1 /*
2 * Copyright (c) 1999, 2021, 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/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/unsafe.hpp"
55 #include "runtime/objectMonitor.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "utilities/macros.hpp"
59 #include "utilities/powerOfTwo.hpp"
60
61 #if INCLUDE_JFR
62 #include "jfr/jfr.hpp"
63 #endif
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return NULL;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
85 !C->directive()->is_intrinsic_disabled(mh) &&
86 !vmIntrinsics::is_disabled_by_flags(mh);
87
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return NULL;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
124 if (C->print_intrinsics() || C->print_inlining()) {
125 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
126 }
127 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
128 if (C->log()) {
129 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
130 vmIntrinsics::name_at(intrinsic_id()),
131 (is_virtual() ? " virtual='1'" : ""),
132 C->unique() - nodes);
133 }
134 // Push the result from the inlined method onto the stack.
135 kit.push_result();
136 C->print_inlining_update(this);
137 return kit.transfer_exceptions_into_jvms();
138 }
139
140 // The intrinsic bailed out
141 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
142 if (jvms->has_method()) {
143 // Not a root compile.
144 const char* msg;
145 if (callee->intrinsic_candidate()) {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
147 } else {
148 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
149 : "failed to inline (intrinsic), method not annotated";
150 }
151 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
152 if (C->print_intrinsics() || C->print_inlining()) {
153 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
154 }
155 } else {
156 // Root compile
157 ResourceMark rm;
158 stringStream msg_stream;
159 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
160 vmIntrinsics::name_at(intrinsic_id()),
161 is_virtual() ? " (virtual)" : "", bci);
162 const char *msg = msg_stream.as_string();
163 log_debug(jit, inlining)("%s", msg);
164 if (C->print_intrinsics() || C->print_inlining()) {
165 tty->print("%s", msg);
166 }
167 }
168 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
169 C->print_inlining_update(this);
170
171 return NULL;
172 }
173
174 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
175 LibraryCallKit kit(jvms, this);
176 Compile* C = kit.C;
177 int nodes = C->unique();
178 _last_predicate = predicate;
179 #ifndef PRODUCT
180 assert(is_predicated() && predicate < predicates_count(), "sanity");
181 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
182 char buf[1000];
183 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
184 tty->print_cr("Predicate for intrinsic %s", str);
185 }
186 #endif
187 ciMethod* callee = kit.callee();
188 const int bci = kit.bci();
189
190 Node* slow_ctl = kit.try_to_predicate(predicate);
191 if (!kit.failing()) {
192 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
193 : "(intrinsic, predicate)";
194 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
195 if (C->print_intrinsics() || C->print_inlining()) {
196 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
197 }
198 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
199 if (C->log()) {
200 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
201 vmIntrinsics::name_at(intrinsic_id()),
202 (is_virtual() ? " virtual='1'" : ""),
203 C->unique() - nodes);
204 }
205 return slow_ctl; // Could be NULL if the check folds.
206 }
207
208 // The intrinsic bailed out
209 if (jvms->has_method()) {
210 // Not a root compile.
211 const char* msg = "failed to generate predicate for intrinsic";
212 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
213 if (C->print_intrinsics() || C->print_inlining()) {
214 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
215 }
216 } else {
217 // Root compile
218 ResourceMark rm;
219 stringStream msg_stream;
220 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
221 vmIntrinsics::name_at(intrinsic_id()),
222 is_virtual() ? " (virtual)" : "", bci);
223 const char *msg = msg_stream.as_string();
224 log_debug(jit, inlining)("%s", msg);
225 if (C->print_intrinsics() || C->print_inlining()) {
226 C->print_inlining_stream()->print("%s", msg);
227 }
228 }
229 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
230 return NULL;
231 }
232
233 bool LibraryCallKit::try_to_inline(int predicate) {
234 // Handle symbolic names for otherwise undistinguished boolean switches:
235 const bool is_store = true;
236 const bool is_compress = true;
237 const bool is_static = true;
238 const bool is_volatile = true;
239
240 if (!jvms()->has_method()) {
241 // Root JVMState has a null method.
