case vmIntrinsics::_remainderUnsigned_i: {
2277 zero_check_int(argument(1));
2278 // Compile-time detect of null-exception
2279 if (stopped()) {
2280 return true; // keep the graph constructed so far
2281 }
2282 n = new UModINode(control(), argument(0), argument(1));
2283 break;
2284 }
2285 case vmIntrinsics::_remainderUnsigned_l: {
2286 zero_check_long(argument(2));
2287 // Compile-time detect of null-exception
2288 if (stopped()) {
2289 return true; // keep the graph constructed so far
2290 }
2291 n = new UModLNode(control(), argument(0), argument(2));
2292 break;
2293 }
2294 default: fatal_unexpected_iid(id); break;
2295 }
2296 set_result(_gvn.transform(n));
2297 return true;
2298 }
2299
2300 //----------------------------inline_unsafe_access----------------------------
2301
2302 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2303 // Attempt to infer a sharper value type from the offset and base type.
2304 ciKlass* sharpened_klass = NULL;
2305
2306 // See if it is an instance field, with an object type.
2307 if (alias_type->field() != NULL) {
2308 if (alias_type->field()->type()->is_klass()) {
2309 sharpened_klass = alias_type->field()->type()->as_klass();
2310 }
2311 }
2312
2313 const TypeOopPtr* result = NULL;
2314 // See if it is a narrow oop array.
2315 if (adr_type->isa_aryptr()) {
2316 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2317 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2318 if (elem_type != NULL && elem_type->is_loaded()) {
2319 // Sharpen the value type.
2320 result = elem_type;
2321 }
2322 }
2323 }
2324
2325 // The sharpened class might be unloaded if there is no class loader
2326 // contraint in place.
2327 if (result == NULL && sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2328 // Sharpen the value type.
2329 result = TypeOopPtr::make_from_klass(sharpened_klass);
2330 }
2331 if (result != NULL) {
2332 #ifndef PRODUCT
2333 if (C->print_intrinsics() || C->print_inlining()) {
2334 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2335 tty->print(" sharpened value: "); result->dump(); tty->cr();
2336 }
2337 #endif
2338 }
2339 return result;
2340 }
2341
2342 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2343 switch (kind) {
2344 case Relaxed:
2345 return MO_UNORDERED;
2346 case Opaque:
2347 return MO_RELAXED;
2348 case Acquire:
2349 return MO_ACQUIRE;
2350 case Release:
2351 return MO_RELEASE;
2352 case Volatile:
2353 return MO_SEQ_CST;
2354 default:
2355 ShouldNotReachHere();
2356 return 0;
2357 }
2358 }
2359
2360 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2361 if (callee()->is_static()) return false; // caller must have the capability!
2362 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2363 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2364 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2365 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2366
2367 if (is_reference_type(type)) {
2368 decorators |= ON_UNKNOWN_OOP_REF;
2369 }
2370
2371 if (unaligned) {
2372 decorators |= C2_UNALIGNED;
2373 }
2374
2375 #ifndef PRODUCT
2376 {
2377 ResourceMark rm;
2378 // Check the signatures.
2379 ciSignature* sig = callee()->signature();
2380 #ifdef ASSERT
2381 if (!is_store) {
2382 // Object getReference(Object base, int/long offset), etc.
2383 BasicType rtype = sig->return_type()->basic_type();
2384 assert(rtype == type, "getter must return the expected value");
2385 assert(sig->count() == 2, "oop getter has 2 arguments");
2386 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2387 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2388 } else {
2389 // void putReference(Object base, int/long offset, Object x), etc.
2390 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2391 assert(sig->count() == 3, "oop putter has 3 arguments");
2392 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2393 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2394 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2395 assert(vtype == type, "putter must accept the expected value");
2396 }
2397 #endif // ASSERT
2398 }
2399 #endif //PRODUCT
2400
2401 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2402
2403 Node* receiver = argument(0); // type: oop
2404
2405 // Build address expression.
2406 Node* heap_base_oop = top();
2407
2408 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2409 Node* base = argument(1); // type: oop
2410 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2411 Node* offset = argument(2); // type: long
2412 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2413 // to be plain byte offsets, which are also the same as those accepted
2414 // by oopDesc::field_addr.
2415 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2416 "fieldOffset must be byte-scaled");
2417 // 32-bit machines ignore the high half!
2418 offset = ConvL2X(offset);
2419
2420 // Save state and restore on bailout
2421 uint old_sp = sp();
2422 SafePointNode* old_map = clone_map();
2423
2424 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2425
2426 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2427 if (type != T_OBJECT) {
2428 decorators |= IN_NATIVE; // off-heap primitive access
2429 } else {
2430 set_map(old_map);
2431 set_sp(old_sp);
2432 return false; // off-heap oop accesses are not supported
2433 }
2434 } else {
2435 heap_base_oop = base; // on-heap or mixed access
2436 }
2437
2438 // Can base be NULL? Otherwise, always on-heap access.
2439 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2440
2441 if (!can_access_non_heap) {
2442 decorators |= IN_HEAP;
2443 }
2444
2445 Node* val = is_store ? argument(4) : NULL;
2446
2447 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2448 if (adr_type == TypePtr::NULL_PTR) {
2449 set_map(old_map);
2450 set_sp(old_sp);
2451 return false; // off-heap access with zero address
2452 }
2453
2454 // Try to categorize the address.
2455 Compile::AliasType* alias_type = C->alias_type(adr_type);
2456 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2457
2458 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2459 alias_type->adr_type() == TypeAryPtr::RANGE) {
2460 set_map(old_map);
2461 set_sp(old_sp);
2462 return false; // not supported
2463 }
2464
2465 bool mismatched = false;
2466 BasicType bt = alias_type->basic_type();
2467 if (bt != T_ILLEGAL) {
2468 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2469 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2470 // Alias type doesn't differentiate between byte[] and boolean[]).
2471 // Use address type to get the element type.
2472 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2473 }
2474 if (is_reference_type(bt, true)) {
2475 // accessing an array field with getReference is not a mismatch
2476 bt = T_OBJECT;
2477 }
2478 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2479 // Don't intrinsify mismatched object accesses
2480 set_map(old_map);
2481 set_sp(old_sp);
2482 return false;
2483 }
2484 mismatched = (bt != type);
2485 } else if (alias_type->adr_type()->isa_oopptr()) {
2486 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2487 }
2488
2489 old_map->destruct(&_gvn);
2490 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2491
2492 if (mismatched) {
2493 decorators |= C2_MISMATCHED;
2494 }
2495
2496 // First guess at the value type.
2497 const Type *value_type = Type::get_const_basic_type(type);
2498
2499 // Figure out the memory ordering.
2500 decorators |= mo_decorator_for_access_kind(kind);
2501
2502 if (!is_store && type == T_OBJECT) {
2503 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2504 if (tjp != NULL) {
2505 value_type = tjp;
2506 }
2507 }
2508
2509 receiver = null_check(receiver);
2510 if (stopped()) {
2511 return true;
2512 }
2513 // Heap pointers get a null-check from the interpreter,
2514 // as a courtesy. However, this is not guaranteed by Unsafe,
2515 // and it is not possible to fully distinguish unintended nulls
2516 // from intended ones in this API.
2517
2518 if (!is_store) {
2519 Node* p = NULL;
2520 // Try to constant fold a load from a constant field
2521 ciField* field = alias_type->field();
2522 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2523 // final or stable field
2524 p = make_constant_from_field(field, heap_base_oop);
2525 }
2526
2527 if (p == NULL) { // Could not constant fold the load
2528 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2529 // Normalize the value returned by getBoolean in the following cases
2530 if (type == T_BOOLEAN &&
2531 (mismatched ||
2532 heap_base_oop == top() || // - heap_base_oop is NULL or
2533 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2534 // and the unsafe access is made to large offset
2535 // (i.e., larger than the maximum offset necessary for any
2536 // field access)
2537 ) {
2538 IdealKit ideal = IdealKit(this);
2539 #define __ ideal.
2540 IdealVariable normalized_result(ideal);
2541 __ declarations_done();
2542 __ set(normalized_result, p);
2543 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2544 __ set(normalized_result, ideal.ConI(1));
2545 ideal.end_if();
2546 final_sync(ideal);
2547 p = __ value(normalized_result);
2548 #undef __
2549 }
2550 }
2551 if (type == T_ADDRESS) {
2552 p = gvn().transform(new CastP2XNode(NULL, p));
2553 p = ConvX2UL(p);
2554 }
2555 // The load node has the control of the preceding MemBarCPUOrder. All
2556 // following nodes will have the control of the MemBarCPUOrder inserted at
2557 // the end of this method. So, pushing the load onto the stack at a later
2558 // point is fine.
2559 set_result(p);
2560 } else {
2561 if (bt == T_ADDRESS) {
2562 // Repackage the long as a pointer.
2563 val = ConvL2X(val);
2564 val = gvn().transform(new CastX2PNode(val));
2565 }
2566 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2567 }
2568
2569 return true;
2570 }
2571
2572 //----------------------------inline_unsafe_load_store----------------------------
2573 // This method serves a couple of different customers (depending on LoadStoreKind):
2574 //
2575 // LS_cmp_swap:
2576 //
2577 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2578 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2579 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2580 //
2581 // LS_cmp_swap_weak:
2582 //
2583 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2584 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2585 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2586 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2587 //
2588 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2589 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2590 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2591 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2592 //
2593 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2594 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2595 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2596 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2597 //
2598 // LS_cmp_exchange:
2599 //
2600 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2601 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2602 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2603 //
2604 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2605 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2606 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2607 //
2608 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2609 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2610 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2611 //
2612 // LS_get_add:
2613 //
2614 // int getAndAddInt( Object o, long offset, int delta)
2615 // long getAndAddLong(Object o, long offset, long delta)
2616 //
2617 // LS_get_set:
2618 //
2619 // int getAndSet(Object o, long offset, int newValue)
2620 // long getAndSet(Object o, long offset, long newValue)
2621 // Object getAndSet(Object o, long offset, Object newValue)
2622 //
2623 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2624 // This basic scheme here is the same as inline_unsafe_access, but
2625 // differs in enough details that combining them would make the code
2626 // overly confusing. (This is a true fact! I originally combined
2627 // them, but even I was confused by it!) As much code/comments as
2628 // possible are retained from inline_unsafe_access though to make
2629 // the correspondences clearer. - dl
2630
2631 if (callee()->is_static()) return false; // caller must have the capability!
2632
2633 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2634 decorators |= mo_decorator_for_access_kind(access_kind);
2635
2636 #ifndef PRODUCT
2637 BasicType rtype;
2638 {
2639 ResourceMark rm;
2640 // Check the signatures.
2641 ciSignature* sig = callee()->signature();
2642 rtype = sig->return_type()->basic_type();
2643 switch(kind) {
2644 case LS_get_add:
2645 case LS_get_set: {
2646 // Check the signatures.
2647 #ifdef ASSERT
2648 assert(rtype == type, "get and set must return the expected type");
2649 assert(sig->count() == 3, "get and set has 3 arguments");
2650 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2651 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2652 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2653 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2654 #endif // ASSERT
2655 break;
2656 }
2657 case LS_cmp_swap:
2658 case LS_cmp_swap_weak: {
2659 // Check the signatures.
2660 #ifdef ASSERT
2661 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2662 assert(sig->count() == 4, "CAS has 4 arguments");
2663 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2664 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2665 #endif // ASSERT
2666 break;
2667 }
2668 case LS_cmp_exchange: {
2669 // Check the signatures.
2670 #ifdef ASSERT
2671 assert(rtype == type, "CAS must return the expected type");
2672 assert(sig->count() == 4, "CAS has 4 arguments");
2673 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2674 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2675 #endif // ASSERT
2676 break;
2677 }
2678 default:
2679 ShouldNotReachHere();
2680 }
2681 }
2682 #endif //PRODUCT
2683
2684 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2685
2686 // Get arguments:
2687 Node* receiver = NULL;
2688 Node* base = NULL;
2689 Node* offset = NULL;
2690 Node* oldval = NULL;
2691 Node* newval = NULL;
2692 switch(kind) {
2693 case LS_cmp_swap:
2694 case LS_cmp_swap_weak:
2695 case LS_cmp_exchange: {
2696 const bool two_slot_type = type2size[type] == 2;
2697 receiver = argument(0); // type: oop
2698 base = argument(1); // type: oop
2699 offset = argument(2); // type: long
2700 oldval = argument(4); // type: oop, int, or long
2701 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2702 break;
2703 }
2704 case LS_get_add:
2705 case LS_get_set: {
2706 receiver = argument(0); // type: oop
2707 base = argument(1); // type: oop
2708 offset = argument(2); // type: long
2709 oldval = NULL;
2710 newval = argument(4); // type: oop, int, or long
2711 break;
2712 }
2713 default:
2714 ShouldNotReachHere();
2715 }
2716
2717 // Build field offset expression.
2718 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2719 // to be plain byte offsets, which are also the same as those accepted
2720 // by oopDesc::field_addr.
2721 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2722 // 32-bit machines ignore the high half of long offsets
2723 offset = ConvL2X(offset);
2724 // Save state and restore on bailout
2725 uint old_sp = sp();
2726 SafePointNode* old_map = clone_map();
2727 Node* adr = make_unsafe_address(base, offset,type, false);
2728 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2729
2730 Compile::AliasType* alias_type = C->alias_type(adr_type);
2731 BasicType bt = alias_type->basic_type();
2732 if (bt != T_ILLEGAL &&
2733 (is_reference_type(bt) != (type == T_OBJECT))) {
2734 // Don't intrinsify mismatched object accesses.
2735 set_map(old_map);
2736 set_sp(old_sp);
2737 return false;
2738 }
2739
2740 old_map->destruct(&_gvn);
2741
2742 // For CAS, unlike inline_unsafe_access, there seems no point in
2743 // trying to refine types. Just use the coarse types here.
2744 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2745 const Type *value_type = Type::get_const_basic_type(type);
2746
2747 switch (kind) {
2748 case LS_get_set:
2749 case LS_cmp_exchange: {
2750 if (type == T_OBJECT) {
2751 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2752 if (tjp != NULL) {
2753 value_type = tjp;
2754 }
2755 }
2756 break;
2757 }
2758 case LS_cmp_swap:
2759 case LS_cmp_swap_weak:
2760 case LS_get_add:
2761 break;
2762 default:
2763 ShouldNotReachHere();
2764 }
2765
2766 // Null check receiver.
2767 receiver = null_check(receiver);
2768 if (stopped()) {
2769 return true;
2770 }
2771
2772 int alias_idx = C->get_alias_index(adr_type);
2773
2774 if (is_reference_type(type)) {
2775 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2776
2777 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2778 // could be delayed during Parse (for example, in adjust_map_after_if()).
2779 // Execute transformation here to avoid barrier generation in such case.
2780 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2781 newval = _gvn.makecon(TypePtr::NULL_PTR);
2782
2783 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2784 // Refine the value to a null constant, when it is known to be null
2785 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2786 }
2787 }
2788
2789 Node* result = NULL;
2790 switch (kind) {
2791 case LS_cmp_exchange: {
2792 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2793 oldval, newval, value_type, type, decorators);
2794 break;
2795 }
2796 case LS_cmp_swap_weak:
2797 decorators |= C2_WEAK_CMPXCHG;
2798 case LS_cmp_swap: {
2799 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2800 oldval, newval, value_type, type, decorators);
2801 break;
2802 }
2803 case LS_get_set: {
2804 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2805 newval, value_type, type, decorators);
2806 break;
2807 }
2808 case LS_get_add: {
2809 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2810 newval, value_type, type, decorators);
2811 break;
2812 }
2813 default:
2814 ShouldNotReachHere();
2815 }
2816
2817 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2818 set_result(result);
2819 return true;
2820 }
2821
2822 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2823 // Regardless of form, don't allow previous ld/st to move down,
2824 // then issue acquire, release, or volatile mem_bar.
2825 insert_mem_bar(Op_MemBarCPUOrder);
2826 switch(id) {
2827 case vmIntrinsics::_loadFence:
2828 insert_mem_bar(Op_LoadFence);
2829 return true;
2830 case vmIntrinsics::_storeFence:
2831 insert_mem_bar(Op_StoreFence);
2832 return true;
2833 case vmIntrinsics::_storeStoreFence:
2834 insert_mem_bar(Op_StoreStoreFence);
2835 return true;
2836 case vmIntrinsics::_fullFence:
2837 insert_mem_bar(Op_MemBarVolatile);
2838 return true;
2839 default:
2840 fatal_unexpected_iid(id);
2841 return false;
2842 }
2843 }
2844
2845 bool LibraryCallKit::inline_onspinwait() {
2846 insert_mem_bar(Op_OnSpinWait);
2847 return true;
2848 }
2849
2850 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2851 if (!kls->is_Con()) {
2852 return true;
2853 }
2854 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2855 if (klsptr == NULL) {
2856 return true;
2857 }
2858 ciInstanceKlass* ik = klsptr->instance_klass();
2859 // don't need a guard for a klass that is already initialized
2860 return !ik->is_initialized();
2861 }
2862
2863 //----------------------------inline_unsafe_writeback0-------------------------
2864 // public native void Unsafe.writeback0(long address)
2865 bool LibraryCallKit::inline_unsafe_writeback0() {
2866 if (!Matcher::has_match_rule(Op_CacheWB)) {
2867 return false;
2868 }
2869 #ifndef PRODUCT
2870 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2871 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2872 ciSignature* sig = callee()->signature();
2873 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2874 #endif
2875 null_check_receiver(); // null-check, then ignore
2876 Node *addr = argument(1);
2877 addr = new CastX2PNode(addr);
2878 addr = _gvn.transform(addr);
2879 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2880 flush = _gvn.transform(flush);
2881 set_memory(flush, TypeRawPtr::BOTTOM);
2882 return true;
2883 }
2884
2885 //----------------------------inline_unsafe_writeback0-------------------------
2886 // public native void Unsafe.writeback0(long address)
2887 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2888 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2889 return false;
2890 }
2891 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2892 return false;
2893 }
2894 #ifndef PRODUCT
2895 assert(Matcher::has_match_rule(Op_CacheWB),
2896 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2897 : "found match rule for CacheWBPostSync but not CacheWB"));
2898
2899 #endif
2900 null_check_receiver(); // null-check, then ignore
2901 Node *sync;
2902 if (is_pre) {
2903 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2904 } else {
2905 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2906 }
2907 sync = _gvn.transform(sync);
2908 set_memory(sync, TypeRawPtr::BOTTOM);
2909 return true;
2910 }
2911
2912 //----------------------------inline_unsafe_allocate---------------------------
2913 // public native Object Unsafe.allocateInstance(Class<?> cls);
2914 bool LibraryCallKit::inline_unsafe_allocate() {
2915 if (callee()->is_static()) return false; // caller must have the capability!
2916
2917 null_check_receiver(); // null-check, then ignore
2918 Node* cls = null_check(argument(1));
2919 if (stopped()) return true;
2920
2921 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2922 kls = null_check(kls);
2923 if (stopped()) return true; // argument was like int.class
2924
2925 Node* test = NULL;
2926 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2927 // Note: The argument might still be an illegal value like
2928 // Serializable.class or Object[].class. The runtime will handle it.
2929 // But we must make an explicit check for initialization.
2930 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2931 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2932 // can generate code to load it as unsigned byte.
2933 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2934 Node* bits = intcon(InstanceKlass::fully_initialized);
2935 test = _gvn.transform(new SubINode(inst, bits));
2936 // The 'test' is non-zero if we need to take a slow path.
2937 }
2938
2939 Node* obj = new_instance(kls, test);
2940 set_result(obj);
2941 return true;
2942 }
2943
2944 //------------------------inline_native_time_funcs--------------
2945 // inline code for System.currentTimeMillis() and System.nanoTime()
2946 // these have the same type and signature
2947 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2948 const TypeFunc* tf = OptoRuntime::void_long_Type();
2949 const TypePtr* no_memory_effects = NULL;
2950 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2951 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2952 #ifdef ASSERT
2953 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2954 assert(value_top == top(), "second value must be top");
2955 #endif
2956 set_result(value);
2957 return true;
2958 }
2959
2960 #ifdef JFR_HAVE_INTRINSICS
2961
2962 /**
2963 * if oop->klass != null
2964 * // normal class
2965 * epoch = _epoch_state ? 2 : 1
2966 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2967 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2968 * }
2969 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2970 * else
2971 * // primitive class
2972 * if oop->array_klass != null
2973 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2974 * else
2975 * id = LAST_TYPE_ID + 1 // void class path
2976 * if (!signaled)
2977 * signaled = true
2978 */
2979 bool LibraryCallKit::inline_native_classID() {
2980 Node* cls = argument(0);
2981
2982 IdealKit ideal(this);
2983 #define __ ideal.