242 assert(map()->memory()->Opcode() == Op_Parm, "");
243 // Insert the memory aliasing node
244 set_all_memory(reset_memory());
245 }
246 assert(merged_memory(), "");
247
248 switch (intrinsic_id()) {
249 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
250 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
251 case vmIntrinsics::_getClass: return inline_native_getClass();
252
253 case vmIntrinsics::_ceil:
254 case vmIntrinsics::_floor:
255 case vmIntrinsics::_rint:
256 case vmIntrinsics::_dsin:
257 case vmIntrinsics::_dcos:
258 case vmIntrinsics::_dtan:
259 case vmIntrinsics::_dabs:
260 case vmIntrinsics::_fabs:
261 case vmIntrinsics::_iabs:
262 case vmIntrinsics::_labs:
263 case vmIntrinsics::_datan2:
264 case vmIntrinsics::_dsqrt:
265 case vmIntrinsics::_dsqrt_strict:
266 case vmIntrinsics::_dexp:
267 case vmIntrinsics::_dlog:
268 case vmIntrinsics::_dlog10:
269 case vmIntrinsics::_dpow:
270 case vmIntrinsics::_dcopySign:
271 case vmIntrinsics::_fcopySign:
272 case vmIntrinsics::_dsignum:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
297 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
298 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
299 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
300
301 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
308 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
309
310 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
311 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
312
313 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
314 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
315 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
316 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
317
318 case vmIntrinsics::_compressStringC:
319 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
320 case vmIntrinsics::_inflateStringC:
321 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
322
323 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
324 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
325 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
326 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
327 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
328 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
329 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
330 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
331 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
332 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
333 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
334 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_INLINE_TYPE,Relaxed, false);
335
336 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
337 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
338 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
339 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
340 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
341 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
342 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
343 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
344 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
345 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_INLINE_TYPE,Relaxed, false);
346
347 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
348 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
349 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
350 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
351 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
352 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
353 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
354 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
355 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
356
357 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
358 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
359 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
360 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
361 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
362 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
363 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
364 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
365 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
366
367 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
373 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
374 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
375 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
376
377 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
378 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
379 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
380 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
381 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
382 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
383 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
384 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
385 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
386
387 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
388 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
389 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
390 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
391 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
392 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
393 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
394 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
395 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
396
397 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
398 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
399 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
400 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