2984 IdealVariable result(ideal); __ declarations_done();
2985 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2986 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2987 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2988
2989
2990 __ if_then(kls, BoolTest::ne, null()); {
2991 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2992 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2993
2994 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
2995 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2996 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2997 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2998 mask = _gvn.transform(new OrLNode(mask, epoch));
2999 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3000
3001 float unlikely = PROB_UNLIKELY(0.999);
3002 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3003 sync_kit(ideal);
3004 make_runtime_call(RC_LEAF,
3005 OptoRuntime::class_id_load_barrier_Type(),
3006 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3007 "class id load barrier",
3008 TypePtr::BOTTOM,
3009 kls);
3010 ideal.sync_kit(this);
3011 } __ end_if();
3012
3013 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3014 } __ else_(); {
3015 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3016 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3017 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3018 __ if_then(array_kls, BoolTest::ne, null()); {
3019 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3020 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3021 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3022 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3023 } __ else_(); {
3024 // void class case
3025 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3026 } __ end_if();
3027
3028 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3029 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3030 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3031 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3032 } __ end_if();
3033 } __ end_if();
3034
3035 final_sync(ideal);
3036 set_result(ideal.value(result));
3037 #undef __
3038 return true;
3039 }
3040
3041 /*
3042 * The intrinsic is a model of this pseudo-code:
3043 *
3044 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3045 * jobject h_event_writer = tl->java_event_writer();
3046 * if (h_event_writer == NULL) {
3047 * return NULL;
3048 * }
3049 * oop threadObj = Thread::threadObj();
3050 * oop vthread = java_lang_Thread::vthread(threadObj);
3051 * traceid tid;
3052 * bool excluded;
3053 * if (vthread != threadObj) { // i.e. current thread is virtual
3054 * tid = java_lang_Thread::tid(vthread);
3055 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3056 * excluded = vthread_epoch_raw & excluded_mask;
3057 * if (!excluded) {
3058 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3059 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3060 * if (vthread_epoch != current_epoch) {
3061 * write_checkpoint();
3062 * }
3063 * }
3064 * } else {
3065 * tid = java_lang_Thread::tid(threadObj);
3066 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3067 * excluded = thread_epoch_raw & excluded_mask;
3068 * }
3069 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3070 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3071 * if (tid_in_event_writer != tid) {
3072 * setField(event_writer, "threadID", tid);
3073 * setField(event_writer, "excluded", excluded);
3074 * }
3075 * return event_writer
3076 */
3077 bool LibraryCallKit::inline_native_getEventWriter() {
3078 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3079
3080 // Save input memory and i_o state.
3081 Node* input_memory_state = reset_memory();
3082 set_all_memory(input_memory_state);
3083 Node* input_io_state = i_o();
3084
3085 Node* excluded_mask = _gvn.intcon(32768);
3086 Node* epoch_mask = _gvn.intcon(32767);
3087
3088 // TLS
3089 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3090
3091 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3092 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3093
3094 // Load the eventwriter jobject handle.
3095 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3096
3097 // Null check the jobject handle.
3098 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3099 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3100 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3101
3102 // False path, jobj is null.
3103 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3104
3105 // True path, jobj is not null.
3106 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3107
3108 set_control(jobj_is_not_null);
3109
3110 // Load the threadObj for the CarrierThread.
3111 Node* threadObj = generate_current_thread(tls_ptr);
3112
3113 // Load the vthread.
3114 Node* vthread = generate_virtual_thread(tls_ptr);
3115
3116 // If vthread != threadObj, this is a virtual thread.
3117 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3118 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3119 IfNode* iff_vthread_not_equal_threadObj =
3120 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3121
3122 // False branch, fallback to threadObj.
3123 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3124 set_control(vthread_equal_threadObj);
3125
3126 // Load the tid field from the vthread object.
3127 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3128
3129 // Load the raw epoch value from the threadObj.
3130 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3131 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3132 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3133
3134 // Mask off the excluded information from the epoch.
3135 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3136
3137 // True branch, this is a virtual thread.
3138 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3139 set_control(vthread_not_equal_threadObj);
3140
3141 // Load the tid field from the vthread object.
3142 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3143
3144 // Load the raw epoch value from the vthread.
3145 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3146 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3147 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3148
3149 // Mask off the excluded information from the epoch.
3150 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3151
3152 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3153 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3154 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3155 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3156
3157 // False branch, vthread is excluded, no need to write epoch info.
3158 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3159
3160 // True branch, vthread is included, update epoch info.
3161 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3162 set_control(included);
3163
3164 // Get epoch value.
3165 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3166
3167 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3168 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3169 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3170
3171 // Compare the epoch in the vthread to the current epoch generation.
3172 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3173 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3174 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3175
3176 // False path, epoch is equal, checkpoint information is valid.
3177 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3178
3179 // True path, epoch is not equal, write a checkpoint for the vthread.
3180 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3181
3182 set_control(epoch_is_not_equal);
3183
3184 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3185 // The call also updates the native thread local thread id and the vthread with the current epoch.
3186 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3187 OptoRuntime::jfr_write_checkpoint_Type(),
3188 StubRoutines::jfr_write_checkpoint(),
3189 "write_checkpoint", TypePtr::BOTTOM);
3190 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3191
3192 // vthread epoch != current epoch
3193 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3194 record_for_igvn(epoch_compare_rgn);
3195 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3196 record_for_igvn(epoch_compare_mem);
3197 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3198 record_for_igvn(epoch_compare_io);
3199
3200 // Update control and phi nodes.
3201 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3202 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3203 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3204 epoch_compare_mem->init_req(_false_path, input_memory_state);
3205 epoch_compare_io->init_req(_true_path, i_o());
3206 epoch_compare_io->init_req(_false_path, input_io_state);
3207
3208 // excluded != true
3209 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3210 record_for_igvn(exclude_compare_rgn);
3211 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3212 record_for_igvn(exclude_compare_mem);
3213 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3214 record_for_igvn(exclude_compare_io);
3215
3216 // Update control and phi nodes.
3217 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3218 exclude_compare_rgn->init_req(_false_path, excluded);
3219 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3220 exclude_compare_mem->init_req(_false_path, input_memory_state);
3221 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3222 exclude_compare_io->init_req(_false_path, input_io_state);
3223
3224 // vthread != threadObj
3225 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3226 record_for_igvn(vthread_compare_rgn);
3227 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3228 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3229 record_for_igvn(vthread_compare_io);
3230 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3231 record_for_igvn(tid);
3232 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3233 record_for_igvn(exclusion);
3234
3235 // Update control and phi nodes.
3236 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3237 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3238 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3239 vthread_compare_mem->init_req(_false_path, input_memory_state);
3240 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3241 vthread_compare_io->init_req(_false_path, input_io_state);
3242 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3243 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3244 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3245 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3246
3247 // Update branch state.
3248 set_control(_gvn.transform(vthread_compare_rgn));
3249 set_all_memory(_gvn.transform(vthread_compare_mem));
3250 set_i_o(_gvn.transform(vthread_compare_io));
3251
3252 // Load the event writer oop by dereferencing the jobject handle.
3253 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3254 assert(klass_EventWriter->is_loaded(), "invariant");
3255 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3256 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3257 const TypeOopPtr* const xtype = aklass->as_instance_type();
3258 Node* event_writer = access_load(jobj, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3259
3260 // Load the current thread id from the event writer object.
3261 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3262 // Get the field offset to, conditionally, store an updated tid value later.
3263 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3264 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3265 // Get the field offset to, conditionally, store an updated exclusion value later.
3266 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3267 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3268
3269 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3270 record_for_igvn(event_writer_tid_compare_rgn);
3271 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3272 record_for_igvn(event_writer_tid_compare_mem);
3273 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3274 record_for_igvn(event_writer_tid_compare_io);
3275
3276 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3277 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3278 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3279 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3280
3281 // False path, tids are the same.
3282 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3283
3284 // True path, tid is not equal, need to update the tid in the event writer.
3285 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3286 record_for_igvn(tid_is_not_equal);
3287
3288 // Store the exclusion state to the event writer.
3289 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3290
3291 // Store the tid to the event writer.
3292 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3293
3294 // Update control and phi nodes.
3295 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3296 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3297 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3298 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3299 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3300 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3301
3302 // Result of top level CFG, Memory, IO and Value.
3303 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3304 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3305 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3306 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3307
3308 // Result control.
3309 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3310 result_rgn->init_req(_false_path, jobj_is_null);
3311
3312 // Result memory.
3313 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3314 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3315
3316 // Result IO.
3317 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3318 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3319
3320 // Result value.
3321 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3322 result_value->init_req(_false_path, null()); // return NULL
3323
3324 // Set output state.
3325 set_control(_gvn.transform(result_rgn));
3326 set_all_memory(_gvn.transform(result_mem));
3327 set_i_o(_gvn.transform(result_io));
3328 set_result(result_rgn, result_value);
3329 return true;
3330 }
3331
3332 /*
3333 * The intrinsic is a model of this pseudo-code:
3334 *
3335 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3336 * if (carrierThread != thread) { // is virtual thread
3337 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3338 * bool excluded = vthread_epoch_raw & excluded_mask;
3339 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3340 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3341 * if (!excluded) {
3342 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3343 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3344 * }
3345 * Atomic::release_store(&tl->_vthread, true);
3346 * return;
3347 * }
3348 * Atomic::release_store(&tl->_vthread, false);
3349 */
3350 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3351 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3352
3353 Node* input_memory_state = reset_memory();
3354 set_all_memory(input_memory_state);
3355
3356 Node* excluded_mask = _gvn.intcon(32768);
3357 Node* epoch_mask = _gvn.intcon(32767);
3358
3359 Node* const carrierThread = generate_current_thread(jt);
3360 // If thread != carrierThread, this is a virtual thread.
3361 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3362 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3363 IfNode* iff_thread_not_equal_carrierThread =
3364 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3365
3366 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3367
3368 // False branch, is carrierThread.
3369 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3370 // Store release
3371 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3372
3373 set_all_memory(input_memory_state);
3374
3375 // True branch, is virtual thread.
3376 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3377 set_control(thread_not_equal_carrierThread);
3378
3379 // Load the raw epoch value from the vthread.
3380 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3381 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3382 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3383
3384 // Mask off the excluded information from the epoch.
3385 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3386
3387 // Load the tid field from the thread.
3388 Node* tid = load_field_from_object(thread, "tid", "J");
3389
3390 // Store the vthread tid to the jfr thread local.
3391 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3392 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3393
3394 // Branch is_excluded to conditionalize updating the epoch .
3395 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3396 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3397 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3398
3399 // True branch, vthread is excluded, no need to write epoch info.
3400 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3401 set_control(excluded);
3402 Node* vthread_is_excluded = _gvn.intcon(1);
3403
3404 // False branch, vthread is included, update epoch info.
3405 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3406 set_control(included);
3407 Node* vthread_is_included = _gvn.intcon(0);
3408
3409 // Get epoch value.
3410 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3411
3412 // Store the vthread epoch to the jfr thread local.
3413 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3414 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3415
3416 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3417 record_for_igvn(excluded_rgn);
3418 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3419 record_for_igvn(excluded_mem);
3420 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3421 record_for_igvn(exclusion);
3422
3423 // Merge the excluded control and memory.
3424 excluded_rgn->init_req(_true_path, excluded);
3425 excluded_rgn->init_req(_false_path, included);
3426 excluded_mem->init_req(_true_path, tid_memory);
3427 excluded_mem->init_req(_false_path, included_memory);
3428 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3429 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3430
3431 // Set intermediate state.
3432 set_control(_gvn.transform(excluded_rgn));
3433 set_all_memory(excluded_mem);
3434
3435 // Store the vthread exclusion state to the jfr thread local.
3436 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3437 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3438
3439 // Store release
3440 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3441
3442 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3443 record_for_igvn(thread_compare_rgn);
3444 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3445 record_for_igvn(thread_compare_mem);
3446 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3447 record_for_igvn(vthread);
3448
3449 // Merge the thread_compare control and memory.
3450 thread_compare_rgn->init_req(_true_path, control());
3451 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3452 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3453 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3454
3455 // Set output state.
3456 set_control(_gvn.transform(thread_compare_rgn));
3457 set_all_memory(_gvn.transform(thread_compare_mem));
3458 }
3459
3460 #endif // JFR_HAVE_INTRINSICS
3461
3462 //------------------------inline_native_currentCarrierThread------------------
3463 bool LibraryCallKit::inline_native_currentCarrierThread() {
3464 Node* junk = NULL;
3465 set_result(generate_current_thread(junk));
3466 return true;
3467 }
3468
3469 //------------------------inline_native_currentThread------------------
3470 bool LibraryCallKit::inline_native_currentThread() {
3471 Node* junk = NULL;
3472 set_result(generate_virtual_thread(junk));
3473 return true;
3474 }
3475
3476 //------------------------inline_native_setVthread------------------
3477 bool LibraryCallKit::inline_native_setCurrentThread() {
3478 assert(C->method()->changes_current_thread(),
3479 "method changes current Thread but is not annotated ChangesCurrentThread");
3480 Node* arr = argument(1);
3481 Node* thread = _gvn.transform(new ThreadLocalNode());
3482 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3483 Node* thread_obj_handle
3484 = make_load(NULL, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3485 thread_obj_handle = _gvn.transform(thread_obj_handle);
3486 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3487 // Stores of oops to native memory not supported yet by BarrierSetC2::store_at_resolved
3488 // access_store_at(NULL, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3489 store_to_memory(control(), thread_obj_handle, arr, T_OBJECT, adr_type, MemNode::unordered);
3490 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3491 return true;
3492 }
3493
3494 Node* LibraryCallKit::extentLocalCache_helper() {
3495 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3496 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3497
3498 bool xk = etype->klass_is_exact();
3499
3500 Node* thread = _gvn.transform(new ThreadLocalNode());
3501 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::extentLocalCache_offset()));
3502 // We cannot use immutable_memory() because we might flip onto a
3503 // different carrier thread, at which point we'll need to use that
3504 // carrier thread's cache.
3505 // return _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
3506 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3507 return make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3508 }
3509
3510 //------------------------inline_native_extentLocalCache------------------
3511 bool LibraryCallKit::inline_native_extentLocalCache() {
3512 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3513 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3514 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3515
3516 // Because we create the extentLocal cache lazily we have to make the
3517 // type of the result BotPTR.
3518 bool xk = etype->klass_is_exact();
3519 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3520 Node* cache_obj_handle = extentLocalCache_helper();
3521 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3522
3523 return true;
3524 }
3525
3526 //------------------------inline_native_setExtentLocalCache------------------
3527 bool LibraryCallKit::inline_native_setExtentLocalCache() {
3528 Node* arr = argument(0);
3529 Node* cache_obj_handle = extentLocalCache_helper();
3530
3531 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3532 store_to_memory(control(), cache_obj_handle, arr, T_OBJECT, adr_type,
3533 MemNode::unordered);
3534
3535 return true;
3536 }
3537
3538 //---------------------------load_mirror_from_klass----------------------------
3539 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3540 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3541 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3542 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3543 // mirror = ((OopHandle)mirror)->resolve();
3544 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3545 }
3546
3547 //-----------------------load_klass_from_mirror_common-------------------------
3548 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3549 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3550 // and branch to the given path on the region.
3551 // If never_see_null, take an uncommon trap on null, so we can optimistically
3552 // compile for the non-null case.
3553 // If the region is NULL, force never_see_null = true.
3554 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3555 bool never_see_null,
3556 RegionNode* region,
3557 int null_path,
3558 int offset) {
3559 if (region == NULL) never_see_null = true;
3560 Node* p = basic_plus_adr(mirror, offset);
3561 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3562 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3563 Node* null_ctl = top();
3564 kls = null_check_oop(kls, &null_ctl, never_see_null);
3565 if (region != NULL) {
3566 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3567 region->init_req(null_path, null_ctl);
3568 } else {
3569 assert(null_ctl == top(), "no loose ends");
3570 }
3571 return kls;
3572 }
3573
3574 //--------------------(inline_native_Class_query helpers)---------------------
3575 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3576 // Fall through if (mods & mask) == bits, take the guard otherwise.
3577 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3578 // Branch around if the given klass has the given modifier bit set.
3579 // Like generate_guard, adds a new path onto the region.
3580 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3581 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3582 Node* mask = intcon(modifier_mask);
3583 Node* bits = intcon(modifier_bits);
3584 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3585 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3586 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3587 return generate_fair_guard(bol, region);
3588 }
3589 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3590 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3591 }
3592 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3593 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3594 }
3595
3596 //-------------------------inline_native_Class_query-------------------
3597 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3598 const Type* return_type = TypeInt::BOOL;
3599 Node* prim_return_value = top(); // what happens if it's a primitive class?
3600 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3601 bool expect_prim = false; // most of these guys expect to work on refs
3602
3603 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3604
3605 Node* mirror = argument(0);
3606 Node* obj = top();
3607
3608 switch (id) {
3609 case vmIntrinsics::_isInstance:
3610 // nothing is an instance of a primitive type
3611 prim_return_value = intcon(0);
3612 obj = argument(1);
3613 break;
3614 case vmIntrinsics::_getModifiers:
3615 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3616 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3617 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3618 break;
3619 case vmIntrinsics::_isInterface:
3620 prim_return_value = intcon(0);
3621 break;
3622 case vmIntrinsics::_isArray:
3623 prim_return_value = intcon(0);
3624 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3625 break;
3626 case vmIntrinsics::_isPrimitive:
3627 prim_return_value = intcon(1);
3628 expect_prim = true; // obviously
3629 break;
3630 case vmIntrinsics::_isHidden:
3631 prim_return_value = intcon(0);
3632 break;
3633 case vmIntrinsics::_getSuperclass:
3634 prim_return_value = null();
3635 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3636 break;
3637 case vmIntrinsics::_getClassAccessFlags:
3638 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3639 return_type = TypeInt::INT; // not bool! 6297094
3640 break;
3641 default:
3642 fatal_unexpected_iid(id);
3643 break;
3644 }
3645
3646 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3647 if (mirror_con == NULL) return false; // cannot happen?
3648
3649 #ifndef PRODUCT
3650 if (C->print_intrinsics() || C->print_inlining()) {
3651 ciType* k = mirror_con->java_mirror_type();
3652 if (k) {
3653 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3654 k->print_name();
3655 tty->cr();
3656 }
3657 }
3658 #endif
3659
3660 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3661 RegionNode* region = new RegionNode(PATH_LIMIT);
3662 record_for_igvn(region);
3663 PhiNode* phi = new PhiNode(region, return_type);
3664
3665 // The mirror will never be null of Reflection.getClassAccessFlags, however
3666 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3667 // if it is. See bug 4774291.
3668
3669 // For Reflection.getClassAccessFlags(), the null check occurs in
3670 // the wrong place; see inline_unsafe_access(), above, for a similar
3671 // situation.
3672 mirror = null_check(mirror);
3673 // If mirror or obj is dead, only null-path is taken.
3674 if (stopped()) return true;
3675
3676 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3677
3678 // Now load the mirror's klass metaobject, and null-check it.
3679 // Side-effects region with the control path if the klass is null.
3680 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3681 // If kls is null, we have a primitive mirror.
3682 phi->init_req(_prim_path, prim_return_value);
3683 if (stopped()) { set_result(region, phi); return true; }
3684 bool safe_for_replace = (region->in(_prim_path) == top());
3685
3686 Node* p; // handy temp
3687 Node* null_ctl;
3688
3689 // Now that we have the non-null klass, we can perform the real query.
3690 // For constant classes, the query will constant-fold in LoadNode::Value.
3691 Node* query_value = top();
3692 switch (id) {
3693 case vmIntrinsics::_isInstance:
3694 // nothing is an instance of a primitive type
3695 query_value = gen_instanceof(obj, kls, safe_for_replace);
3696 break;
3697
3698 case vmIntrinsics::_getModifiers:
3699 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3700 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3701 break;
3702
3703 case vmIntrinsics::_isInterface:
3704 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3705 if (generate_interface_guard(kls, region) != NULL)
3706 // A guard was added. If the guard is taken, it was an interface.
3707 phi->add_req(intcon(1));
3708 // If we fall through, it's a plain class.
3709 query_value = intcon(0);
3710 break;
3711
3712 case vmIntrinsics::_isArray:
3713 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3714 if (generate_array_guard(kls, region) != NULL)
3715 // A guard was added. If the guard is taken, it was an array.
3716 phi->add_req(intcon(1));
3717 // If we fall through, it's a plain class.
3718 query_value = intcon(0);
3719 break;
3720
3721 case vmIntrinsics::_isPrimitive:
3722 query_value = intcon(0); // "normal" path produces false
3723 break;
3724
3725 case vmIntrinsics::_isHidden:
3726 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3727 if (generate_hidden_class_guard(kls, region) != NULL)
3728 // A guard was added. If the guard is taken, it was an hidden class.
3729 phi->add_req(intcon(1));
3730 // If we fall through, it's a plain class.
3731 query_value = intcon(0);
3732 break;
3733
3734
3735 case vmIntrinsics::_getSuperclass:
3736 // The rules here are somewhat unfortunate, but we can still do better
3737 // with random logic than with a JNI call.
3738 // Interfaces store null or Object as _super, but must report null.
3739 // Arrays store an intermediate super as _super, but must report Object.
3740 // Other types can report the actual _super.
3741 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3742 if (generate_interface_guard(kls, region) != NULL)
3743 // A guard was added. If the guard is taken, it was an interface.
3744 phi->add_req(null());
3745 if (generate_array_guard(kls, region) != NULL)
3746 // A guard was added. If the guard is taken, it was an array.
3747 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3748 // If we fall through, it's a plain class. Get its _super.
3749 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3750 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3751 null_ctl = top();
3752 kls = null_check_oop(kls, &null_ctl);
3753 if (null_ctl != top()) {
3754 // If the guard is taken, Object.superClass is null (both klass and mirror).
3755 region->add_req(null_ctl);
3756 phi ->add_req(null());
3757 }
3758 if (!stopped()) {
3759 query_value = load_mirror_from_klass(kls);
3760 }
3761 break;
3762
3763 case vmIntrinsics::_getClassAccessFlags:
3764 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3765 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3766 break;
3767
3768 default:
3769 fatal_unexpected_iid(id);
3770 break;
3771 }
3772
3773 // Fall-through is the normal case of a query to a real class.
3774 phi->init_req(1, query_value);
3775 region->init_req(1, control());
3776
3777 C->set_has_split_ifs(true); // Has chance for split-if optimization
3778 set_result(region, phi);
3779 return true;
3780 }
3781
3782 //-------------------------inline_Class_cast-------------------
3783 bool LibraryCallKit::inline_Class_cast() {
3784 Node* mirror = argument(0); // Class
3785 Node* obj = argument(1);
3786 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3787 if (mirror_con == NULL) {
3788 return false; // dead path (mirror->is_top()).
3789 }
3790 if (obj == NULL || obj->is_top()) {
3791 return false; // dead path
3792 }
3793 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3794
3795 // First, see if Class.cast() can be folded statically.
3796 // java_mirror_type() returns non-null for compile-time Class constants.
3797 ciType* tm = mirror_con->java_mirror_type();
3798 if (tm != NULL && tm->is_klass() &&
3799 tp != NULL) {
3800 if (!tp->is_loaded()) {
3801 // Don't use intrinsic when class is not loaded.
3802 return false;
3803 } else {
3804 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass()), tp->as_klass_type());
3805 if (static_res == Compile::SSC_always_true) {
3806 // isInstance() is true - fold the code.
3807 set_result(obj);
3808 return true;
3809 } else if (static_res == Compile::SSC_always_false) {
3810 // Don't use intrinsic, have to throw ClassCastException.
3811 // If the reference is null, the non-intrinsic bytecode will
3812 // be optimized appropriately.
3813 return false;
3814 }
3815 }
3816 }
3817
3818 // Bailout intrinsic and do normal inlining if exception path is frequent.
3819 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3820 return false;
3821 }
3822
3823 // Generate dynamic checks.
3824 // Class.cast() is java implementation of _checkcast bytecode.
3825 // Do checkcast (Parse::do_checkcast()) optimizations here.
3826
3827 mirror = null_check(mirror);
3828 // If mirror is dead, only null-path is taken.
3829 if (stopped()) {
3830 return true;
3831 }
3832
3833 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3834 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3835 RegionNode* region = new RegionNode(PATH_LIMIT);
3836 record_for_igvn(region);
3837
3838 // Now load the mirror's klass metaobject, and null-check it.
3839 // If kls is null, we have a primitive mirror and
3840 // nothing is an instance of a primitive type.
3841 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3842
3843 Node* res = top();
3844 if (!stopped()) {
3845 Node* bad_type_ctrl = top();
3846 // Do checkcast optimizations.
3847 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3848 region->init_req(_bad_type_path, bad_type_ctrl);
3849 }
3850 if (region->in(_prim_path) != top() ||
3851 region->in(_bad_type_path) != top()) {
3852 // Let Interpreter throw ClassCastException.
3853 PreserveJVMState pjvms(this);
3854 set_control(_gvn.transform(region));
3855 uncommon_trap(Deoptimization::Reason_intrinsic,
3856 Deoptimization::Action_maybe_recompile);
3857 }
3858 if (!stopped()) {
3859 set_result(res);
3860 }
3861 return true;
3862 }
3863
3864
3865 //--------------------------inline_native_subtype_check------------------------
3866 // This intrinsic takes the JNI calls out of the heart of
3867 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3868 bool LibraryCallKit::inline_native_subtype_check() {
3869 // Pull both arguments off the stack.
3870 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3871 args[0] = argument(0);
3872 args[1] = argument(1);
3873 Node* klasses[2]; // corresponding Klasses: superk, subk
3874 klasses[0] = klasses[1] = top();
3875
3876 enum {
3877 // A full decision tree on {superc is prim, subc is prim}:
3878 _prim_0_path = 1, // {P,N} => false
3879 // {P,P} & superc!=subc => false
3880 _prim_same_path, // {P,P} & superc==subc => true
3881 _prim_1_path, // {N,P} => false
3882 _ref_subtype_path, // {N,N} & subtype check wins => true
3883 _both_ref_path, // {N,N} & subtype check loses => false
3884 PATH_LIMIT
3885 };
3886
3887 RegionNode* region = new RegionNode(PATH_LIMIT);
3888 Node* phi = new PhiNode(region, TypeInt::BOOL);
3889 record_for_igvn(region);
3890
3891 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3892 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3893 int class_klass_offset = java_lang_Class::klass_offset();
3894
3895 // First null-check both mirrors and load each mirror's klass metaobject.
3896 int which_arg;
3897 for (which_arg = 0; which_arg <= 1; which_arg++) {
3898 Node* arg = args[which_arg];
3899 arg = null_check(arg);
3900 if (stopped()) break;
3901 args[which_arg] = arg;
3902
3903 Node* p = basic_plus_adr(arg, class_klass_offset);
3904 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3905 klasses[which_arg] = _gvn.transform(kls);
3906 }
3907
3908 // Having loaded both klasses, test each for null.
3909 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3910 for (which_arg = 0; which_arg <= 1; which_arg++) {
3911 Node* kls = klasses[which_arg];
3912 Node* null_ctl = top();
3913 kls = null_check_oop(kls, &null_ctl, never_see_null);
3914 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3915 region->init_req(prim_path, null_ctl);
3916 if (stopped()) break;
3917 klasses[which_arg] = kls;
3918 }
3919
3920 if (!stopped()) {
3921 // now we have two reference types, in klasses[0..1]
3922 Node* subk = klasses[1]; // the argument to isAssignableFrom
3923 Node* superk = klasses[0]; // the receiver
3924 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3925 // now we have a successful reference subtype check
3926 region->set_req(_ref_subtype_path, control());
3927 }
3928
3929 // If both operands are primitive (both klasses null), then
3930 // we must return true when they are identical primitives.
3931 // It is convenient to test this after the first null klass check.
3932 set_control(region->in(_prim_0_path)); // go back to first null check
3933 if (!stopped()) {
3934 // Since superc is primitive, make a guard for the superc==subc case.
3935 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3936 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3937 generate_guard(bol_eq, region, PROB_FAIR);
3938 if (region->req() == PATH_LIMIT+1) {
3939 // A guard was added. If the added guard is taken, superc==subc.
3940 region->swap_edges(PATH_LIMIT, _prim_same_path);
3941 region->del_req(PATH_LIMIT);
3942 }
3943 region->set_req(_prim_0_path, control()); // Not equal after all.
3944 }
3945
3946 // these are the only paths that produce 'true':
3947 phi->set_req(_prim_same_path, intcon(1));
3948 phi->set_req(_ref_subtype_path, intcon(1));
3949
3950 // pull together the cases:
3951 assert(region->req() == PATH_LIMIT, "sane region");
3952 for (uint i = 1; i < region->req(); i++) {
3953 Node* ctl = region->in(i);
3954 if (ctl == NULL || ctl == top()) {
3955 region->set_req(i, top());
3956 phi ->set_req(i, top());
3957 } else if (phi->in(i) == NULL) {
3958 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3959 }
3960 }
3961
3962 set_control(_gvn.transform(region));
3963 set_result(_gvn.transform(phi));
3964 return true;
3965 }
3966
3967 //---------------------generate_array_guard_common------------------------
3968 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3969 bool obj_array, bool not_array) {
3970
3971 if (stopped()) {
3972 return NULL;
3973 }
3974
3975 // If obj_array/non_array==false/false:
3976 // Branch around if the given klass is in fact an array (either obj or prim).
3977 // If obj_array/non_array==false/true:
3978 // Branch around if the given klass is not an array klass of any kind.
3979 // If obj_array/non_array==true/true:
3980 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3981 // If obj_array/non_array==true/false:
3982 // Branch around if the kls is an oop array (Object[] or subtype)
3983 //
3984 // Like generate_guard, adds a new path onto the region.
3985 jint layout_con = 0;
3986 Node* layout_val = get_layout_helper(kls, layout_con);
3987 if (layout_val == NULL) {
3988 bool query = (obj_array
3989 ? Klass::layout_helper_is_objArray(layout_con)
3990 : Klass::layout_helper_is_array(layout_con));
3991 if (query == not_array) {
3992 return NULL; // never a branch
3993 } else { // always a branch
3994 Node* always_branch = control();
3995 if (region != NULL)
3996 region->add_req(always_branch);
3997 set_control(top());
3998 return always_branch;
3999 }
4000 }
4001 // Now test the correct condition.
4002 jint nval = (obj_array
4003 ? (jint)(Klass::_lh_array_tag_type_value
4004 << Klass::_lh_array_tag_shift)
4005 : Klass::_lh_neutral_value);
4006 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4007 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4008 // invert the test if we are looking for a non-array
4009 if (not_array) btest = BoolTest(btest).negate();
4010 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4011 return generate_fair_guard(bol, region);
4012 }
4013
4014
4015 //-----------------------inline_native_newArray--------------------------
4016 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4017 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4018 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4019 Node* mirror;
4020 Node* count_val;
4021 if (uninitialized) {
4022 mirror = argument(1);
4023 count_val = argument(2);
4024 } else {
4025 mirror = argument(0);
4026 count_val = argument(1);
4027 }
4028
4029 mirror = null_check(mirror);
4030 // If mirror or obj is dead, only null-path is taken.
4031 if (stopped()) return true;
4032
4033 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4034 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4035 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4036 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4037 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4038
4039 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4040 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4041 result_reg, _slow_path);
4042 Node* normal_ctl = control();
4043 Node* no_array_ctl = result_reg->in(_slow_path);
4044
4045 // Generate code for the slow case. We make a call to newArray().
4046 set_control(no_array_ctl);
4047 if (!stopped()) {
4048 // Either the input type is void.class, or else the
4049 // array klass has not yet been cached. Either the
4050 // ensuing call will throw an exception, or else it
4051 // will cache the array klass for next time.
4052 PreserveJVMState pjvms(this);
4053 CallJavaNode* slow_call = NULL;
4054 if (uninitialized) {
4055 // Generate optimized virtual call (holder class 'Unsafe' is final)
4056 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
4057 } else {
4058 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
4059 }
4060 Node* slow_result = set_results_for_java_call(slow_call);
4061 // this->control() comes from set_results_for_java_call
4062 result_reg->set_req(_slow_path, control());
4063 result_val->set_req(_slow_path, slow_result);
4064 result_io ->set_req(_slow_path, i_o());
4065 result_mem->set_req(_slow_path, reset_memory());
4066 }
4067
4068 set_control(normal_ctl);
4069 if (!stopped()) {
4070 // Normal case: The array type has been cached in the java.lang.Class.
4071 // The following call works fine even if the array type is polymorphic.
4072 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4073 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4074 result_reg->init_req(_normal_path, control());
4075 result_val->init_req(_normal_path, obj);
4076 result_io ->init_req(_normal_path, i_o());
4077 result_mem->init_req(_normal_path, reset_memory());
4078
4079 if (uninitialized) {
4080 // Mark the allocation so that zeroing is skipped
4081 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
4082 alloc->maybe_set_complete(&_gvn);
4083 }
4084 }
4085
4086 // Return the combined state.
4087 set_i_o( _gvn.transform(result_io) );
4088 set_all_memory( _gvn.transform(result_mem));
4089
4090 C->set_has_split_ifs(true); // Has chance for split-if optimization
4091 set_result(result_reg, result_val);
4092 return true;
4093 }
4094
4095 //----------------------inline_native_getLength--------------------------
4096 // public static native int java.lang.reflect.Array.getLength(Object array);
4097 bool LibraryCallKit::inline_native_getLength() {
4098 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4099
4100 Node* array = null_check(argument(0));
4101 // If array is dead, only null-path is taken.
4102 if (stopped()) return true;
4103
4104 // Deoptimize if it is a non-array.
4105 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
4106
4107 if (non_array != NULL) {
4108 PreserveJVMState pjvms(this);
4109 set_control(non_array);
4110 uncommon_trap(Deoptimization::Reason_intrinsic,
4111 Deoptimization::Action_maybe_recompile);
4112 }
4113
4114 // If control is dead, only non-array-path is taken.
4115 if (stopped()) return true;
4116
4117 // The works fine even if the array type is polymorphic.
4118 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4119 Node* result = load_array_length(array);
4120
4121 C->set_has_split_ifs(true); // Has chance for split-if optimization
4122 set_result(result);
4123 return true;
4124 }
4125
4126 //------------------------inline_array_copyOf----------------------------
4127 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4128 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4129 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4130 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4131
4132 // Get the arguments.
4133 Node* original = argument(0);
4134 Node* start = is_copyOfRange? argument(1): intcon(0);
4135 Node* end = is_copyOfRange? argument(2): argument(1);
4136 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4137
4138 Node* newcopy = NULL;
4139
4140 // Set the original stack and the reexecute bit for the interpreter to reexecute
4141 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4142 { PreserveReexecuteState preexecs(this);
4143 jvms()->set_should_reexecute(true);
4144
4145 array_type_mirror = null_check(array_type_mirror);
4146 original = null_check(original);
4147
4148 // Check if a null path was taken unconditionally.
4149 if (stopped()) return true;
4150
4151 Node* orig_length = load_array_length(original);
4152
4153 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4154 klass_node = null_check(klass_node);
4155
4156 RegionNode* bailout = new RegionNode(1);
4157 record_for_igvn(bailout);
4158
4159 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4160 // Bail out if that is so.
4161 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4162 if (not_objArray != NULL) {
4163 // Improve the klass node's type from the new optimistic assumption:
4164 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4165 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4166 Node* cast = new CastPPNode(klass_node, akls);
4167 cast->init_req(0, control());
4168 klass_node = _gvn.transform(cast);
4169 }
4170
4171 // Bail out if either start or end is negative.
4172 generate_negative_guard(start, bailout, &start);
4173 generate_negative_guard(end, bailout, &end);
4174
4175 Node* length = end;
4176 if (_gvn.type(start) != TypeInt::ZERO) {
4177 length = _gvn.transform(new SubINode(end, start));
4178 }
4179
4180 // Bail out if length is negative.
4181 // Without this the new_array would throw
4182 // NegativeArraySizeException but IllegalArgumentException is what
4183 // should be thrown
4184 generate_negative_guard(length, bailout, &length);
4185
4186 if (bailout->req() > 1) {
4187 PreserveJVMState pjvms(this);
4188 set_control(_gvn.transform(bailout));
4189 uncommon_trap(Deoptimization::Reason_intrinsic,
4190 Deoptimization::Action_maybe_recompile);
4191 }
4192
4193 if (!stopped()) {
4194 // How many elements will we copy from the original?
4195 // The answer is MinI(orig_length - start, length).
4196 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4197 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4198
4199 // Generate a direct call to the right arraycopy function(s).
4200 // We know the copy is disjoint but we might not know if the
4201 // oop stores need checking.
4202 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4203 // This will fail a store-check if x contains any non-nulls.
4204
4205 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4206 // loads/stores but it is legal only if we're sure the
4207 // Arrays.copyOf would succeed. So we need all input arguments
4208 // to the copyOf to be validated, including that the copy to the
4209 // new array won't trigger an ArrayStoreException. That subtype
4210 // check can be optimized if we know something on the type of
4211 // the input array from type speculation.
4212 if (_gvn.type(klass_node)->singleton()) {
4213 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4214 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4215
4216 int test = C->static_subtype_check(superk, subk);
4217 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4218 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4219 if (t_original->speculative_type() != NULL) {
4220 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4221 }
4222 }
4223 }
4224
4225 bool validated = false;
4226 // Reason_class_check rather than Reason_intrinsic because we
4227 // want to intrinsify even if this traps.
4228 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4229 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4230
4231 if (not_subtype_ctrl != top()) {
4232 PreserveJVMState pjvms(this);
4233 set_control(not_subtype_ctrl);
4234 uncommon_trap(Deoptimization::Reason_class_check,
4235 Deoptimization::Action_make_not_entrant);
4236 assert(stopped(), "Should be stopped");
4237 }
4238 validated = true;
4239 }
4240
4241 if (!stopped()) {
4242 newcopy = new_array(klass_node, length, 0); // no arguments to push
4243
4244 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4245 load_object_klass(original), klass_node);
4246 if (!is_copyOfRange) {
4247 ac->set_copyof(validated);
4248 } else {
4249 ac->set_copyofrange(validated);
4250 }
4251 Node* n = _gvn.transform(ac);
4252 if (n == ac) {
4253 ac->connect_outputs(this);
4254 } else {
4255 assert(validated, "shouldn't transform if all arguments not validated");
4256 set_all_memory(n);
4257 }
4258 }
4259 }
4260 } // original reexecute is set back here
4261
4262 C->set_has_split_ifs(true); // Has chance for split-if optimization
4263 if (!stopped()) {
4264 set_result(newcopy);
4265 }
4266 return true;
4267 }
4268
4269
4270 //----------------------generate_virtual_guard---------------------------
4271 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4272 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4273 RegionNode* slow_region) {
4274 ciMethod* method = callee();
4275 int vtable_index = method->vtable_index();
4276 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4277 "bad index %d", vtable_index);
4278 // Get the Method* out of the appropriate vtable entry.
4279 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4280 vtable_index*vtableEntry::size_in_bytes() +
4281 vtableEntry::method_offset_in_bytes();
4282 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4283 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4284
4285 // Compare the target method with the expected method (e.g., Object.hashCode).
4286 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4287
4288 Node* native_call = makecon(native_call_addr);
4289 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4290 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4291
4292 return generate_slow_guard(test_native, slow_region);
4293 }
4294
4295 //-----------------------generate_method_call----------------------------
4296 // Use generate_method_call to make a slow-call to the real
4297 // method if the fast path fails. An alternative would be to
4298 // use a stub like OptoRuntime::slow_arraycopy_Java.
4299 // This only works for expanding the current library call,
4300 // not another intrinsic. (E.g., don't use this for making an
4301 // arraycopy call inside of the copyOf intrinsic.)
4302 CallJavaNode*
4303 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4304 // When compiling the intrinsic method itself, do not use this technique.
4305 guarantee(callee() != C->method(), "cannot make slow-call to self");
4306
4307 ciMethod* method = callee();
4308 // ensure the JVMS we have will be correct for this call
4309 guarantee(method_id == method->intrinsic_id(), "must match");
4310
4311 const TypeFunc* tf = TypeFunc::make(method);
4312 CallJavaNode* slow_call;
4313 if (is_static) {
4314 assert(!is_virtual, "");
4315 slow_call = new CallStaticJavaNode(C, tf,
4316 SharedRuntime::get_resolve_static_call_stub(), method);
4317 } else if (is_virtual) {
4318 null_check_receiver();
4319 int vtable_index = Method::invalid_vtable_index;
4320 if (UseInlineCaches) {
4321 // Suppress the vtable call
4322 } else {
4323 // hashCode and clone are not a miranda methods,
4324 // so the vtable index is fixed.
4325 // No need to use the linkResolver to get it.
4326 vtable_index = method->vtable_index();
4327 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4328 "bad index %d", vtable_index);
4329 }
4330 slow_call = new CallDynamicJavaNode(tf,
4331 SharedRuntime::get_resolve_virtual_call_stub(),
4332 method, vtable_index);
4333 } else { // neither virtual nor static: opt_virtual
4334 null_check_receiver();
4335 slow_call = new CallStaticJavaNode(C, tf,
4336 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4337 slow_call->set_optimized_virtual(true);
4338 }
4339 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4340 // To be able to issue a direct call (optimized virtual or virtual)
4341 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4342 // about the method being invoked should be attached to the call site to
4343 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4344 slow_call->set_override_symbolic_info(true);
4345 }
4346 set_arguments_for_java_call(slow_call);
4347 set_edges_for_java_call(slow_call);
4348 return slow_call;
4349 }
4350
4351
4352 /**
4353 * Build special case code for calls to hashCode on an object. This call may
4354 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4355 * slightly different code.
4356 */
4357 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4358 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4359 assert(!(is_virtual && is_static), "either virtual, special, or static");
4360
4361 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4362
4363 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4364 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4365 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4366 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4367 Node* obj = NULL;
4368 if (!is_static) {
4369 // Check for hashing null object
4370 obj = null_check_receiver();
4371 if (stopped()) return true; // unconditionally null
4372 result_reg->init_req(_null_path, top());
4373 result_val->init_req(_null_path, top());
4374 } else {
4375 // Do a null check, and return zero if null.
4376 // System.identityHashCode(null) == 0
4377 obj = argument(0);
4378 Node* null_ctl = top();
4379 obj = null_check_oop(obj, &null_ctl);
4380 result_reg->init_req(_null_path, null_ctl);
4381 result_val->init_req(_null_path, _gvn.intcon(0));
4382 }
4383
4384 // Unconditionally null? Then return right away.
4385 if (stopped()) {
4386 set_control( result_reg->in(_null_path));
4387 if (!stopped())
4388 set_result(result_val->in(_null_path));
4389 return true;
4390 }
4391
4392 // We only go to the fast case code if we pass a number of guards. The
4393 // paths which do not pass are accumulated in the slow_region.
4394 RegionNode* slow_region = new RegionNode(1);
4395 record_for_igvn(slow_region);
4396
4397 // If this is a virtual call, we generate a funny guard. We pull out
4398 // the vtable entry corresponding to hashCode() from the target object.
4399 // If the target method which we are calling happens to be the native
4400 // Object hashCode() method, we pass the guard. We do not need this
4401 // guard for non-virtual calls -- the caller is known to be the native
4402 // Object hashCode().
4403 if (is_virtual) {
4404 // After null check, get the object's klass.
4405 Node* obj_klass = load_object_klass(obj);
4406 generate_virtual_guard(obj_klass, slow_region);
4407 }
4408
4409 // Get the header out of the object, use LoadMarkNode when available
4410 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4411 // The control of the load must be NULL. Otherwise, the load can move before
4412 // the null check after castPP removal.
4413 Node* no_ctrl = NULL;
4414 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4415
4416 // Test the header to see if it is unlocked.
4417 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4418 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4419 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4420 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4421 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4422
4423 generate_slow_guard(test_unlocked, slow_region);
4424
4425 // Get the hash value and check to see that it has been properly assigned.
4426 // We depend on hash_mask being at most 32 bits and avoid the use of
4427 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4428 // vm: see markWord.hpp.