401 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
402 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
403 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
404 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
405 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
406
407 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
408 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
409 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
410 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
411 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
412 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
413 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
414 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
415 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
416
417 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
422
423 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
443
444 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
459
460 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
461 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
464
465 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
470
471 case vmIntrinsics::_loadFence:
472 case vmIntrinsics::_storeFence:
473 case vmIntrinsics::_storeStoreFence:
474 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
475
476 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
477
478 case vmIntrinsics::_currentThread: return inline_native_currentThread();
479
480 #ifdef JFR_HAVE_INTRINSICS
481 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
482 case vmIntrinsics::_getClassId: return inline_native_classID();
483 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
484 #endif
485 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
486 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
487 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
488 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
489 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
490 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
491 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
492 case vmIntrinsics::_getLength: return inline_native_getLength();
493 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
494 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
495 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
496 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
497 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
498 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
499 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
500
501 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
502 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
503
504 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
505
506 case vmIntrinsics::_isInstance:
507 case vmIntrinsics::_getModifiers:
508 case vmIntrinsics::_isInterface:
509 case vmIntrinsics::_isArray:
510 case vmIntrinsics::_isPrimitive:
511 case vmIntrinsics::_isHidden:
512 case vmIntrinsics::_getSuperclass:
513 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
514
515 case vmIntrinsics::_asPrimaryType:
516 case vmIntrinsics::_asValueType: return inline_primitive_Class_conversion(intrinsic_id());
517
518 case vmIntrinsics::_floatToRawIntBits:
519 case vmIntrinsics::_floatToIntBits:
520 case vmIntrinsics::_intBitsToFloat:
521 case vmIntrinsics::_doubleToRawLongBits:
522 case vmIntrinsics::_doubleToLongBits:
523 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
524
525 case vmIntrinsics::_numberOfLeadingZeros_i:
526 case vmIntrinsics::_numberOfLeadingZeros_l:
527 case vmIntrinsics::_numberOfTrailingZeros_i:
528 case vmIntrinsics::_numberOfTrailingZeros_l:
529 case vmIntrinsics::_bitCount_i:
530 case vmIntrinsics::_bitCount_l:
531 case vmIntrinsics::_reverseBytes_i:
532 case vmIntrinsics::_reverseBytes_l:
533 case vmIntrinsics::_reverseBytes_s:
534 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
535
536 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
537
538 case vmIntrinsics::_Reference_get: return inline_reference_get();
539 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
540 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
541
542 case vmIntrinsics::_Class_cast: return inline_Class_cast();
543
544 case vmIntrinsics::_aescrypt_encryptBlock:
545 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
546
547 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
548 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
549 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
550
551 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
552 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
553 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
554
555 case vmIntrinsics::_counterMode_AESCrypt:
556 return inline_counterMode_AESCrypt(intrinsic_id());
557
558 case vmIntrinsics::_galoisCounterMode_AESCrypt:
559 return inline_galoisCounterMode_AESCrypt();
560
561 case vmIntrinsics::_md5_implCompress:
562 case vmIntrinsics::_sha_implCompress:
563 case vmIntrinsics::_sha2_implCompress:
564 case vmIntrinsics::_sha5_implCompress:
565 case vmIntrinsics::_sha3_implCompress:
566 return inline_digestBase_implCompress(intrinsic_id());
567
568 case vmIntrinsics::_digestBase_implCompressMB:
569 return inline_digestBase_implCompressMB(predicate);
570
571 case vmIntrinsics::_multiplyToLen:
572 return inline_multiplyToLen();
573
574 case vmIntrinsics::_squareToLen:
575 return inline_squareToLen();
576
577 case vmIntrinsics::_mulAdd:
578 return inline_mulAdd();
579
580 case vmIntrinsics::_montgomeryMultiply:
581 return inline_montgomeryMultiply();
582 case vmIntrinsics::_montgomerySquare:
583 return inline_montgomerySquare();
584
585 case vmIntrinsics::_bigIntegerRightShiftWorker:
586 return inline_bigIntegerShift(true);