4429 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4430 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4431 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4432 // This hack lets the hash bits live anywhere in the mark object now, as long
4433 // as the shift drops the relevant bits into the low 32 bits. Note that
4434 // Java spec says that HashCode is an int so there's no point in capturing
4435 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4436 hshifted_header = ConvX2I(hshifted_header);
4437 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4438
4439 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4440 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4441 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4442
4443 generate_slow_guard(test_assigned, slow_region);
4444
4445 Node* init_mem = reset_memory();
4446 // fill in the rest of the null path:
4447 result_io ->init_req(_null_path, i_o());
4448 result_mem->init_req(_null_path, init_mem);
4449
4450 result_val->init_req(_fast_path, hash_val);
4451 result_reg->init_req(_fast_path, control());
4452 result_io ->init_req(_fast_path, i_o());
4453 result_mem->init_req(_fast_path, init_mem);
4454
4455 // Generate code for the slow case. We make a call to hashCode().
4456 set_control(_gvn.transform(slow_region));
4457 if (!stopped()) {
4458 // No need for PreserveJVMState, because we're using up the present state.
4459 set_all_memory(init_mem);
4460 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4461 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4462 Node* slow_result = set_results_for_java_call(slow_call);
4463 // this->control() comes from set_results_for_java_call
4464 result_reg->init_req(_slow_path, control());
4465 result_val->init_req(_slow_path, slow_result);
4466 result_io ->set_req(_slow_path, i_o());
4467 result_mem ->set_req(_slow_path, reset_memory());
4468 }
4469
4470 // Return the combined state.
4471 set_i_o( _gvn.transform(result_io) );
4472 set_all_memory( _gvn.transform(result_mem));
4473
4474 set_result(result_reg, result_val);
4475 return true;
4476 }
4477
4478 //---------------------------inline_native_getClass----------------------------
4479 // public final native Class<?> java.lang.Object.getClass();
4480 //
4481 // Build special case code for calls to getClass on an object.
4482 bool LibraryCallKit::inline_native_getClass() {
4483 Node* obj = null_check_receiver();
4484 if (stopped()) return true;
4485 set_result(load_mirror_from_klass(load_object_klass(obj)));
4486 return true;
4487 }
4488
4489 //-----------------inline_native_Reflection_getCallerClass---------------------
4490 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4491 //
4492 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4493 //
4494 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4495 // in that it must skip particular security frames and checks for
4496 // caller sensitive methods.
4497 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4498 #ifndef PRODUCT
4499 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4500 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4501 }
4502 #endif
4503
4504 if (!jvms()->has_method()) {
4505 #ifndef PRODUCT
4506 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4507 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4508 }
4509 #endif
4510 return false;
4511 }
4512
4513 // Walk back up the JVM state to find the caller at the required
4514 // depth.
4515 JVMState* caller_jvms = jvms();
4516
4517 // Cf. JVM_GetCallerClass
4518 // NOTE: Start the loop at depth 1 because the current JVM state does
4519 // not include the Reflection.getCallerClass() frame.
4520 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4521 ciMethod* m = caller_jvms->method();
4522 switch (n) {
4523 case 0:
4524 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4525 break;
4526 case 1:
4527 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4528 if (!m->caller_sensitive()) {
4529 #ifndef PRODUCT
4530 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4531 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4532 }
4533 #endif
4534 return false; // bail-out; let JVM_GetCallerClass do the work
4535 }
4536 break;
4537 default:
4538 if (!m->is_ignored_by_security_stack_walk()) {
4539 // We have reached the desired frame; return the holder class.
4540 // Acquire method holder as java.lang.Class and push as constant.
4541 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4542 ciInstance* caller_mirror = caller_klass->java_mirror();
4543 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4544
4545 #ifndef PRODUCT
4546 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4547 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());
4548 tty->print_cr(" JVM state at this point:");
4549 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4550 ciMethod* m = jvms()->of_depth(i)->method();
4551 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4552 }
4553 }
4554 #endif
4555 return true;
4556 }
4557 break;
4558 }
4559 }
4560
4561 #ifndef PRODUCT
4562 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4563 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4564 tty->print_cr(" JVM state at this point:");
4565 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4566 ciMethod* m = jvms()->of_depth(i)->method();
4567 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4568 }
4569 }
4570 #endif
4571
4572 return false; // bail-out; let JVM_GetCallerClass do the work
4573 }
4574
4575 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4576 Node* arg = argument(0);
4577 Node* result = NULL;
4578
4579 switch (id) {
4580 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4581 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4582 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4583 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4584
4585 case vmIntrinsics::_doubleToLongBits: {
4586 // two paths (plus control) merge in a wood
4587 RegionNode *r = new RegionNode(3);
4588 Node *phi = new PhiNode(r, TypeLong::LONG);
4589
4590 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4591 // Build the boolean node
4592 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4593
4594 // Branch either way.
4595 // NaN case is less traveled, which makes all the difference.
4596 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4597 Node *opt_isnan = _gvn.transform(ifisnan);
4598 assert( opt_isnan->is_If(), "Expect an IfNode");
4599 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4600 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4601
4602 set_control(iftrue);
4603
4604 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4605 Node *slow_result = longcon(nan_bits); // return NaN
4606 phi->init_req(1, _gvn.transform( slow_result ));
4607 r->init_req(1, iftrue);
4608
4609 // Else fall through
4610 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4611 set_control(iffalse);
4612
4613 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4614 r->init_req(2, iffalse);
4615
4616 // Post merge
4617 set_control(_gvn.transform(r));
4618 record_for_igvn(r);
4619
4620 C->set_has_split_ifs(true); // Has chance for split-if optimization
4621 result = phi;
4622 assert(result->bottom_type()->isa_long(), "must be");
4623 break;
4624 }
4625
4626 case vmIntrinsics::_floatToIntBits: {
4627 // two paths (plus control) merge in a wood
4628 RegionNode *r = new RegionNode(3);
4629 Node *phi = new PhiNode(r, TypeInt::INT);
4630
4631 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4632 // Build the boolean node
4633 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4634
4635 // Branch either way.
4636 // NaN case is less traveled, which makes all the difference.
4637 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4638 Node *opt_isnan = _gvn.transform(ifisnan);
4639 assert( opt_isnan->is_If(), "Expect an IfNode");
4640 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4641 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4642
4643 set_control(iftrue);
4644
4645 static const jint nan_bits = 0x7fc00000;
4646 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4647 phi->init_req(1, _gvn.transform( slow_result ));
4648 r->init_req(1, iftrue);
4649
4650 // Else fall through
4651 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4652 set_control(iffalse);
4653
4654 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4655 r->init_req(2, iffalse);
4656
4657 // Post merge
4658 set_control(_gvn.transform(r));
4659 record_for_igvn(r);
4660
4661 C->set_has_split_ifs(true); // Has chance for split-if optimization
4662 result = phi;
4663 assert(result->bottom_type()->isa_int(), "must be");
4664 break;
4665 }
4666
4667 default:
4668 fatal_unexpected_iid(id);
4669 break;
4670 }
4671 set_result(_gvn.transform(result));
4672 return true;
4673 }
4674
4675 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4676 Node* arg = argument(0);
4677 Node* result = NULL;
4678
4679 switch (id) {
4680 case vmIntrinsics::_floatIsInfinite:
4681 result = new IsInfiniteFNode(arg);
4682 break;
4683 case vmIntrinsics::_doubleIsInfinite:
4684 result = new IsInfiniteDNode(arg);
4685 break;
4686 default:
4687 fatal_unexpected_iid(id);
4688 break;
4689 }
4690 set_result(_gvn.transform(result));
4691 return true;
4692 }
4693
4694 //----------------------inline_unsafe_copyMemory-------------------------
4695 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4696
4697 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4698 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4699 const Type* base_t = gvn.type(base);
4700
4701 bool in_native = (base_t == TypePtr::NULL_PTR);
4702 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4703 bool is_mixed = !in_heap && !in_native;
4704
4705 if (is_mixed) {
4706 return true; // mixed accesses can touch both on-heap and off-heap memory
4707 }
4708 if (in_heap) {
4709 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4710 if (!is_prim_array) {
4711 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4712 // there's not enough type information available to determine proper memory slice for it.
4713 return true;
4714 }
4715 }
4716 return false;
4717 }
4718
4719 bool LibraryCallKit::inline_unsafe_copyMemory() {
4720 if (callee()->is_static()) return false; // caller must have the capability!
4721 null_check_receiver(); // null-check receiver
4722 if (stopped()) return true;
4723
4724 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4725
4726 Node* src_base = argument(1); // type: oop
4727 Node* src_off = ConvL2X(argument(2)); // type: long
4728 Node* dst_base = argument(4); // type: oop
4729 Node* dst_off = ConvL2X(argument(5)); // type: long
4730 Node* size = ConvL2X(argument(7)); // type: long
4731
4732 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4733 "fieldOffset must be byte-scaled");
4734
4735 Node* src_addr = make_unsafe_address(src_base, src_off);
4736 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4737
4738 Node* thread = _gvn.transform(new ThreadLocalNode());
4739 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4740 BasicType doing_unsafe_access_bt = T_BYTE;
4741 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4742
4743 // update volatile field
4744 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4745
4746 int flags = RC_LEAF | RC_NO_FP;
4747
4748 const TypePtr* dst_type = TypePtr::BOTTOM;
4749
4750 // Adjust memory effects of the runtime call based on input values.
4751 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4752 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4753 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4754
4755 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4756 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4757 flags |= RC_NARROW_MEM; // narrow in memory
4758 }
4759 }
4760
4761 // Call it. Note that the length argument is not scaled.
4762 make_runtime_call(flags,
4763 OptoRuntime::fast_arraycopy_Type(),
4764 StubRoutines::unsafe_arraycopy(),
4765 "unsafe_arraycopy",
4766 dst_type,
4767 src_addr, dst_addr, size XTOP);
4768
4769 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4770
4771 return true;
4772 }
4773
4774 #undef XTOP
4775
4776 //------------------------clone_coping-----------------------------------
4777 // Helper function for inline_native_clone.
4778 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4779 assert(obj_size != NULL, "");
4780 Node* raw_obj = alloc_obj->in(1);
4781 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4782
4783 AllocateNode* alloc = NULL;
4784 if (ReduceBulkZeroing) {
4785 // We will be completely responsible for initializing this object -
4786 // mark Initialize node as complete.
4787 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4788 // The object was just allocated - there should be no any stores!
4789 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4790 // Mark as complete_with_arraycopy so that on AllocateNode
4791 // expansion, we know this AllocateNode is initialized by an array
4792 // copy and a StoreStore barrier exists after the array copy.
4793 alloc->initialization()->set_complete_with_arraycopy();
4794 }
4795
4796 Node* size = _gvn.transform(obj_size);
4797 access_clone(obj, alloc_obj, size, is_array);
4798
4799 // Do not let reads from the cloned object float above the arraycopy.
4800 if (alloc != NULL) {
4801 // Do not let stores that initialize this object be reordered with
4802 // a subsequent store that would make this object accessible by
4803 // other threads.
4804 // Record what AllocateNode this StoreStore protects so that
4805 // escape analysis can go from the MemBarStoreStoreNode to the
4806 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4807 // based on the escape status of the AllocateNode.
4808 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4809 } else {
4810 insert_mem_bar(Op_MemBarCPUOrder);
4811 }
4812 }
4813
4814 //------------------------inline_native_clone----------------------------
4815 // protected native Object java.lang.Object.clone();
4816 //
4817 // Here are the simple edge cases:
4818 // null receiver => normal trap
4819 // virtual and clone was overridden => slow path to out-of-line clone
4820 // not cloneable or finalizer => slow path to out-of-line Object.clone
4821 //
4822 // The general case has two steps, allocation and copying.
4823 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4824 //
4825 // Copying also has two cases, oop arrays and everything else.
4826 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4827 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4828 //
4829 // These steps fold up nicely if and when the cloned object's klass
4830 // can be sharply typed as an object array, a type array, or an instance.
4831 //
4832 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4833 PhiNode* result_val;
4834
4835 // Set the reexecute bit for the interpreter to reexecute
4836 // the bytecode that invokes Object.clone if deoptimization happens.
4837 { PreserveReexecuteState preexecs(this);
4838 jvms()->set_should_reexecute(true);
4839
4840 Node* obj = null_check_receiver();
4841 if (stopped()) return true;
4842
4843 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4844
4845 // If we are going to clone an instance, we need its exact type to
4846 // know the number and types of fields to convert the clone to
4847 // loads/stores. Maybe a speculative type can help us.
4848 if (!obj_type->klass_is_exact() &&
4849 obj_type->speculative_type() != NULL &&
4850 obj_type->speculative_type()->is_instance_klass()) {
4851 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4852 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4853 !spec_ik->has_injected_fields()) {
4854 if (!obj_type->isa_instptr() ||
4855 obj_type->is_instptr()->instance_klass()->has_subklass()) {
4856 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4857 }
4858 }
4859 }
4860
4861 // Conservatively insert a memory barrier on all memory slices.
4862 // Do not let writes into the original float below the clone.
4863 insert_mem_bar(Op_MemBarCPUOrder);
4864
4865 // paths into result_reg:
4866 enum {
4867 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4868 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4869 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4870 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4871 PATH_LIMIT
4872 };
4873 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4874 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4875 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4876 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4877 record_for_igvn(result_reg);
4878
4879 Node* obj_klass = load_object_klass(obj);
4880 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4881 if (array_ctl != NULL) {
4882 // It's an array.
4883 PreserveJVMState pjvms(this);
4884 set_control(array_ctl);
4885 Node* obj_length = load_array_length(obj);
4886 Node* obj_size = NULL;
4887 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4888
4889 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4890 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4891 // If it is an oop array, it requires very special treatment,
4892 // because gc barriers are required when accessing the array.
4893 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4894 if (is_obja != NULL) {
4895 PreserveJVMState pjvms2(this);
4896 set_control(is_obja);
4897 // Generate a direct call to the right arraycopy function(s).
4898 // Clones are always tightly coupled.
4899 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4900 ac->set_clone_oop_array();
4901 Node* n = _gvn.transform(ac);
4902 assert(n == ac, "cannot disappear");
4903 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4904
4905 result_reg->init_req(_objArray_path, control());
4906 result_val->init_req(_objArray_path, alloc_obj);
4907 result_i_o ->set_req(_objArray_path, i_o());
4908 result_mem ->set_req(_objArray_path, reset_memory());
4909 }
4910 }
4911 // Otherwise, there are no barriers to worry about.
4912 // (We can dispense with card marks if we know the allocation
4913 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4914 // causes the non-eden paths to take compensating steps to
4915 // simulate a fresh allocation, so that no further
4916 // card marks are required in compiled code to initialize
4917 // the object.)
4918
4919 if (!stopped()) {
4920 copy_to_clone(obj, alloc_obj, obj_size, true);
4921
4922 // Present the results of the copy.
4923 result_reg->init_req(_array_path, control());
4924 result_val->init_req(_array_path, alloc_obj);
4925 result_i_o ->set_req(_array_path, i_o());
4926 result_mem ->set_req(_array_path, reset_memory());
4927 }
4928 }
4929
4930 // We only go to the instance fast case code if we pass a number of guards.
4931 // The paths which do not pass are accumulated in the slow_region.
4932 RegionNode* slow_region = new RegionNode(1);
4933 record_for_igvn(slow_region);
4934 if (!stopped()) {
4935 // It's an instance (we did array above). Make the slow-path tests.
4936 // If this is a virtual call, we generate a funny guard. We grab
4937 // the vtable entry corresponding to clone() from the target object.
4938 // If the target method which we are calling happens to be the
4939 // Object clone() method, we pass the guard. We do not need this
4940 // guard for non-virtual calls; the caller is known to be the native
4941 // Object clone().
4942 if (is_virtual) {
4943 generate_virtual_guard(obj_klass, slow_region);
4944 }
4945
4946 // The object must be easily cloneable and must not have a finalizer.
4947 // Both of these conditions may be checked in a single test.
4948 // We could optimize the test further, but we don't care.
4949 generate_access_flags_guard(obj_klass,
4950 // Test both conditions:
4951 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4952 // Must be cloneable but not finalizer:
4953 JVM_ACC_IS_CLONEABLE_FAST,
4954 slow_region);
4955 }
4956
4957 if (!stopped()) {
4958 // It's an instance, and it passed the slow-path tests.
4959 PreserveJVMState pjvms(this);
4960 Node* obj_size = NULL;
4961 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4962 // is reexecuted if deoptimization occurs and there could be problems when merging
4963 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4964 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4965
4966 copy_to_clone(obj, alloc_obj, obj_size, false);
4967
4968 // Present the results of the slow call.
4969 result_reg->init_req(_instance_path, control());
4970 result_val->init_req(_instance_path, alloc_obj);
4971 result_i_o ->set_req(_instance_path, i_o());
4972 result_mem ->set_req(_instance_path, reset_memory());
4973 }
4974
4975 // Generate code for the slow case. We make a call to clone().
4976 set_control(_gvn.transform(slow_region));
4977 if (!stopped()) {
4978 PreserveJVMState pjvms(this);
4979 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4980 // We need to deoptimize on exception (see comment above)
4981 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4982 // this->control() comes from set_results_for_java_call
4983 result_reg->init_req(_slow_path, control());
4984 result_val->init_req(_slow_path, slow_result);
4985 result_i_o ->set_req(_slow_path, i_o());
4986 result_mem ->set_req(_slow_path, reset_memory());
4987 }
4988
4989 // Return the combined state.
4990 set_control( _gvn.transform(result_reg));
4991 set_i_o( _gvn.transform(result_i_o));
4992 set_all_memory( _gvn.transform(result_mem));
4993 } // original reexecute is set back here
4994
4995 set_result(_gvn.transform(result_val));
4996 return true;
4997 }
4998
4999 // If we have a tightly coupled allocation, the arraycopy may take care
5000 // of the array initialization. If one of the guards we insert between
5001 // the allocation and the arraycopy causes a deoptimization, an
5002 // uninitialized array will escape the compiled method. To prevent that
5003 // we set the JVM state for uncommon traps between the allocation and
5004 // the arraycopy to the state before the allocation so, in case of
5005 // deoptimization, we'll reexecute the allocation and the
5006 // initialization.
5007 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5008 if (alloc != NULL) {
5009 ciMethod* trap_method = alloc->jvms()->method();
5010 int trap_bci = alloc->jvms()->bci();
5011
5012 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5013 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5014 // Make sure there's no store between the allocation and the
5015 // arraycopy otherwise visible side effects could be rexecuted
5016 // in case of deoptimization and cause incorrect execution.
5017 bool no_interfering_store = true;
5018 Node* mem = alloc->in(TypeFunc::Memory);
5019 if (mem->is_MergeMem()) {
5020 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5021 Node* n = mms.memory();
5022 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5023 assert(n->is_Store(), "what else?");
5024 no_interfering_store = false;
5025 break;
5026 }
5027 }
5028 } else {
5029 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5030 Node* n = mms.memory();
5031 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5032 assert(n->is_Store(), "what else?");
5033 no_interfering_store = false;
5034 break;
5035 }
5036 }
5037 }
5038
5039 if (no_interfering_store) {
5040 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5041 uint size = alloc->req();
5042 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5043 old_jvms->set_map(sfpt);
5044 for (uint i = 0; i < size; i++) {
5045 sfpt->init_req(i, alloc->in(i));
5046 }
5047 // re-push array length for deoptimization
5048 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5049 old_jvms->set_sp(old_jvms->sp()+1);
5050 old_jvms->set_monoff(old_jvms->monoff()+1);
5051 old_jvms->set_scloff(old_jvms->scloff()+1);
5052 old_jvms->set_endoff(old_jvms->endoff()+1);
5053 old_jvms->set_should_reexecute(true);
5054
5055 sfpt->set_i_o(map()->i_o());
5056 sfpt->set_memory(map()->memory());
5057 sfpt->set_control(map()->control());
5058
5059 JVMState* saved_jvms = jvms();
5060 saved_reexecute_sp = _reexecute_sp;
5061
5062 set_jvms(sfpt->jvms());
5063 _reexecute_sp = jvms()->sp();
5064
5065 return saved_jvms;
5066 }
5067 }
5068 }
5069 return NULL;
5070 }
5071
5072 // In case of a deoptimization, we restart execution at the
5073 // allocation, allocating a new array. We would leave an uninitialized
5074 // array in the heap that GCs wouldn't expect. Move the allocation
5075 // after the traps so we don't allocate the array if we
5076 // deoptimize. This is possible because tightly_coupled_allocation()
5077 // guarantees there's no observer of the allocated array at this point
5078 // and the control flow is simple enough.
5079 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
5080 int saved_reexecute_sp, uint new_idx) {
5081 if (saved_jvms != NULL && !stopped()) {
5082 assert(alloc != NULL, "only with a tightly coupled allocation");
5083 // restore JVM state to the state at the arraycopy
5084 saved_jvms->map()->set_control(map()->control());
5085 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
5086 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
5087 // If we've improved the types of some nodes (null check) while
5088 // emitting the guards, propagate them to the current state
5089 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
5090 set_jvms(saved_jvms);
5091 _reexecute_sp = saved_reexecute_sp;
5092
5093 // Remove the allocation from above the guards
5094 CallProjections callprojs;
5095 alloc->extract_projections(&callprojs, true);
5096 InitializeNode* init = alloc->initialization();
5097 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5098 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5099 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5100
5101 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5102 // the allocation (i.e. is only valid if the allocation succeeds):
5103 // 1) replace CastIINode with AllocateArrayNode's length here
5104 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5105 //
5106 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5107 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5108 Node* init_control = init->proj_out(TypeFunc::Control);
5109 Node* alloc_length = alloc->Ideal_length();
5110 #ifdef ASSERT
5111 Node* prev_cast = NULL;
5112 #endif
5113 for (uint i = 0; i < init_control->outcnt(); i++) {
5114 Node* init_out = init_control->raw_out(i);
5115 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5116 #ifdef ASSERT
5117 if (prev_cast == NULL) {
5118 prev_cast = init_out;
5119 } else {
5120 if (prev_cast->cmp(*init_out) == false) {
5121 prev_cast->dump();
5122 init_out->dump();
5123 assert(false, "not equal CastIINode");
5124 }
5125 }
5126 #endif
5127 C->gvn_replace_by(init_out, alloc_length);
5128 }
5129 }
5130 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5131
5132 // move the allocation here (after the guards)
5133 _gvn.hash_delete(alloc);
5134 alloc->set_req(TypeFunc::Control, control());
5135 alloc->set_req(TypeFunc::I_O, i_o());
5136 Node *mem = reset_memory();
5137 set_all_memory(mem);
5138 alloc->set_req(TypeFunc::Memory, mem);
5139 set_control(init->proj_out_or_null(TypeFunc::Control));
5140 set_i_o(callprojs.fallthrough_ioproj);
5141
5142 // Update memory as done in GraphKit::set_output_for_allocation()
5143 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5144 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5145 if (ary_type->isa_aryptr() && length_type != NULL) {
5146 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5147 }
5148 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5149 int elemidx = C->get_alias_index(telemref);
5150 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5151 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5152
5153 Node* allocx = _gvn.transform(alloc);
5154 assert(allocx == alloc, "where has the allocation gone?");
5155 assert(dest->is_CheckCastPP(), "not an allocation result?");
5156
5157 _gvn.hash_delete(dest);
5158 dest->set_req(0, control());
5159 Node* destx = _gvn.transform(dest);
5160 assert(destx == dest, "where has the allocation result gone?");
5161
5162 array_ideal_length(alloc, ary_type, true);
5163 }
5164 }
5165
5166
5167 //------------------------------inline_arraycopy-----------------------
5168 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5169 // Object dest, int destPos,
5170 // int length);
5171 bool LibraryCallKit::inline_arraycopy() {
5172 // Get the arguments.
5173 Node* src = argument(0); // type: oop
5174 Node* src_offset = argument(1); // type: int
5175 Node* dest = argument(2); // type: oop
5176 Node* dest_offset = argument(3); // type: int
5177 Node* length = argument(4); // type: int
5178
5179 uint new_idx = C->unique();
5180
5181 // Check for allocation before we add nodes that would confuse
5182 // tightly_coupled_allocation()
5183 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5184
5185 int saved_reexecute_sp = -1;
5186 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5187 // See arraycopy_restore_alloc_state() comment
5188 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5189 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5190 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5191 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5192
5193 // The following tests must be performed
5194 // (1) src and dest are arrays.
5195 // (2) src and dest arrays must have elements of the same BasicType
5196 // (3) src and dest must not be null.
5197 // (4) src_offset must not be negative.
5198 // (5) dest_offset must not be negative.
5199 // (6) length must not be negative.
5200 // (7) src_offset + length must not exceed length of src.
5201 // (8) dest_offset + length must not exceed length of dest.
5202 // (9) each element of an oop array must be assignable
5203
5204 // (3) src and dest must not be null.
5205 // always do this here because we need the JVM state for uncommon traps
5206 Node* null_ctl = top();
5207 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5208 assert(null_ctl->is_top(), "no null control here");
5209 dest = null_check(dest, T_ARRAY);
5210
5211 if (!can_emit_guards) {
5212 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5213 // guards but the arraycopy node could still take advantage of a
5214 // tightly allocated allocation. tightly_coupled_allocation() is
5215 // called again to make sure it takes the null check above into
5216 // account: the null check is mandatory and if it caused an
5217 // uncommon trap to be emitted then the allocation can't be
5218 // considered tightly coupled in this context.
5219 alloc = tightly_coupled_allocation(dest);
5220 }
5221
5222 bool validated = false;
5223
5224 const Type* src_type = _gvn.type(src);
5225 const Type* dest_type = _gvn.type(dest);
5226 const TypeAryPtr* top_src = src_type->isa_aryptr();
5227 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5228
5229 // Do we have the type of src?
5230 bool has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5231 // Do we have the type of dest?
5232 bool has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5233 // Is the type for src from speculation?
5234 bool src_spec = false;
5235 // Is the type for dest from speculation?
5236 bool dest_spec = false;
5237
5238 if ((!has_src || !has_dest) && can_emit_guards) {
5239 // We don't have sufficient type information, let's see if
5240 // speculative types can help. We need to have types for both src
5241 // and dest so that it pays off.
5242
5243 // Do we already have or could we have type information for src
5244 bool could_have_src = has_src;
5245 // Do we already have or could we have type information for dest
5246 bool could_have_dest = has_dest;
5247
5248 ciKlass* src_k = NULL;
5249 if (!has_src) {
5250 src_k = src_type->speculative_type_not_null();
5251 if (src_k != NULL && src_k->is_array_klass()) {
5252 could_have_src = true;
5253 }
5254 }
5255
5256 ciKlass* dest_k = NULL;
5257 if (!has_dest) {
5258 dest_k = dest_type->speculative_type_not_null();
5259 if (dest_k != NULL && dest_k->is_array_klass()) {
5260 could_have_dest = true;
5261 }
5262 }
5263
5264 if (could_have_src && could_have_dest) {
5265 // This is going to pay off so emit the required guards
5266 if (!has_src) {
5267 src = maybe_cast_profiled_obj(src, src_k, true);
5268 src_type = _gvn.type(src);
5269 top_src = src_type->isa_aryptr();
5270 has_src = (top_src != NULL && top_src->elem() != Type::BOTTOM);
5271 src_spec = true;
5272 }
5273 if (!has_dest) {
5274 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5275 dest_type = _gvn.type(dest);
5276 top_dest = dest_type->isa_aryptr();
5277 has_dest = (top_dest != NULL && top_dest->elem() != Type::BOTTOM);
5278 dest_spec = true;
5279 }
5280 }
5281 }
5282
5283 if (has_src && has_dest && can_emit_guards) {
5284 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5285 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5286 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5287 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5288
5289 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5290 // If both arrays are object arrays then having the exact types
5291 // for both will remove the need for a subtype check at runtime
5292 // before the call and may make it possible to pick a faster copy
5293 // routine (without a subtype check on every element)
5294 // Do we have the exact type of src?
5295 bool could_have_src = src_spec;
5296 // Do we have the exact type of dest?
5297 bool could_have_dest = dest_spec;
5298 ciKlass* src_k = NULL;
5299 ciKlass* dest_k = NULL;
5300 if (!src_spec) {
5301 src_k = src_type->speculative_type_not_null();
5302 if (src_k != NULL && src_k->is_array_klass()) {
5303 could_have_src = true;
5304 }
5305 }
5306 if (!dest_spec) {
5307 dest_k = dest_type->speculative_type_not_null();
5308 if (dest_k != NULL && dest_k->is_array_klass()) {
5309 could_have_dest = true;
5310 }
5311 }
5312 if (could_have_src && could_have_dest) {
5313 // If we can have both exact types, emit the missing guards
5314 if (could_have_src && !src_spec) {
5315 src = maybe_cast_profiled_obj(src, src_k, true);
5316 }
5317 if (could_have_dest && !dest_spec) {
5318 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5319 }
5320 }
5321 }
5322 }
5323
5324 ciMethod* trap_method = method();
5325 int trap_bci = bci();
5326 if (saved_jvms != NULL) {
5327 trap_method = alloc->jvms()->method();
5328 trap_bci = alloc->jvms()->bci();
5329 }
5330
5331 bool negative_length_guard_generated = false;
5332
5333 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5334 can_emit_guards &&
5335 !src->is_top() && !dest->is_top()) {
5336 // validate arguments: enables transformation the ArrayCopyNode
5337 validated = true;
5338
5339 RegionNode* slow_region = new RegionNode(1);
5340 record_for_igvn(slow_region);
5341
5342 // (1) src and dest are arrays.
5343 generate_non_array_guard(load_object_klass(src), slow_region);
5344 generate_non_array_guard(load_object_klass(dest), slow_region);
5345
5346 // (2) src and dest arrays must have elements of the same BasicType
5347 // done at macro expansion or at Ideal transformation time
5348
5349 // (4) src_offset must not be negative.
5350 generate_negative_guard(src_offset, slow_region);
5351
5352 // (5) dest_offset must not be negative.
5353 generate_negative_guard(dest_offset, slow_region);
5354
5355 // (7) src_offset + length must not exceed length of src.
5356 generate_limit_guard(src_offset, length,
5357 load_array_length(src),
5358 slow_region);
5359
5360 // (8) dest_offset + length must not exceed length of dest.
5361 generate_limit_guard(dest_offset, length,
5362 load_array_length(dest),
5363 slow_region);
5364
5365 // (6) length must not be negative.
5366 // This is also checked in generate_arraycopy() during macro expansion, but
5367 // we also have to check it here for the case where the ArrayCopyNode will
5368 // be eliminated by Escape Analysis.
5369 if (EliminateAllocations) {
5370 generate_negative_guard(length, slow_region);
5371 negative_length_guard_generated = true;
5372 }
5373
5374 // (9) each element of an oop array must be assignable
5375 Node* dest_klass = load_object_klass(dest);
5376 if (src != dest) {
5377 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5378
5379 if (not_subtype_ctrl != top()) {
5380 PreserveJVMState pjvms(this);
5381 set_control(not_subtype_ctrl);
5382 uncommon_trap(Deoptimization::Reason_intrinsic,
5383 Deoptimization::Action_make_not_entrant);
5384 assert(stopped(), "Should be stopped");
5385 }
5386 }
5387 {
5388 PreserveJVMState pjvms(this);
5389 set_control(_gvn.transform(slow_region));
5390 uncommon_trap(Deoptimization::Reason_intrinsic,
5391 Deoptimization::Action_make_not_entrant);
5392 assert(stopped(), "Should be stopped");
5393 }
5394
5395 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5396 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5397 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5398 }
5399
5400 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5401
5402 if (stopped()) {
5403 return true;
5404 }
5405
5406 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5407 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5408 // so the compiler has a chance to eliminate them: during macro expansion,
5409 // we have to set their control (CastPP nodes are eliminated).
5410 load_object_klass(src), load_object_klass(dest),
5411 load_array_length(src), load_array_length(dest));
5412
5413 ac->set_arraycopy(validated);
5414
5415 Node* n = _gvn.transform(ac);
5416 if (n == ac) {
5417 ac->connect_outputs(this);
5418 } else {
5419 assert(validated, "shouldn't transform if all arguments not validated");
5420 set_all_memory(n);
5421 }
5422 clear_upper_avx();
5423
5424
5425 return true;
5426 }
5427
5428
5429 // Helper function which determines if an arraycopy immediately follows
5430 // an allocation, with no intervening tests or other escapes for the object.
5431 AllocateArrayNode*
5432 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5433 if (stopped()) return NULL; // no fast path
5434 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5435
5436 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5437 if (alloc == NULL) return NULL;
5438
5439 Node* rawmem = memory(Compile::AliasIdxRaw);
5440 // Is the allocation's memory state untouched?
5441 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5442 // Bail out if there have been raw-memory effects since the allocation.
5443 // (Example: There might have been a call or safepoint.)
5444 return NULL;
5445 }
5446 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5447 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5448 return NULL;
5449 }
5450
5451 // There must be no unexpected observers of this allocation.
5452 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5453 Node* obs = ptr->fast_out(i);
5454 if (obs != this->map()) {
5455 return NULL;
5456 }
5457 }
5458
5459 // This arraycopy must unconditionally follow the allocation of the ptr.
5460 Node* alloc_ctl = ptr->in(0);
5461 Node* ctl = control();
5462 while (ctl != alloc_ctl) {
5463 // There may be guards which feed into the slow_region.
5464 // Any other control flow means that we might not get a chance
5465 // to finish initializing the allocated object.
5466 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5467 IfNode* iff = ctl->in(0)->as_If();
5468 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5469 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5470 // One more try: Various low-level checks bottom out in
5471 // uncommon traps. If the debug-info of the trap omits
5472 // any reference to the allocation, as we've already
5473 // observed, then there can be no objection to the trap.
5474 bool found_trap = false;
5475 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5476 Node* obs = not_ctl->fast_out(j);
5477 if (obs->in(0) == not_ctl && obs->is_Call() &&
5478 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5479 found_trap = true; break;
5480 }
5481 }
5482 if (found_trap) {
5483 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5484 continue;
5485 }
5486 }
5487 return NULL;
5488 }
5489
5490 // If we get this far, we have an allocation which immediately
5491 // precedes the arraycopy, and we can take over zeroing the new object.
5492 // The arraycopy will finish the initialization, and provide
5493 // a new control state to which we will anchor the destination pointer.
5494
5495 return alloc;
5496 }
5497
5498 //-------------inline_encodeISOArray-----------------------------------
5499 // encode char[] to byte[] in ISO_8859_1 or ASCII
5500 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5501 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5502 // no receiver since it is static method
5503 Node *src = argument(0);
5504 Node *src_offset = argument(1);
5505 Node *dst = argument(2);
5506 Node *dst_offset = argument(3);
5507 Node *length = argument(4);
5508
5509 src = must_be_not_null(src, true);
5510 dst = must_be_not_null(dst, true);
5511
5512 const Type* src_type = src->Value(&_gvn);
5513 const Type* dst_type = dst->Value(&_gvn);
5514 const TypeAryPtr* top_src = src_type->isa_aryptr();
5515 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5516 if (top_src == NULL || top_src->elem() == Type::BOTTOM ||
5517 top_dest == NULL || top_dest->elem() == Type::BOTTOM) {
5518 // failed array check
5519 return false;
5520 }
5521
5522 // Figure out the size and type of the elements we will be copying.
5523 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
5524 BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
5525 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5526 return false;
5527 }
5528
5529 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5530 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5531 // 'src_start' points to src array + scaled offset
5532 // 'dst_start' points to dst array + scaled offset
5533
5534 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5535 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5536 enc = _gvn.transform(enc);
5537 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5538 set_memory(res_mem, mtype);
5539 set_result(enc);
5540 clear_upper_avx();
5541
5542 return true;
5543 }
5544
5545 //-------------inline_multiplyToLen-----------------------------------
5546 bool LibraryCallKit::inline_multiplyToLen() {
5547 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5548
5549 address stubAddr = StubRoutines::multiplyToLen();
5550 if (stubAddr == NULL) {
5551 return false; // Intrinsic's stub is not implemented on this platform
5552 }
5553 const char* stubName = "multiplyToLen";
5554
5555 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5556
5557 // no receiver because it is a static method
5558 Node* x = argument(0);
5559 Node* xlen = argument(1);
5560 Node* y = argument(2);
5561 Node* ylen = argument(3);
5562 Node* z = argument(4);
5563
5564 x = must_be_not_null(x, true);
5565 y = must_be_not_null(y, true);
5566
5567 const Type* x_type = x->Value(&_gvn);
5568 const Type* y_type = y->Value(&_gvn);
5569 const TypeAryPtr* top_x = x_type->isa_aryptr();
5570 const TypeAryPtr* top_y = y_type->isa_aryptr();
5571 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5572 top_y == NULL || top_y->elem() == Type::BOTTOM) {
5573 // failed array check
5574 return false;
5575 }
5576
5577 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5578 BasicType y_elem = y_type->isa_aryptr()->elem()->array_element_basic_type();
5579 if (x_elem != T_INT || y_elem != T_INT) {
5580 return false;
5581 }
5582
5583 // Set the original stack and the reexecute bit for the interpreter to reexecute
5584 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5585 // on the return from z array allocation in runtime.
5586 { PreserveReexecuteState preexecs(this);
5587 jvms()->set_should_reexecute(true);
5588
5589 Node* x_start = array_element_address(x, intcon(0), x_elem);
5590 Node* y_start = array_element_address(y, intcon(0), y_elem);
5591 // 'x_start' points to x array + scaled xlen
5592 // 'y_start' points to y array + scaled ylen
5593
5594 // Allocate the result array
5595 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5596 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5597 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5598
5599 IdealKit ideal(this);
5600
5601 #define __ ideal.
5602 Node* one = __ ConI(1);
5603 Node* zero = __ ConI(0);
5604 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5605 __ set(need_alloc, zero);
5606 __ set(z_alloc, z);
5607 __ if_then(z, BoolTest::eq, null()); {
5608 __ increment (need_alloc, one);
5609 } __ else_(); {
5610 // Update graphKit memory and control from IdealKit.
5611 sync_kit(ideal);
5612 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5613 cast->init_req(0, control());
5614 _gvn.set_type(cast, cast->bottom_type());
5615 C->record_for_igvn(cast);
5616
5617 Node* zlen_arg = load_array_length(cast);
5618 // Update IdealKit memory and control from graphKit.
5619 __ sync_kit(this);
5620 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5621 __ increment (need_alloc, one);
5622 } __ end_if();
5623 } __ end_if();
5624
5625 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5626 // Update graphKit memory and control from IdealKit.
5627 sync_kit(ideal);
5628 Node * narr = new_array(klass_node, zlen, 1);
5629 // Update IdealKit memory and control from graphKit.
5630 __ sync_kit(this);
5631 __ set(z_alloc, narr);
5632 } __ end_if();
5633
5634 sync_kit(ideal);
5635 z = __ value(z_alloc);
5636 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5637 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5638 // Final sync IdealKit and GraphKit.
5639 final_sync(ideal);
5640 #undef __
5641
5642 Node* z_start = array_element_address(z, intcon(0), T_INT);
5643
5644 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5645 OptoRuntime::multiplyToLen_Type(),
5646 stubAddr, stubName, TypePtr::BOTTOM,
5647 x_start, xlen, y_start, ylen, z_start, zlen);
5648 } // original reexecute is set back here
5649
5650 C->set_has_split_ifs(true); // Has chance for split-if optimization
5651 set_result(z);
5652 return true;
5653 }
5654
5655 //-------------inline_squareToLen------------------------------------
5656 bool LibraryCallKit::inline_squareToLen() {
5657 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5658
5659 address stubAddr = StubRoutines::squareToLen();
5660 if (stubAddr == NULL) {
5661 return false; // Intrinsic's stub is not implemented on this platform
5662 }
5663 const char* stubName = "squareToLen";
5664
5665 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5666
5667 Node* x = argument(0);
5668 Node* len = argument(1);
5669 Node* z = argument(2);
5670 Node* zlen = argument(3);
5671
5672 x = must_be_not_null(x, true);
5673 z = must_be_not_null(z, true);
5674
5675 const Type* x_type = x->Value(&_gvn);
5676 const Type* z_type = z->Value(&_gvn);
5677 const TypeAryPtr* top_x = x_type->isa_aryptr();
5678 const TypeAryPtr* top_z = z_type->isa_aryptr();
5679 if (top_x == NULL || top_x->elem() == Type::BOTTOM ||
5680 top_z == NULL || top_z->elem() == Type::BOTTOM) {
5681 // failed array check
5682 return false;
5683 }
5684
5685 BasicType x_elem = x_type->isa_aryptr()->elem()->array_element_basic_type();
5686 BasicType z_elem = z_type->isa_aryptr()->elem()->array_element_basic_type();
5687 if (x_elem != T_INT || z_elem != T_INT) {
5688 return false;
5689 }
5690
5691
5692 Node* x_start = array_element_address(x, intcon(0), x_elem);
5693 Node* z_start = array_element_address(z, intcon(0), z_elem);
5694
5695 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5696 OptoRuntime::squareToLen_Type(),
5697 stubAddr, stubName, TypePtr::BOTTOM,
5698 x_start, len, z_start, zlen);
5699
5700 set_result(z);
5701 return true;
5702 }
5703
5704 //-------------inline_mulAdd------------------------------------------
5705 bool LibraryCallKit::inline_mulAdd() {
5706 assert(UseMulAddIntrinsic, "not implemented on this platform");
5707
5708 address stubAddr = StubRoutines::mulAdd();
5709 if (stubAddr == NULL) {
5710 return false; // Intrinsic's stub is not implemented on this platform
5711 }
5712 const char* stubName = "mulAdd";
5713
5714 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5715
5716 Node* out = argument(0);
5717 Node* in = argument(1);
5718 Node* offset = argument(2);
5719 Node* len = argument(3);
5720 Node* k = argument(4);
5721
5722 out = must_be_not_null(out, true);
5723
5724 const Type* out_type = out->Value(&_gvn);
5725 const Type* in_type = in->Value(&_gvn);
5726 const TypeAryPtr* top_out = out_type->isa_aryptr();
5727 const TypeAryPtr* top_in = in_type->isa_aryptr();
5728 if (top_out == NULL || top_out->elem() == Type::BOTTOM ||
5729 top_in == NULL || top_in->elem() == Type::BOTTOM) {
5730 // failed array check
5731 return false;
5732 }
5733
5734 BasicType out_elem = out_type->isa_aryptr()->elem()->array_element_basic_type();
5735 BasicType in_elem = in_type->isa_aryptr()->elem()->array_element_basic_type();
5736 if (out_elem != T_INT || in_elem != T_INT) {
5737 return false;
5738 }
5739
5740 Node* outlen = load_array_length(out);
5741 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5742 Node* out_start = array_element_address(out, intcon(0), out_elem);
5743 Node* in_start = array_element_address(in, intcon(0), in_elem);
5744
5745 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5746 OptoRuntime::mulAdd_Type(),
5747 stubAddr, stubName, TypePtr::BOTTOM,
5748 out_start,in_start, new_offset, len, k);
5749 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5750 set_result(result);
5751 return true;
5752 }
5753
5754 //-------------inline_montgomeryMultiply-----------------------------------
5755 bool LibraryCallKit::inline_montgomeryMultiply() {
5756 address stubAddr = StubRoutines::montgomeryMultiply();
5757 if (stubAddr == NULL) {
5758 return false; // Intrinsic's stub is not implemented on this platform
5759 }
5760
5761 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5762 const char* stubName = "montgomery_multiply";
5763
5764 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5765
5766 Node* a = argument(0);
5767 Node* b = argument(1);
5768 Node* n = argument(2);
5769 Node* len = argument(3);
5770 Node* inv = argument(4);
5771 Node* m = argument(6);
5772
5773 const Type* a_type = a->Value(&_gvn);
5774 const TypeAryPtr* top_a = a_type->isa_aryptr();
5775 const Type* b_type = b->Value(&_gvn);
5776 const TypeAryPtr* top_b = b_type->isa_aryptr();
5777 const Type* n_type = a->Value(&_gvn);
5778 const TypeAryPtr* top_n = n_type->isa_aryptr();
5779 const Type* m_type = a->Value(&_gvn);
5780 const TypeAryPtr* top_m = m_type->isa_aryptr();
5781 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5782 top_b == NULL || top_b->elem() == Type::BOTTOM ||
5783 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5784 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5785 // failed array check
5786 return false;
5787 }
5788
5789 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5790 BasicType b_elem = b_type->isa_aryptr()->elem()->array_element_basic_type();
5791 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5792 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5793 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5794 return false;
5795 }
5796
5797 // Make the call
5798 {
5799 Node* a_start = array_element_address(a, intcon(0), a_elem);
5800 Node* b_start = array_element_address(b, intcon(0), b_elem);
5801 Node* n_start = array_element_address(n, intcon(0), n_elem);
5802 Node* m_start = array_element_address(m, intcon(0), m_elem);
5803
5804 Node* call = make_runtime_call(RC_LEAF,
5805 OptoRuntime::montgomeryMultiply_Type(),
5806 stubAddr, stubName, TypePtr::BOTTOM,
5807 a_start, b_start, n_start, len, inv, top(),
5808 m_start);
5809 set_result(m);
5810 }
5811
5812 return true;
5813 }
5814
5815 bool LibraryCallKit::inline_montgomerySquare() {
5816 address stubAddr = StubRoutines::montgomerySquare();
5817 if (stubAddr == NULL) {
5818 return false; // Intrinsic's stub is not implemented on this platform
5819 }
5820
5821 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5822 const char* stubName = "montgomery_square";
5823
5824 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5825
5826 Node* a = argument(0);
5827 Node* n = argument(1);
5828 Node* len = argument(2);
5829 Node* inv = argument(3);
5830 Node* m = argument(5);
5831
5832 const Type* a_type = a->Value(&_gvn);
5833 const TypeAryPtr* top_a = a_type->isa_aryptr();
5834 const Type* n_type = a->Value(&_gvn);
5835 const TypeAryPtr* top_n = n_type->isa_aryptr();
5836 const Type* m_type = a->Value(&_gvn);
5837 const TypeAryPtr* top_m = m_type->isa_aryptr();
5838 if (top_a == NULL || top_a->elem() == Type::BOTTOM ||
5839 top_n == NULL || top_n->elem() == Type::BOTTOM ||
5840 top_m == NULL || top_m->elem() == Type::BOTTOM) {
5841 // failed array check
5842 return false;
5843 }
5844
5845 BasicType a_elem = a_type->isa_aryptr()->elem()->array_element_basic_type();
5846 BasicType n_elem = n_type->isa_aryptr()->elem()->array_element_basic_type();
5847 BasicType m_elem = m_type->isa_aryptr()->elem()->array_element_basic_type();
5848 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5849 return false;
5850 }
5851
5852 // Make the call
5853 {
5854 Node* a_start = array_element_address(a, intcon(0), a_elem);
5855 Node* n_start = array_element_address(n, intcon(0), n_elem);
5856 Node* m_start = array_element_address(m, intcon(0), m_elem);
5857
5858 Node* call = make_runtime_call(RC_LEAF,
5859 OptoRuntime::montgomerySquare_Type(),
5860 stubAddr, stubName, TypePtr::BOTTOM,
5861 a_start, n_start, len, inv, top(),
5862 m_start);
5863 set_result(m);
5864 }
5865
5866 return true;
5867 }
5868
5869 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5870 address stubAddr = NULL;
5871 const char* stubName = NULL;
5872
5873 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5874 if (stubAddr == NULL) {
5875 return false; // Intrinsic's stub is not implemented on this platform
5876 }
5877
5878 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5879
5880 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5881
5882 Node* newArr = argument(0);
5883 Node* oldArr = argument(1);
5884 Node* newIdx = argument(2);
5885 Node* shiftCount = argument(3);
5886 Node* numIter = argument(4);
5887
5888 const Type* newArr_type = newArr->Value(&_gvn);
5889 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5890 const Type* oldArr_type = oldArr->Value(&_gvn);
5891 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5892 if (top_newArr == NULL || top_newArr->elem() == Type::BOTTOM || top_oldArr == NULL
5893 || top_oldArr->elem() == Type::BOTTOM) {
5894 return false;
5895 }
5896
5897 BasicType newArr_elem = newArr_type->isa_aryptr()->elem()->array_element_basic_type();
5898 BasicType oldArr_elem = oldArr_type->isa_aryptr()->elem()->array_element_basic_type();
5899 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5900 return false;
5901 }
5902
5903 // Make the call
5904 {
5905 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5906 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5907
5908 Node* call = make_runtime_call(RC_LEAF,
5909 OptoRuntime::bigIntegerShift_Type(),
5910 stubAddr,
5911 stubName,
5912 TypePtr::BOTTOM,
5913 newArr_start,
5914 oldArr_start,
5915 newIdx,
5916 shiftCount,
5917 numIter);
5918 }
5919
5920 return true;
5921 }
5922
5923 //-------------inline_vectorizedMismatch------------------------------
5924 bool LibraryCallKit::inline_vectorizedMismatch() {
5925 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5926
5927 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5928 Node* obja = argument(0); // Object
5929 Node* aoffset = argument(1); // long
5930 Node* objb = argument(3); // Object
5931 Node* boffset = argument(4); // long
5932 Node* length = argument(6); // int
5933 Node* scale = argument(7); // int
5934
5935 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5936 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5937 if (obja_t == NULL || obja_t->elem() == Type::BOTTOM ||
5938 objb_t == NULL || objb_t->elem() == Type::BOTTOM ||
5939 scale == top()) {
5940 return false; // failed input validation
5941 }
5942
5943 Node* obja_adr = make_unsafe_address(obja, aoffset);
5944 Node* objb_adr = make_unsafe_address(objb, boffset);
5945
5946 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5947 //
5948 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5949 // if (length <= inline_limit) {
5950 // inline_path:
5951 // vmask = VectorMaskGen length
5952 // vload1 = LoadVectorMasked obja, vmask
5953 // vload2 = LoadVectorMasked objb, vmask
5954 // result1 = VectorCmpMasked vload1, vload2, vmask
5955 // } else {
5956 // call_stub_path:
5957 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5958 // }
5959 // exit_block:
5960 // return Phi(result1, result2);
5961 //
5962 enum { inline_path = 1, // input is small enough to process it all at once
5963 stub_path = 2, // input is too large; call into the VM
5964 PATH_LIMIT = 3
5965 };
5966
5967 Node* exit_block = new RegionNode(PATH_LIMIT);
5968 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5969 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5970
5971 Node* call_stub_path = control();
5972
5973 BasicType elem_bt = T_ILLEGAL;
5974
5975 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5976 if (scale_t->is_con()) {
5977 switch (scale_t->get_con()) {
5978 case 0: elem_bt = T_BYTE; break;
5979 case 1: elem_bt = T_SHORT; break;
5980 case 2: elem_bt = T_INT; break;
5981 case 3: elem_bt = T_LONG; break;
5982
5983 default: elem_bt = T_ILLEGAL; break; // not supported
5984 }
5985 }
5986
5987 int inline_limit = 0;
5988 bool do_partial_inline = false;
5989
5990 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5991 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5992 do_partial_inline = inline_limit >= 16;
5993 }
5994
5995 if (do_partial_inline) {
5996 assert(elem_bt != T_ILLEGAL, "sanity");
5997
5998 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5999 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6000 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6001
6002 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6003 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6004 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6005
6006 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
6007
6008 if (!stopped()) {
6009 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6010
6011 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6012 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6013 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6014 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6015
6016 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6017 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6018 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6019 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6020
6021 exit_block->init_req(inline_path, control());
6022 memory_phi->init_req(inline_path, map()->memory());
6023 result_phi->init_req(inline_path, result);
6024
6025 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6026 clear_upper_avx();
6027 }
6028 }
6029 }
6030
6031 if (call_stub_path != NULL) {
6032 set_control(call_stub_path);
6033
6034 Node* call = make_runtime_call(RC_LEAF,
6035 OptoRuntime::vectorizedMismatch_Type(),
6036 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6037 obja_adr, objb_adr, length, scale);
6038
6039 exit_block->init_req(stub_path, control());
6040 memory_phi->init_req(stub_path, map()->memory());
6041 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6042 }
6043
6044 exit_block = _gvn.transform(exit_block);
6045 memory_phi = _gvn.transform(memory_phi);
6046 result_phi = _gvn.transform(result_phi);
6047
6048 set_control(exit_block);
6049 set_all_memory(memory_phi);
6050 set_result(result_phi);
6051
6052 return true;
6053 }
6054
6055 /**
6056 * Calculate CRC32 for byte.
6057 * int java.util.zip.CRC32.update(int crc, int b)
6058 */
6059 bool LibraryCallKit::inline_updateCRC32() {
6060 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6061 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6062 // no receiver since it is static method
6063 Node* crc = argument(0); // type: int
6064 Node* b = argument(1); // type: int
6065
6066 /*
6067 * int c = ~ crc;
6068 * b = timesXtoThe32[(b ^ c) & 0xFF];
6069 * b = b ^ (c >>> 8);
6070 * crc = ~b;
6071 */
6072
6073 Node* M1 = intcon(-1);
6074 crc = _gvn.transform(new XorINode(crc, M1));
6075 Node* result = _gvn.transform(new XorINode(crc, b));
6076 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6077
6078 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6079 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6080 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6081 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6082
6083 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6084 result = _gvn.transform(new XorINode(crc, result));
6085 result = _gvn.transform(new XorINode(result, M1));
6086 set_result(result);
6087 return true;
6088 }
6089
6090 /**
6091 * Calculate CRC32 for byte[] array.
6092 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6093 */
6094 bool LibraryCallKit::inline_updateBytesCRC32() {
6095 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6096 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6097 // no receiver since it is static method
6098 Node* crc = argument(0); // type: int
6099 Node* src = argument(1); // type: oop
6100 Node* offset = argument(2); // type: int
6101 Node* length = argument(3); // type: int
6102
6103 const Type* src_type = src->Value(&_gvn);
6104 const TypeAryPtr* top_src = src_type->isa_aryptr();
6105 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6106 // failed array check
6107 return false;
6108 }
6109
6110 // Figure out the size and type of the elements we will be copying.
6111 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6112 if (src_elem != T_BYTE) {
6113 return false;
6114 }
6115
6116 // 'src_start' points to src array + scaled offset
6117 src = must_be_not_null(src, true);
6118 Node* src_start = array_element_address(src, offset, src_elem);
6119
6120 // We assume that range check is done by caller.
6121 // TODO: generate range check (offset+length < src.length) in debug VM.
6122
6123 // Call the stub.
6124 address stubAddr = StubRoutines::updateBytesCRC32();
6125 const char *stubName = "updateBytesCRC32";
6126
6127 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6128 stubAddr, stubName, TypePtr::BOTTOM,
6129 crc, src_start, length);
6130 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6131 set_result(result);
6132 return true;
6133 }
6134
6135 /**
6136 * Calculate CRC32 for ByteBuffer.
6137 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6138 */
6139 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6140 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6141 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6142 // no receiver since it is static method
6143 Node* crc = argument(0); // type: int
6144 Node* src = argument(1); // type: long
6145 Node* offset = argument(3); // type: int
6146 Node* length = argument(4); // type: int
6147
6148 src = ConvL2X(src); // adjust Java long to machine word
6149 Node* base = _gvn.transform(new CastX2PNode(src));
6150 offset = ConvI2X(offset);
6151
6152 // 'src_start' points to src array + scaled offset
6153 Node* src_start = basic_plus_adr(top(), base, offset);
6154
6155 // Call the stub.
6156 address stubAddr = StubRoutines::updateBytesCRC32();
6157 const char *stubName = "updateBytesCRC32";
6158
6159 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6160 stubAddr, stubName, TypePtr::BOTTOM,
6161 crc, src_start, length);
6162 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6163 set_result(result);
6164 return true;
6165 }
6166
6167 //------------------------------get_table_from_crc32c_class-----------------------
6168 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6169 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6170 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6171
6172 return table;
6173 }
6174
6175 //------------------------------inline_updateBytesCRC32C-----------------------
6176 //
6177 // Calculate CRC32C for byte[] array.
6178 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6179 //
6180 bool LibraryCallKit::inline_updateBytesCRC32C() {
6181 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6182 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6183 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6184 // no receiver since it is a static method
6185 Node* crc = argument(0); // type: int
6186 Node* src = argument(1); // type: oop
6187 Node* offset = argument(2); // type: int
6188 Node* end = argument(3); // type: int
6189
6190 Node* length = _gvn.transform(new SubINode(end, offset));
6191
6192 const Type* src_type = src->Value(&_gvn);
6193 const TypeAryPtr* top_src = src_type->isa_aryptr();
6194 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6195 // failed array check
6196 return false;
6197 }
6198
6199 // Figure out the size and type of the elements we will be copying.
6200 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6201 if (src_elem != T_BYTE) {
6202 return false;
6203 }
6204
6205 // 'src_start' points to src array + scaled offset
6206 src = must_be_not_null(src, true);
6207 Node* src_start = array_element_address(src, offset, src_elem);
6208
6209 // static final int[] byteTable in class CRC32C
6210 Node* table = get_table_from_crc32c_class(callee()->holder());
6211 table = must_be_not_null(table, true);
6212 Node* table_start = array_element_address(table, intcon(0), T_INT);
6213
6214 // We assume that range check is done by caller.
6215 // TODO: generate range check (offset+length < src.length) in debug VM.
6216
6217 // Call the stub.
6218 address stubAddr = StubRoutines::updateBytesCRC32C();
6219 const char *stubName = "updateBytesCRC32C";
6220
6221 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6222 stubAddr, stubName, TypePtr::BOTTOM,
6223 crc, src_start, length, table_start);
6224 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6225 set_result(result);
6226 return true;
6227 }
6228
6229 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6230 //
6231 // Calculate CRC32C for DirectByteBuffer.
6232 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6233 //
6234 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6235 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6236 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6237 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6238 // no receiver since it is a static method
6239 Node* crc = argument(0); // type: int
6240 Node* src = argument(1); // type: long
6241 Node* offset = argument(3); // type: int
6242 Node* end = argument(4); // type: int
6243
6244 Node* length = _gvn.transform(new SubINode(end, offset));
6245
6246 src = ConvL2X(src); // adjust Java long to machine word
6247 Node* base = _gvn.transform(new CastX2PNode(src));
6248 offset = ConvI2X(offset);
6249
6250 // 'src_start' points to src array + scaled offset
6251 Node* src_start = basic_plus_adr(top(), base, offset);
6252
6253 // static final int[] byteTable in class CRC32C
6254 Node* table = get_table_from_crc32c_class(callee()->holder());
6255 table = must_be_not_null(table, true);
6256 Node* table_start = array_element_address(table, intcon(0), T_INT);
6257
6258 // Call the stub.
6259 address stubAddr = StubRoutines::updateBytesCRC32C();
6260 const char *stubName = "updateBytesCRC32C";
6261
6262 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6263 stubAddr, stubName, TypePtr::BOTTOM,
6264 crc, src_start, length, table_start);
6265 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6266 set_result(result);
6267 return true;
6268 }
6269
6270 //------------------------------inline_updateBytesAdler32----------------------
6271 //
6272 // Calculate Adler32 checksum for byte[] array.
6273 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6274 //
6275 bool LibraryCallKit::inline_updateBytesAdler32() {
6276 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6277 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6278 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6279 // no receiver since it is static method
6280 Node* crc = argument(0); // type: int
6281 Node* src = argument(1); // type: oop
6282 Node* offset = argument(2); // type: int
6283 Node* length = argument(3); // type: int
6284
6285 const Type* src_type = src->Value(&_gvn);
6286 const TypeAryPtr* top_src = src_type->isa_aryptr();
6287 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
6288 // failed array check
6289 return false;
6290 }
6291
6292 // Figure out the size and type of the elements we will be copying.
6293 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
6294 if (src_elem != T_BYTE) {
6295 return false;
6296 }
6297
6298 // 'src_start' points to src array + scaled offset
6299 Node* src_start = array_element_address(src, offset, src_elem);
6300
6301 // We assume that range check is done by caller.
6302 // TODO: generate range check (offset+length < src.length) in debug VM.
6303
6304 // Call the stub.
6305 address stubAddr = StubRoutines::updateBytesAdler32();
6306 const char *stubName = "updateBytesAdler32";
6307
6308 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6309 stubAddr, stubName, TypePtr::BOTTOM,
6310 crc, src_start, length);
6311 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6312 set_result(result);
6313 return true;
6314 }
6315
6316 //------------------------------inline_updateByteBufferAdler32---------------
6317 //
6318 // Calculate Adler32 checksum for DirectByteBuffer.
6319 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6320 //
6321 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6322 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6323 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6324 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6325 // no receiver since it is static method
6326 Node* crc = argument(0); // type: int
6327 Node* src = argument(1); // type: long
6328 Node* offset = argument(3); // type: int
6329 Node* length = argument(4); // type: int
6330
6331 src = ConvL2X(src); // adjust Java long to machine word
6332 Node* base = _gvn.transform(new CastX2PNode(src));
6333 offset = ConvI2X(offset);
6334
6335 // 'src_start' points to src array + scaled offset
6336 Node* src_start = basic_plus_adr(top(), base, offset);
6337
6338 // Call the stub.
6339 address stubAddr = StubRoutines::updateBytesAdler32();
6340 const char *stubName = "updateBytesAdler32";
6341
6342 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6343 stubAddr, stubName, TypePtr::BOTTOM,
6344 crc, src_start, length);
6345
6346 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6347 set_result(result);
6348 return true;
6349 }
6350
6351 //----------------------------inline_reference_get----------------------------
6352 // public T java.lang.ref.Reference.get();
6353 bool LibraryCallKit::inline_reference_get() {
6354 const int referent_offset = java_lang_ref_Reference::referent_offset();
6355
6356 // Get the argument:
6357 Node* reference_obj = null_check_receiver();
6358 if (stopped()) return true;
6359
6360 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6361 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6362 decorators, /*is_static*/ false, NULL);
6363 if (result == NULL) return false;
6364
6365 // Add memory barrier to prevent commoning reads from this field
6366 // across safepoint since GC can change its value.
6367 insert_mem_bar(Op_MemBarCPUOrder);
6368
6369 set_result(result);
6370 return true;
6371 }
6372
6373 //----------------------------inline_reference_refersTo0----------------------------
6374 // bool java.lang.ref.Reference.refersTo0();
6375 // bool java.lang.ref.PhantomReference.refersTo0();
6376 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6377 // Get arguments:
6378 Node* reference_obj = null_check_receiver();
6379 Node* other_obj = argument(1);
6380 if (stopped()) return true;
6381
6382 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6383 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6384 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6385 decorators, /*is_static*/ false, NULL);
6386 if (referent == NULL) return false;
6387
6388 // Add memory barrier to prevent commoning reads from this field
6389 // across safepoint since GC can change its value.
6390 insert_mem_bar(Op_MemBarCPUOrder);
6391
6392 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6393 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6394 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6395
6396 RegionNode* region = new RegionNode(3);
6397 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6398
6399 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6400 region->init_req(1, if_true);
6401 phi->init_req(1, intcon(1));
6402
6403 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6404 region->init_req(2, if_false);
6405 phi->init_req(2, intcon(0));
6406
6407 set_control(_gvn.transform(region));
6408 record_for_igvn(region);
6409 set_result(_gvn.transform(phi));
6410 return true;
6411 }
6412
6413
6414 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6415 DecoratorSet decorators, bool is_static,
6416 ciInstanceKlass* fromKls) {
6417 if (fromKls == NULL) {
6418 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6419 assert(tinst != NULL, "obj is null");
6420 assert(tinst->is_loaded(), "obj is not loaded");
6421 fromKls = tinst->instance_klass();
6422 } else {
6423 assert(is_static, "only for static field access");
6424 }
6425 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6426 ciSymbol::make(fieldTypeString),
6427 is_static);
6428
6429 assert(field != NULL, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6430 if (field == NULL) return (Node *) NULL;
6431
6432 if (is_static) {
6433 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6434 fromObj = makecon(tip);
6435 }
6436
6437 // Next code copied from Parse::do_get_xxx():
6438
6439 // Compute address and memory type.
6440 int offset = field->offset_in_bytes();
6441 bool is_vol = field->is_volatile();
6442 ciType* field_klass = field->type();
6443 assert(field_klass->is_loaded(), "should be loaded");
6444 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6445 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6446 BasicType bt = field->layout_type();
6447
6448 // Build the resultant type of the load
6449 const Type *type;
6450 if (bt == T_OBJECT) {
6451 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6452 } else {
6453 type = Type::get_const_basic_type(bt);
6454 }
6455
6456 if (is_vol) {
6457 decorators |= MO_SEQ_CST;
6458 }
6459
6460 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6461 }
6462
6463 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6464 bool is_exact /* true */, bool is_static /* false */,
6465 ciInstanceKlass * fromKls /* NULL */) {
6466 if (fromKls == NULL) {
6467 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6468 assert(tinst != NULL, "obj is null");
6469 assert(tinst->is_loaded(), "obj is not loaded");
6470 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6471 fromKls = tinst->instance_klass();
6472 }
6473 else {
6474 assert(is_static, "only for static field access");
6475 }
6476 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6477 ciSymbol::make(fieldTypeString),
6478 is_static);
6479
6480 assert(field != NULL, "undefined field");
6481 assert(!field->is_volatile(), "not defined for volatile fields");
6482
6483 if (is_static) {
6484 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6485 fromObj = makecon(tip);
6486 }
6487
6488 // Next code copied from Parse::do_get_xxx():
6489
6490 // Compute address and memory type.
6491 int offset = field->offset_in_bytes();
6492 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6493
6494 return adr;
6495 }
6496
6497 //------------------------------inline_aescrypt_Block-----------------------
6498 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6499 address stubAddr = NULL;
6500 const char *stubName;
6501 assert(UseAES, "need AES instruction support");
6502
6503 switch(id) {
6504 case vmIntrinsics::_aescrypt_encryptBlock:
6505 stubAddr = StubRoutines::aescrypt_encryptBlock();
6506 stubName = "aescrypt_encryptBlock";
6507 break;
6508 case vmIntrinsics::_aescrypt_decryptBlock:
6509 stubAddr = StubRoutines::aescrypt_decryptBlock();
6510 stubName = "aescrypt_decryptBlock";
6511 break;
6512 default:
6513 break;
6514 }
6515 if (stubAddr == NULL) return false;
6516
6517 Node* aescrypt_object = argument(0);
6518 Node* src = argument(1);
6519 Node* src_offset = argument(2);
6520 Node* dest = argument(3);
6521 Node* dest_offset = argument(4);
6522
6523 src = must_be_not_null(src, true);
6524 dest = must_be_not_null(dest, true);
6525
6526 // (1) src and dest are arrays.
6527 const Type* src_type = src->Value(&_gvn);
6528 const Type* dest_type = dest->Value(&_gvn);
6529 const TypeAryPtr* top_src = src_type->isa_aryptr();
6530 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6531 assert (top_src != NULL && top_src->elem() != Type::BOTTOM && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6532
6533 // for the quick and dirty code we will skip all the checks.
6534 // we are just trying to get the call to be generated.
6535 Node* src_start = src;
6536 Node* dest_start = dest;
6537 if (src_offset != NULL || dest_offset != NULL) {
6538 assert(src_offset != NULL && dest_offset != NULL, "");
6539 src_start = array_element_address(src, src_offset, T_BYTE);
6540 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6541 }
6542
6543 // now need to get the start of its expanded key array
6544 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6545 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6546 if (k_start == NULL) return false;
6547
6548 // Call the stub.
6549 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6550 stubAddr, stubName, TypePtr::BOTTOM,
6551 src_start, dest_start, k_start);
6552
6553 return true;
6554 }
6555
6556 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6557 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6558 address stubAddr = NULL;
6559 const char *stubName = NULL;
6560
6561 assert(UseAES, "need AES instruction support");
6562
6563 switch(id) {
6564 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6565 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6566 stubName = "cipherBlockChaining_encryptAESCrypt";
6567 break;
6568 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6569 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6570 stubName = "cipherBlockChaining_decryptAESCrypt";
6571 break;
6572 default:
6573 break;
6574 }
6575 if (stubAddr == NULL) return false;
6576
6577 Node* cipherBlockChaining_object = argument(0);
6578 Node* src = argument(1);
6579 Node* src_offset = argument(2);
6580 Node* len = argument(3);
6581 Node* dest = argument(4);
6582 Node* dest_offset = argument(5);
6583
6584 src = must_be_not_null(src, false);
6585 dest = must_be_not_null(dest, false);
6586
6587 // (1) src and dest are arrays.
6588 const Type* src_type = src->Value(&_gvn);
6589 const Type* dest_type = dest->Value(&_gvn);
6590 const TypeAryPtr* top_src = src_type->isa_aryptr();
6591 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6592 assert (top_src != NULL && top_src->elem() != Type::BOTTOM
6593 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6594
6595 // checks are the responsibility of the caller
6596 Node* src_start = src;
6597 Node* dest_start = dest;
6598 if (src_offset != NULL || dest_offset != NULL) {
6599 assert(src_offset != NULL && dest_offset != NULL, "");
6600 src_start = array_element_address(src, src_offset, T_BYTE);
6601 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6602 }
6603
6604 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6605 // (because of the predicated logic executed earlier).
6606 // so we cast it here safely.
6607 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6608
6609 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6610 if (embeddedCipherObj == NULL) return false;
6611
6612 // cast it to what we know it will be at runtime
6613 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6614 assert(tinst != NULL, "CBC obj is null");
6615 assert(tinst->is_loaded(), "CBC obj is not loaded");
6616 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6617 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6618
6619 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6620 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6621 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6622 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6623 aescrypt_object = _gvn.transform(aescrypt_object);
6624
6625 // we need to get the start of the aescrypt_object's expanded key array
6626 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6627 if (k_start == NULL) return false;
6628
6629 // similarly, get the start address of the r vector
6630 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6631 if (objRvec == NULL) return false;
6632 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6633
6634 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6635 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6636 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6637 stubAddr, stubName, TypePtr::BOTTOM,
6638 src_start, dest_start, k_start, r_start, len);
6639
6640 // return cipher length (int)
6641 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6642 set_result(retvalue);
6643 return true;
6644 }
6645
6646 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6647 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6648 address stubAddr = NULL;
6649 const char *stubName = NULL;
6650
6651 assert(UseAES, "need AES instruction support");
6652
6653 switch (id) {
6654 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6655 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6656 stubName = "electronicCodeBook_encryptAESCrypt";
6657 break;
6658 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6659 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6660 stubName = "electronicCodeBook_decryptAESCrypt";
6661 break;
6662 default:
6663 break;
6664 }
6665
6666 if (stubAddr == NULL) return false;
6667
6668 Node* electronicCodeBook_object = argument(0);
6669 Node* src = argument(1);
6670 Node* src_offset = argument(2);
6671 Node* len = argument(3);
6672 Node* dest = argument(4);
6673 Node* dest_offset = argument(5);
6674
6675 // (1) src and dest are arrays.
6676 const Type* src_type = src->Value(&_gvn);
6677 const Type* dest_type = dest->Value(&_gvn);
6678 const TypeAryPtr* top_src = src_type->isa_aryptr();
6679 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6680 assert(top_src != NULL && top_src->elem() != Type::BOTTOM
6681 && top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6682
6683 // checks are the responsibility of the caller
6684 Node* src_start = src;
6685 Node* dest_start = dest;
6686 if (src_offset != NULL || dest_offset != NULL) {
6687 assert(src_offset != NULL && dest_offset != NULL, "");
6688 src_start = array_element_address(src, src_offset, T_BYTE);
6689 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6690 }
6691
6692 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6693 // (because of the predicated logic executed earlier).
6694 // so we cast it here safely.
6695 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6696
6697 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6698 if (embeddedCipherObj == NULL) return false;
6699
6700 // cast it to what we know it will be at runtime
6701 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6702 assert(tinst != NULL, "ECB obj is null");
6703 assert(tinst->is_loaded(), "ECB obj is not loaded");
6704 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6705 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6706
6707 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6708 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6709 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6710 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6711 aescrypt_object = _gvn.transform(aescrypt_object);
6712
6713 // we need to get the start of the aescrypt_object's expanded key array
6714 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6715 if (k_start == NULL) return false;
6716
6717 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6718 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6719 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6720 stubAddr, stubName, TypePtr::BOTTOM,
6721 src_start, dest_start, k_start, len);
6722
6723 // return cipher length (int)
6724 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6725 set_result(retvalue);
6726 return true;
6727 }
6728
6729 //------------------------------inline_counterMode_AESCrypt-----------------------
6730 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6731 assert(UseAES, "need AES instruction support");
6732 if (!UseAESCTRIntrinsics) return false;
6733
6734 address stubAddr = NULL;
6735 const char *stubName = NULL;
6736 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6737 stubAddr = StubRoutines::counterMode_AESCrypt();
6738 stubName = "counterMode_AESCrypt";
6739 }
6740 if (stubAddr == NULL) return false;
6741
6742 Node* counterMode_object = argument(0);
6743 Node* src = argument(1);
6744 Node* src_offset = argument(2);
6745 Node* len = argument(3);
6746 Node* dest = argument(4);
6747 Node* dest_offset = argument(5);
6748
6749 // (1) src and dest are arrays.
6750 const Type* src_type = src->Value(&_gvn);
6751 const Type* dest_type = dest->Value(&_gvn);
6752 const TypeAryPtr* top_src = src_type->isa_aryptr();
6753 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6754 assert(top_src != NULL && top_src->elem() != Type::BOTTOM &&
6755 top_dest != NULL && top_dest->elem() != Type::BOTTOM, "args are strange");
6756
6757 // checks are the responsibility of the caller
6758 Node* src_start = src;
6759 Node* dest_start = dest;
6760 if (src_offset != NULL || dest_offset != NULL) {
6761 assert(src_offset != NULL && dest_offset != NULL, "");
6762 src_start = array_element_address(src, src_offset, T_BYTE);
6763 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6764 }
6765
6766 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6767 // (because of the predicated logic executed earlier).
6768 // so we cast it here safely.
6769 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6770 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6771 if (embeddedCipherObj == NULL) return false;
6772 // cast it to what we know it will be at runtime
6773 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6774 assert(tinst != NULL, "CTR obj is null");
6775 assert(tinst->is_loaded(), "CTR obj is not loaded");
6776 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6777 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6778 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6779 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6780 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6781 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6782 aescrypt_object = _gvn.transform(aescrypt_object);
6783 // we need to get the start of the aescrypt_object's expanded key array
6784 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6785 if (k_start == NULL) return false;
6786 // similarly, get the start address of the r vector
6787 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6788 if (obj_counter == NULL) return false;
6789 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6790
6791 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6792 if (saved_encCounter == NULL) return false;
6793 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6794 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6795
6796 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6797 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6798 OptoRuntime::counterMode_aescrypt_Type(),
6799 stubAddr, stubName, TypePtr::BOTTOM,
6800 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6801
6802 // return cipher length (int)
6803 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6804 set_result(retvalue);
6805 return true;
6806 }
6807
6808 //------------------------------get_key_start_from_aescrypt_object-----------------------
6809 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6810 #if defined(PPC64) || defined(S390)
6811 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6812 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6813 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6814 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6815 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6816 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6817 if (objSessionK == NULL) {
6818 return (Node *) NULL;
6819 }
6820 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6821 #else
6822 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6823 #endif // PPC64
6824 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6825 if (objAESCryptKey == NULL) return (Node *) NULL;
6826
6827 // now have the array, need to get the start address of the K array
6828 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6829 return k_start;
6830 }
6831
6832 //----------------------------inline_cipherBlockChaining_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 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6842 // The receiver was checked for NULL already.
6843 Node* objCBC = argument(0);
6844
6845 Node* src = argument(1);
6846 Node* dest = argument(4);
6847
6848 // Load embeddedCipher field of CipherBlockChaining object.
6849 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6850
6851 // get AESCrypt klass for instanceOf check
6852 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6853 // will have same classloader as CipherBlockChaining object
6854 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6855 assert(tinst != NULL, "CBCobj is null");
6856 assert(tinst->is_loaded(), "CBCobj is not loaded");
6857
6858 // we want to do an instanceof comparison against the AESCrypt class
6859 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6860 if (!klass_AESCrypt->is_loaded()) {
6861 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6862 Node* ctrl = control();
6863 set_control(top()); // no regular fast path
6864 return ctrl;
6865 }
6866
6867 src = must_be_not_null(src, true);
6868 dest = must_be_not_null(dest, true);
6869
6870 // Resolve oops to stable for CmpP below.
6871 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6872
6873 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6874 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6875 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6876
6877 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6878
6879 // for encryption, we are done
6880 if (!decrypting)
6881 return instof_false; // even if it is NULL
6882
6883 // for decryption, we need to add a further check to avoid
6884 // taking the intrinsic path when cipher and plain are the same
6885 // see the original java code for why.
6886 RegionNode* region = new RegionNode(3);
6887 region->init_req(1, instof_false);
6888
6889 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6890 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6891 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6892 region->init_req(2, src_dest_conjoint);
6893
6894 record_for_igvn(region);
6895 return _gvn.transform(region);
6896 }
6897
6898 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6899 // Return node representing slow path of predicate check.
6900 // the pseudo code we want to emulate with this predicate is:
6901 // for encryption:
6902 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6903 // for decryption:
6904 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6905 // note cipher==plain is more conservative than the original java code but that's OK
6906 //
6907 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6908 // The receiver was checked for NULL already.
6909 Node* objECB = argument(0);
6910
6911 // Load embeddedCipher field of ElectronicCodeBook object.
6912 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6913
6914 // get AESCrypt klass for instanceOf check
6915 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6916 // will have same classloader as ElectronicCodeBook object
6917 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6918 assert(tinst != NULL, "ECBobj is null");
6919 assert(tinst->is_loaded(), "ECBobj is not loaded");
6920
6921 // we want to do an instanceof comparison against the AESCrypt class
6922 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6923 if (!klass_AESCrypt->is_loaded()) {
6924 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6925 Node* ctrl = control();
6926 set_control(top()); // no regular fast path
6927 return ctrl;
6928 }
6929 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6930
6931 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6932 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6933 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6934
6935 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6936
6937 // for encryption, we are done
6938 if (!decrypting)
6939 return instof_false; // even if it is NULL
6940
6941 // for decryption, we need to add a further check to avoid
6942 // taking the intrinsic path when cipher and plain are the same
6943 // see the original java code for why.
6944 RegionNode* region = new RegionNode(3);
6945 region->init_req(1, instof_false);
6946 Node* src = argument(1);
6947 Node* dest = argument(4);
6948 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6949 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6950 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6951 region->init_req(2, src_dest_conjoint);
6952
6953 record_for_igvn(region);
6954 return _gvn.transform(region);
6955 }
6956
6957 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6958 // Return node representing slow path of predicate check.
6959 // the pseudo code we want to emulate with this predicate is:
6960 // for encryption:
6961 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6962 // for decryption:
6963 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6964 // note cipher==plain is more conservative than the original java code but that's OK
6965 //
6966
6967 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6968 // The receiver was checked for NULL already.
6969 Node* objCTR = argument(0);
6970
6971 // Load embeddedCipher field of CipherBlockChaining object.
6972 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6973
6974 // get AESCrypt klass for instanceOf check
6975 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6976 // will have same classloader as CipherBlockChaining object
6977 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6978 assert(tinst != NULL, "CTRobj is null");
6979 assert(tinst->is_loaded(), "CTRobj is not loaded");
6980
6981 // we want to do an instanceof comparison against the AESCrypt class
6982 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6983 if (!klass_AESCrypt->is_loaded()) {
6984 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6985 Node* ctrl = control();
6986 set_control(top()); // no regular fast path
6987 return ctrl;
6988 }
6989
6990 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6991 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6992 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6993 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6994 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6995
6996 return instof_false; // even if it is NULL
6997 }
6998
6999 //------------------------------inline_ghash_processBlocks
7000 bool LibraryCallKit::inline_ghash_processBlocks() {
7001 address stubAddr;
7002 const char *stubName;
7003 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7004
7005 stubAddr = StubRoutines::ghash_processBlocks();
7006 stubName = "ghash_processBlocks";
7007
7008 Node* data = argument(0);
7009 Node* offset = argument(1);
7010 Node* len = argument(2);
7011 Node* state = argument(3);
7012 Node* subkeyH = argument(4);
7013
7014 state = must_be_not_null(state, true);
7015 subkeyH = must_be_not_null(subkeyH, true);
7016 data = must_be_not_null(data, true);
7017
7018 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7019 assert(state_start, "state is NULL");
7020 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7021 assert(subkeyH_start, "subkeyH is NULL");
7022 Node* data_start = array_element_address(data, offset, T_BYTE);
7023 assert(data_start, "data is NULL");
7024
7025 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7026 OptoRuntime::ghash_processBlocks_Type(),
7027 stubAddr, stubName, TypePtr::BOTTOM,
7028 state_start, subkeyH_start, data_start, len);
7029 return true;
7030 }
7031
7032 bool LibraryCallKit::inline_base64_encodeBlock() {
7033 address stubAddr;
7034 const char *stubName;
7035 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7036 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7037 stubAddr = StubRoutines::base64_encodeBlock();
7038 stubName = "encodeBlock";
7039
7040 if (!stubAddr) return false;
7041 Node* base64obj = argument(0);
7042 Node* src = argument(1);
7043 Node* offset = argument(2);
7044 Node* len = argument(3);
7045 Node* dest = argument(4);
7046 Node* dp = argument(5);
7047 Node* isURL = argument(6);
7048
7049 src = must_be_not_null(src, true);
7050 dest = must_be_not_null(dest, true);
7051
7052 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7053 assert(src_start, "source array is NULL");
7054 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7055 assert(dest_start, "destination array is NULL");
7056
7057 Node* base64 = make_runtime_call(RC_LEAF,
7058 OptoRuntime::base64_encodeBlock_Type(),
7059 stubAddr, stubName, TypePtr::BOTTOM,
7060 src_start, offset, len, dest_start, dp, isURL);
7061 return true;
7062 }
7063
7064 bool LibraryCallKit::inline_base64_decodeBlock() {
7065 address stubAddr;
7066 const char *stubName;
7067 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7068 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7069 stubAddr = StubRoutines::base64_decodeBlock();
7070 stubName = "decodeBlock";
7071
7072 if (!stubAddr) return false;
7073 Node* base64obj = argument(0);
7074 Node* src = argument(1);
7075 Node* src_offset = argument(2);
7076 Node* len = argument(3);
7077 Node* dest = argument(4);
7078 Node* dest_offset = argument(5);
7079 Node* isURL = argument(6);
7080 Node* isMIME = argument(7);
7081
7082 src = must_be_not_null(src, true);
7083 dest = must_be_not_null(dest, true);
7084
7085 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7086 assert(src_start, "source array is NULL");
7087 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7088 assert(dest_start, "destination array is NULL");
7089
7090 Node* call = make_runtime_call(RC_LEAF,
7091 OptoRuntime::base64_decodeBlock_Type(),
7092 stubAddr, stubName, TypePtr::BOTTOM,
7093 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7094 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7095 set_result(result);
7096 return true;
7097 }
7098
7099 //------------------------------inline_digestBase_implCompress-----------------------
7100 //
7101 // Calculate MD5 for single-block byte[] array.
7102 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7103 //
7104 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7105 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7106 //
7107 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7108 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7109 //
7110 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7111 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7112 //
7113 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7114 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7115 //
7116 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7117 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7118
7119 Node* digestBase_obj = argument(0);
7120 Node* src = argument(1); // type oop
7121 Node* ofs = argument(2); // type int
7122
7123 const Type* src_type = src->Value(&_gvn);
7124 const TypeAryPtr* top_src = src_type->isa_aryptr();
7125 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7126 // failed array check
7127 return false;
7128 }
7129 // Figure out the size and type of the elements we will be copying.
7130 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7131 if (src_elem != T_BYTE) {
7132 return false;
7133 }
7134 // 'src_start' points to src array + offset
7135 src = must_be_not_null(src, true);
7136 Node* src_start = array_element_address(src, ofs, src_elem);
7137 Node* state = NULL;
7138 Node* digest_length = NULL;
7139 address stubAddr;
7140 const char *stubName;
7141
7142 switch(id) {
7143 case vmIntrinsics::_md5_implCompress:
7144 assert(UseMD5Intrinsics, "need MD5 instruction support");
7145 state = get_state_from_digest_object(digestBase_obj, T_INT);
7146 stubAddr = StubRoutines::md5_implCompress();
7147 stubName = "md5_implCompress";
7148 break;
7149 case vmIntrinsics::_sha_implCompress:
7150 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7151 state = get_state_from_digest_object(digestBase_obj, T_INT);
7152 stubAddr = StubRoutines::sha1_implCompress();
7153 stubName = "sha1_implCompress";
7154 break;
7155 case vmIntrinsics::_sha2_implCompress:
7156 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7157 state = get_state_from_digest_object(digestBase_obj, T_INT);
7158 stubAddr = StubRoutines::sha256_implCompress();
7159 stubName = "sha256_implCompress";
7160 break;
7161 case vmIntrinsics::_sha5_implCompress:
7162 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7163 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7164 stubAddr = StubRoutines::sha512_implCompress();
7165 stubName = "sha512_implCompress";
7166 break;
7167 case vmIntrinsics::_sha3_implCompress:
7168 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7169 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7170 stubAddr = StubRoutines::sha3_implCompress();
7171 stubName = "sha3_implCompress";
7172 digest_length = get_digest_length_from_digest_object(digestBase_obj);
7173 if (digest_length == NULL) return false;
7174 break;
7175 default:
7176 fatal_unexpected_iid(id);
7177 return false;
7178 }
7179 if (state == NULL) return false;
7180
7181 assert(stubAddr != NULL, "Stub is generated");
7182 if (stubAddr == NULL) return false;
7183
7184 // Call the stub.
7185 Node* call;
7186 if (digest_length == NULL) {
7187 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7188 stubAddr, stubName, TypePtr::BOTTOM,
7189 src_start, state);
7190 } else {
7191 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7192 stubAddr, stubName, TypePtr::BOTTOM,
7193 src_start, state, digest_length);
7194 }
7195
7196 return true;
7197 }
7198
7199 //------------------------------inline_digestBase_implCompressMB-----------------------
7200 //
7201 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7202 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7203 //
7204 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7205 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7206 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7207 assert((uint)predicate < 5, "sanity");
7208 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7209
7210 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7211 Node* src = argument(1); // byte[] array
7212 Node* ofs = argument(2); // type int
7213 Node* limit = argument(3); // type int
7214
7215 const Type* src_type = src->Value(&_gvn);
7216 const TypeAryPtr* top_src = src_type->isa_aryptr();
7217 if (top_src == NULL || top_src->elem() == Type::BOTTOM) {
7218 // failed array check
7219 return false;
7220 }
7221 // Figure out the size and type of the elements we will be copying.
7222 BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
7223 if (src_elem != T_BYTE) {
7224 return false;
7225 }
7226 // 'src_start' points to src array + offset
7227 src = must_be_not_null(src, false);
7228 Node* src_start = array_element_address(src, ofs, src_elem);
7229
7230 const char* klass_digestBase_name = NULL;
7231 const char* stub_name = NULL;
7232 address stub_addr = NULL;
7233 BasicType elem_type = T_INT;
7234
7235 switch (predicate) {
7236 case 0:
7237 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7238 klass_digestBase_name = "sun/security/provider/MD5";
7239 stub_name = "md5_implCompressMB";
7240 stub_addr = StubRoutines::md5_implCompressMB();
7241 }
7242 break;
7243 case 1:
7244 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7245 klass_digestBase_name = "sun/security/provider/SHA";
7246 stub_name = "sha1_implCompressMB";
7247 stub_addr = StubRoutines::sha1_implCompressMB();
7248 }
7249 break;
7250 case 2:
7251 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7252 klass_digestBase_name = "sun/security/provider/SHA2";
7253 stub_name = "sha256_implCompressMB";
7254 stub_addr = StubRoutines::sha256_implCompressMB();
7255 }
7256 break;
7257 case 3:
7258 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7259 klass_digestBase_name = "sun/security/provider/SHA5";
7260 stub_name = "sha512_implCompressMB";
7261 stub_addr = StubRoutines::sha512_implCompressMB();
7262 elem_type = T_LONG;
7263 }
7264 break;
7265 case 4:
7266 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7267 klass_digestBase_name = "sun/security/provider/SHA3";
7268 stub_name = "sha3_implCompressMB";
7269 stub_addr = StubRoutines::sha3_implCompressMB();
7270 elem_type = T_BYTE;
7271 }
7272 break;
7273 default:
7274 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7275 }
7276 if (klass_digestBase_name != NULL) {
7277 assert(stub_addr != NULL, "Stub is generated");
7278 if (stub_addr == NULL) return false;
7279
7280 // get DigestBase klass to lookup for SHA klass
7281 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7282 assert(tinst != NULL, "digestBase_obj is not instance???");
7283 assert(tinst->is_loaded(), "DigestBase is not loaded");
7284
7285 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7286 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7287 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7288 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7289 }
7290 return false;
7291 }
7292
7293 //------------------------------inline_digestBase_implCompressMB-----------------------
7294 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7295 BasicType elem_type, address stubAddr, const char *stubName,
7296 Node* src_start, Node* ofs, Node* limit) {
7297 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7298 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7299 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7300 digest_obj = _gvn.transform(digest_obj);
7301
7302 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7303 if (state == NULL) return false;
7304
7305 Node* digest_length = NULL;
7306 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7307 digest_length = get_digest_length_from_digest_object(digest_obj);
7308 if (digest_length == NULL) return false;
7309 }
7310
7311 // Call the stub.
7312 Node* call;
7313 if (digest_length == NULL) {
7314 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7315 OptoRuntime::digestBase_implCompressMB_Type(false),
7316 stubAddr, stubName, TypePtr::BOTTOM,
7317 src_start, state, ofs, limit);
7318 } else {
7319 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7320 OptoRuntime::digestBase_implCompressMB_Type(true),
7321 stubAddr, stubName, TypePtr::BOTTOM,
7322 src_start, state, digest_length, ofs, limit);
7323 }
7324
7325 // return ofs (int)
7326 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7327 set_result(result);
7328
7329 return true;
7330 }
7331
7332 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7333 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7334 assert(UseAES, "need AES instruction support");
7335 address stubAddr = NULL;
7336 const char *stubName = NULL;
7337 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7338 stubName = "galoisCounterMode_AESCrypt";
7339
7340 if (stubAddr == NULL) return false;
7341
7342 Node* in = argument(0);
7343 Node* inOfs = argument(1);
7344 Node* len = argument(2);
7345 Node* ct = argument(3);
7346 Node* ctOfs = argument(4);
7347 Node* out = argument(5);
7348 Node* outOfs = argument(6);
7349 Node* gctr_object = argument(7);
7350 Node* ghash_object = argument(8);
7351
7352 // (1) in, ct and out are arrays.
7353 const Type* in_type = in->Value(&_gvn);
7354 const Type* ct_type = ct->Value(&_gvn);
7355 const Type* out_type = out->Value(&_gvn);
7356 const TypeAryPtr* top_in = in_type->isa_aryptr();
7357 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7358 const TypeAryPtr* top_out = out_type->isa_aryptr();
7359 assert(top_in != NULL && top_in->elem() != Type::BOTTOM &&
7360 top_ct != NULL && top_ct->elem() != Type::BOTTOM &&
7361 top_out != NULL && top_out->elem() != Type::BOTTOM, "args are strange");
7362
7363 // checks are the responsibility of the caller
7364 Node* in_start = in;
7365 Node* ct_start = ct;
7366 Node* out_start = out;
7367 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7368 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7369 in_start = array_element_address(in, inOfs, T_BYTE);
7370 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7371 out_start = array_element_address(out, outOfs, T_BYTE);
7372 }
7373
7374 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7375 // (because of the predicated logic executed earlier).
7376 // so we cast it here safely.
7377 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7378 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7379 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7380 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7381 Node* state = load_field_from_object(ghash_object, "state", "[J");
7382
7383 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7384 return false;
7385 }
7386 // cast it to what we know it will be at runtime
7387 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7388 assert(tinst != NULL, "GCTR obj is null");
7389 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7390 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7391 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7392 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7393 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7394 const TypeOopPtr* xtype = aklass->as_instance_type();
7395 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7396 aescrypt_object = _gvn.transform(aescrypt_object);
7397 // we need to get the start of the aescrypt_object's expanded key array
7398 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7399 if (k_start == NULL) return false;
7400 // similarly, get the start address of the r vector
7401 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7402 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7403 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7404
7405
7406 // Call the stub, passing params
7407 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7408 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7409 stubAddr, stubName, TypePtr::BOTTOM,
7410 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7411
7412 // return cipher length (int)
7413 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7414 set_result(retvalue);
7415
7416 return true;
7417 }
7418
7419 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7420 // Return node representing slow path of predicate check.
7421 // the pseudo code we want to emulate with this predicate is:
7422 // for encryption:
7423 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7424 // for decryption:
7425 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7426 // note cipher==plain is more conservative than the original java code but that's OK
7427 //
7428
7429 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7430 // The receiver was checked for NULL already.
7431 Node* objGCTR = argument(7);
7432 // Load embeddedCipher field of GCTR object.
7433 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7434 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7435
7436 // get AESCrypt klass for instanceOf check
7437 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7438 // will have same classloader as CipherBlockChaining object
7439 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7440 assert(tinst != NULL, "GCTR obj is null");
7441 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7442
7443 // we want to do an instanceof comparison against the AESCrypt class
7444 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7445 if (!klass_AESCrypt->is_loaded()) {
7446 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7447 Node* ctrl = control();
7448 set_control(top()); // no regular fast path
7449 return ctrl;
7450 }
7451
7452 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7453 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7454 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7455 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7456 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7457
7458 return instof_false; // even if it is NULL
7459 }
7460
7461 //------------------------------get_state_from_digest_object-----------------------
7462 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7463 const char* state_type;
7464 switch (elem_type) {
7465 case T_BYTE: state_type = "[B"; break;
7466 case T_INT: state_type = "[I"; break;
7467 case T_LONG: state_type = "[J"; break;
7468 default: ShouldNotReachHere();
7469 }
7470 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7471 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7472 if (digest_state == NULL) return (Node *) NULL;
7473
7474 // now have the array, need to get the start address of the state array
7475 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7476 return state;
7477 }
7478
7479 //------------------------------get_digest_length_from_sha3_object----------------------------------
7480 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7481 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7482 assert (digest_length != NULL, "sanity");
7483 return digest_length;
7484 }
7485
7486 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7487 // Return node representing slow path of predicate check.
7488 // the pseudo code we want to emulate with this predicate is:
7489 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7490 //
7491 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7492 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7493 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7494 assert((uint)predicate < 5, "sanity");
7495
7496 // The receiver was checked for NULL already.
7497 Node* digestBaseObj = argument(0);
7498
7499 // get DigestBase klass for instanceOf check
7500 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7501 assert(tinst != NULL, "digestBaseObj is null");
7502 assert(tinst->is_loaded(), "DigestBase is not loaded");
7503
7504 const char* klass_name = NULL;
7505 switch (predicate) {
7506 case 0:
7507 if (UseMD5Intrinsics) {
7508 // we want to do an instanceof comparison against the MD5 class
7509 klass_name = "sun/security/provider/MD5";
7510 }
7511 break;
7512 case 1:
7513 if (UseSHA1Intrinsics) {
7514 // we want to do an instanceof comparison against the SHA class
7515 klass_name = "sun/security/provider/SHA";
7516 }
7517 break;
7518 case 2:
7519 if (UseSHA256Intrinsics) {
7520 // we want to do an instanceof comparison against the SHA2 class
7521 klass_name = "sun/security/provider/SHA2";
7522 }
7523 break;
7524 case 3:
7525 if (UseSHA512Intrinsics) {
7526 // we want to do an instanceof comparison against the SHA5 class
7527 klass_name = "sun/security/provider/SHA5";
7528 }
7529 break;
7530 case 4:
7531 if (UseSHA3Intrinsics) {
7532 // we want to do an instanceof comparison against the SHA3 class
7533 klass_name = "sun/security/provider/SHA3";
7534 }
7535 break;
7536 default:
7537 fatal("unknown SHA intrinsic predicate: %d", predicate);
7538 }
7539
7540 ciKlass* klass = NULL;
7541 if (klass_name != NULL) {
7542 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7543 }
7544 if ((klass == NULL) || !klass->is_loaded()) {
7545 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7546 Node* ctrl = control();
7547 set_control(top()); // no intrinsic path
7548 return ctrl;
7549 }
7550 ciInstanceKlass* instklass = klass->as_instance_klass();
7551
7552 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7553 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7554 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7555 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7556
7557 return instof_false; // even if it is NULL
7558 }
7559
7560 bool LibraryCallKit::inline_continuation_do_yield() {
7561 address call_addr = StubRoutines::cont_doYield();
7562 const TypeFunc* tf = OptoRuntime::continuation_doYield_Type();
7563 Node* call = make_runtime_call(RC_NO_LEAF, tf, call_addr, "doYield", TypeRawPtr::BOTTOM);
7564 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7565 set_result(result);
7566 return true;
7567 }
7568
7569 //-------------inline_fma-----------------------------------
7570 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7571 Node *a = NULL;
7572 Node *b = NULL;
7573 Node *c = NULL;
7574 Node* result = NULL;
7575 switch (id) {
7576 case vmIntrinsics::_fmaD:
7577 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7578 // no receiver since it is static method
7579 a = round_double_node(argument(0));
7580 b = round_double_node(argument(2));
7581 c = round_double_node(argument(4));
7582 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7583 break;
7584 case vmIntrinsics::_fmaF:
7585 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7586 a = argument(0);
7587 b = argument(1);
7588 c = argument(2);
7589 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7590 break;
7591 default:
7592 fatal_unexpected_iid(id); break;
7593 }
7594 set_result(result);
7595 return true;
7596 }
7597
7598 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7599 // argument(0) is receiver
7600 Node* codePoint = argument(1);
7601 Node* n = NULL;
7602
7603 switch (id) {
7604 case vmIntrinsics::_isDigit :
7605 n = new DigitNode(control(), codePoint);
7606 break;
7607 case vmIntrinsics::_isLowerCase :
7608 n = new LowerCaseNode(control(), codePoint);
7609 break;
7610 case vmIntrinsics::_isUpperCase :
7611 n = new UpperCaseNode(control(), codePoint);
7612 break;
7613 case vmIntrinsics::_isWhitespace :
7614 n = new WhitespaceNode(control(), codePoint);
7615 break;
7616 default:
7617 fatal_unexpected_iid(id);
7618 }
7619
7620 set_result(_gvn.transform(n));
7621 return true;
7622 }
7623
7624 //------------------------------inline_fp_min_max------------------------------
7625 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7626 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7627
7628 // The intrinsic should be used only when the API branches aren't predictable,
7629 // the last one performing the most important comparison. The following heuristic
7630 // uses the branch statistics to eventually bail out if necessary.
7631
7632 ciMethodData *md = callee()->method_data();
7633
7634 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7635 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7636
7637 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7638 // Bail out if the call-site didn't contribute enough to the statistics.
7639 return false;
7640 }
7641
7642 uint taken = 0, not_taken = 0;
7643
7644 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7645 if (p->is_BranchData()) {
7646 taken = ((ciBranchData*)p)->taken();
7647 not_taken = ((ciBranchData*)p)->not_taken();
7648 }
7649 }
7650
7651 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7652 balance = balance < 0 ? -balance : balance;
7653 if ( balance > 0.2 ) {
7654 // Bail out if the most important branch is predictable enough.
7655 return false;
7656 }
7657 }
7658 */
7659
7660 Node *a = NULL;
7661 Node *b = NULL;
7662 Node *n = NULL;
7663 switch (id) {
7664 case vmIntrinsics::_maxF:
7665 case vmIntrinsics::_minF:
7666 case vmIntrinsics::_maxF_strict:
7667 case vmIntrinsics::_minF_strict:
7668 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7669 a = argument(0);
7670 b = argument(1);
7671 break;
7672 case vmIntrinsics::_maxD:
7673 case vmIntrinsics::_minD:
7674 case vmIntrinsics::_maxD_strict:
7675 case vmIntrinsics::_minD_strict:
7676 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7677 a = round_double_node(argument(0));
7678 b = round_double_node(argument(2));
7679 break;
7680 default:
7681 fatal_unexpected_iid(id);
7682 break;
7683 }
7684 switch (id) {
7685 case vmIntrinsics::_maxF:
7686 case vmIntrinsics::_maxF_strict:
7687 n = new MaxFNode(a, b);
7688 break;
7689 case vmIntrinsics::_minF:
7690 case vmIntrinsics::_minF_strict:
7691 n = new MinFNode(a, b);
7692 break;
7693 case vmIntrinsics::_maxD:
7694 case vmIntrinsics::_maxD_strict:
7695 n = new MaxDNode(a, b);
7696 break;
7697 case vmIntrinsics::_minD:
7698 case vmIntrinsics::_minD_strict:
7699 n = new MinDNode(a, b);
7700 break;
7701 default:
7702 fatal_unexpected_iid(id);
7703 break;
7704 }
7705 set_result(_gvn.transform(n));
7706 return true;
7707 }
7708
7709 bool LibraryCallKit::inline_profileBoolean() {
7710 Node* counts = argument(1);
7711 const TypeAryPtr* ary = NULL;
7712 ciArray* aobj = NULL;
7713 if (counts->is_Con()
7714 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7715 && (aobj = ary->const_oop()->as_array()) != NULL
7716 && (aobj->length() == 2)) {
7717 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7718 jint false_cnt = aobj->element_value(0).as_int();
7719 jint true_cnt = aobj->element_value(1).as_int();
7720
7721 if (C->log() != NULL) {
7722 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7723 false_cnt, true_cnt);
7724 }
7725
7726 if (false_cnt + true_cnt == 0) {
7727 // According to profile, never executed.
7728 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7729 Deoptimization::Action_reinterpret);
7730 return true;
7731 }
7732
7733 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7734 // is a number of each value occurrences.
7735 Node* result = argument(0);
7736 if (false_cnt == 0 || true_cnt == 0) {
7737 // According to profile, one value has been never seen.
7738 int expected_val = (false_cnt == 0) ? 1 : 0;
7739
7740 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7741 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7742
7743 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7744 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7745 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7746
7747 { // Slow path: uncommon trap for never seen value and then reexecute
7748 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7749 // the value has been seen at least once.
7750 PreserveJVMState pjvms(this);
7751 PreserveReexecuteState preexecs(this);
7752 jvms()->set_should_reexecute(true);
7753
7754 set_control(slow_path);
7755 set_i_o(i_o());
7756
7757 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7758 Deoptimization::Action_reinterpret);
7759 }
7760 // The guard for never seen value enables sharpening of the result and
7761 // returning a constant. It allows to eliminate branches on the same value
7762 // later on.
7763 set_control(fast_path);
7764 result = intcon(expected_val);
7765 }
7766 // Stop profiling.
7767 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7768 // By replacing method body with profile data (represented as ProfileBooleanNode
7769 // on IR level) we effectively disable profiling.
7770 // It enables full speed execution once optimized code is generated.
7771 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7772 C->record_for_igvn(profile);
7773 set_result(profile);
7774 return true;
7775 } else {
7776 // Continue profiling.
7777 // Profile data isn't available at the moment. So, execute method's bytecode version.
7778 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7779 // is compiled and counters aren't available since corresponding MethodHandle
7780 // isn't a compile-time constant.
7781 return false;
7782 }
7783 }
7784
7785 bool LibraryCallKit::inline_isCompileConstant() {
7786 Node* n = argument(0);
7787 set_result(n->is_Con() ? intcon(1) : intcon(0));
7788 return true;
7789 }
7790
7791 //------------------------------- inline_getObjectSize --------------------------------------
7792 //
7793 // Calculate the runtime size of the object/array.
7794 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7795 //
7796 bool LibraryCallKit::inline_getObjectSize() {
7797 Node* obj = argument(3);
7798 Node* klass_node = load_object_klass(obj);
7799
7800 jint layout_con = Klass::_lh_neutral_value;
7801 Node* layout_val = get_layout_helper(klass_node, layout_con);
7802 int layout_is_con = (layout_val == NULL);
7803
7804 if (layout_is_con) {
7805 // Layout helper is constant, can figure out things at compile time.
7806
7807 if (Klass::layout_helper_is_instance(layout_con)) {
7808 // Instance case: layout_con contains the size itself.
7809 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7810 set_result(size);
7811 } else {
7812 // Array case: size is round(header + element_size*arraylength).
7813 // Since arraylength is different for every array instance, we have to
7814 // compute the whole thing at runtime.
7815
7816 Node* arr_length = load_array_length(obj);
7817
7818 int round_mask = MinObjAlignmentInBytes - 1;
7819 int hsize = Klass::layout_helper_header_size(layout_con);
7820 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7821
7822 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7823 round_mask = 0; // strength-reduce it if it goes away completely
7824 }
7825 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7826 Node* header_size = intcon(hsize + round_mask);
7827
7828 Node* lengthx = ConvI2X(arr_length);
7829 Node* headerx = ConvI2X(header_size);
7830
7831 Node* abody = lengthx;
7832 if (eshift != 0) {
7833 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7834 }
7835 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7836 if (round_mask != 0) {
7837 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7838 }
7839 size = ConvX2L(size);
7840 set_result(size);
7841 }
7842 } else {
7843 // Layout helper is not constant, need to test for array-ness at runtime.
7844
7845 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7846 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7847 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7848 record_for_igvn(result_reg);
7849
7850 Node* array_ctl = generate_array_guard(klass_node, NULL);
7851 if (array_ctl != NULL) {
7852 // Array case: size is round(header + element_size*arraylength).
7853 // Since arraylength is different for every array instance, we have to
7854 // compute the whole thing at runtime.
7855
7856 PreserveJVMState pjvms(this);
7857 set_control(array_ctl);
7858 Node* arr_length = load_array_length(obj);
7859
7860 int round_mask = MinObjAlignmentInBytes - 1;
7861 Node* mask = intcon(round_mask);
7862
7863 Node* hss = intcon(Klass::_lh_header_size_shift);
7864 Node* hsm = intcon(Klass::_lh_header_size_mask);
7865 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7866 header_size = _gvn.transform(new AndINode(header_size, hsm));
7867 header_size = _gvn.transform(new AddINode(header_size, mask));
7868
7869 // There is no need to mask or shift this value.
7870 // The semantics of LShiftINode include an implicit mask to 0x1F.
7871 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7872 Node* elem_shift = layout_val;
7873
7874 Node* lengthx = ConvI2X(arr_length);
7875 Node* headerx = ConvI2X(header_size);
7876
7877 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7878 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7879 if (round_mask != 0) {
7880 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7881 }
7882 size = ConvX2L(size);
7883
7884 result_reg->init_req(_array_path, control());
7885 result_val->init_req(_array_path, size);
7886 }
7887
7888 if (!stopped()) {
7889 // Instance case: the layout helper gives us instance size almost directly,
7890 // but we need to mask out the _lh_instance_slow_path_bit.
7891 Node* size = ConvI2X(layout_val);
7892 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7893 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7894 size = _gvn.transform(new AndXNode(size, mask));
7895 size = ConvX2L(size);
7896
7897 result_reg->init_req(_instance_path, control());
7898 result_val->init_req(_instance_path, size);
7899 }
7900
7901 set_result(result_reg, result_val);
7902 }
7903
7904 return true;
7905 }
7906
7907 //------------------------------- inline_blackhole --------------------------------------
7908 //
7909 // Make sure all arguments to this node are alive.
7910 // This matches methods that were requested to be blackholed through compile commands.
7911 //
7912 bool LibraryCallKit::inline_blackhole() {
7913 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7914 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7915 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7916
7917 // Bind call arguments as blackhole arguments to keep them alive
7918 Node* bh = insert_mem_bar(Op_Blackhole);
7919 uint nargs = callee()->arg_size();
7920 for (uint i = 0; i < nargs; i++) {
7921 bh->add_req(argument(i));
7922 }
7923
7924 return true;
7925 }