587 case vmIntrinsics::_bigIntegerLeftShiftWorker:
588 return inline_bigIntegerShift(false);
589
590 case vmIntrinsics::_vectorizedMismatch:
591 return inline_vectorizedMismatch();
592
593 case vmIntrinsics::_ghash_processBlocks:
594 return inline_ghash_processBlocks();
595 case vmIntrinsics::_base64_encodeBlock:
596 return inline_base64_encodeBlock();
597 case vmIntrinsics::_base64_decodeBlock:
598 return inline_base64_decodeBlock();
599
600 case vmIntrinsics::_encodeISOArray:
601 case vmIntrinsics::_encodeByteISOArray:
602 return inline_encodeISOArray(false);
603 case vmIntrinsics::_encodeAsciiArray:
604 return inline_encodeISOArray(true);
605
606 case vmIntrinsics::_updateCRC32:
607 return inline_updateCRC32();
608 case vmIntrinsics::_updateBytesCRC32:
609 return inline_updateBytesCRC32();
610 case vmIntrinsics::_updateByteBufferCRC32:
611 return inline_updateByteBufferCRC32();
612
613 case vmIntrinsics::_updateBytesCRC32C:
614 return inline_updateBytesCRC32C();
615 case vmIntrinsics::_updateDirectByteBufferCRC32C:
616 return inline_updateDirectByteBufferCRC32C();
617
618 case vmIntrinsics::_updateBytesAdler32:
619 return inline_updateBytesAdler32();
620 case vmIntrinsics::_updateByteBufferAdler32:
621 return inline_updateByteBufferAdler32();
622
623 case vmIntrinsics::_profileBoolean:
624 return inline_profileBoolean();
625 case vmIntrinsics::_isCompileConstant:
626 return inline_isCompileConstant();
627
628 case vmIntrinsics::_hasNegatives:
629 return inline_hasNegatives();
630
631 case vmIntrinsics::_fmaD:
632 case vmIntrinsics::_fmaF:
633 return inline_fma(intrinsic_id());
634
635 case vmIntrinsics::_isDigit:
636 case vmIntrinsics::_isLowerCase:
637 case vmIntrinsics::_isUpperCase:
638 case vmIntrinsics::_isWhitespace:
639 return inline_character_compare(intrinsic_id());
640
641 case vmIntrinsics::_min:
642 case vmIntrinsics::_max:
643 case vmIntrinsics::_min_strict:
644 case vmIntrinsics::_max_strict:
645 return inline_min_max(intrinsic_id());
646
647 case vmIntrinsics::_maxF:
648 case vmIntrinsics::_minF:
649 case vmIntrinsics::_maxD:
650 case vmIntrinsics::_minD:
651 case vmIntrinsics::_maxF_strict:
652 case vmIntrinsics::_minF_strict:
653 case vmIntrinsics::_maxD_strict:
654 case vmIntrinsics::_minD_strict:
655 return inline_fp_min_max(intrinsic_id());
656
657 case vmIntrinsics::_VectorUnaryOp:
658 return inline_vector_nary_operation(1);
659 case vmIntrinsics::_VectorBinaryOp:
660 return inline_vector_nary_operation(2);
661 case vmIntrinsics::_VectorTernaryOp:
662 return inline_vector_nary_operation(3);
663 case vmIntrinsics::_VectorBroadcastCoerced:
664 return inline_vector_broadcast_coerced();
665 case vmIntrinsics::_VectorShuffleIota:
666 return inline_vector_shuffle_iota();
667 case vmIntrinsics::_VectorMaskOp:
668 return inline_vector_mask_operation();
669 case vmIntrinsics::_VectorShuffleToVector:
670 return inline_vector_shuffle_to_vector();
671 case vmIntrinsics::_VectorLoadOp:
672 return inline_vector_mem_operation(/*is_store=*/false);
673 case vmIntrinsics::_VectorLoadMaskedOp:
674 return inline_vector_mem_masked_operation(/*is_store*/false);
675 case vmIntrinsics::_VectorStoreOp:
676 return inline_vector_mem_operation(/*is_store=*/true);
677 case vmIntrinsics::_VectorStoreMaskedOp:
678 return inline_vector_mem_masked_operation(/*is_store=*/true);
679 case vmIntrinsics::_VectorGatherOp:
680 return inline_vector_gather_scatter(/*is_scatter*/ false);
681 case vmIntrinsics::_VectorScatterOp:
682 return inline_vector_gather_scatter(/*is_scatter*/ true);
683 case vmIntrinsics::_VectorReductionCoerced:
684 return inline_vector_reduction();
685 case vmIntrinsics::_VectorTest:
686 return inline_vector_test();
687 case vmIntrinsics::_VectorBlend:
688 return inline_vector_blend();
689 case vmIntrinsics::_VectorRearrange:
690 return inline_vector_rearrange();
691 case vmIntrinsics::_VectorCompare:
692 return inline_vector_compare();
693 case vmIntrinsics::_VectorBroadcastInt:
694 return inline_vector_broadcast_int();
695 case vmIntrinsics::_VectorConvert:
696 return inline_vector_convert();
697 case vmIntrinsics::_VectorInsert:
698 return inline_vector_insert();
699 case vmIntrinsics::_VectorExtract:
700 return inline_vector_extract();
701
702 case vmIntrinsics::_getObjectSize:
703 return inline_getObjectSize();
704
705 case vmIntrinsics::_blackhole:
706 return inline_blackhole();
707
708 default:
709 // If you get here, it may be that someone has added a new intrinsic
710 // to the list in vmIntrinsics.hpp without implementing it here.
711 #ifndef PRODUCT
712 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
713 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
714 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
715 }
716 #endif
717 return false;
718 }
719 }
720
721 Node* LibraryCallKit::try_to_predicate(int predicate) {
722 if (!jvms()->has_method()) {
723 // Root JVMState has a null method.
724 assert(map()->memory()->Opcode() == Op_Parm, "");
725 // Insert the memory aliasing node
726 set_all_memory(reset_memory());
727 }
728 assert(merged_memory(), "");
729
730 switch (intrinsic_id()) {
731 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
732 return inline_cipherBlockChaining_AESCrypt_predicate(false);
733 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
734 return inline_cipherBlockChaining_AESCrypt_predicate(true);
735 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
736 return inline_electronicCodeBook_AESCrypt_predicate(false);
737 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
738 return inline_electronicCodeBook_AESCrypt_predicate(true);
739 case vmIntrinsics::_counterMode_AESCrypt:
740 return inline_counterMode_AESCrypt_predicate();
741 case vmIntrinsics::_digestBase_implCompressMB:
742 return inline_digestBase_implCompressMB_predicate(predicate);
743 case vmIntrinsics::_galoisCounterMode_AESCrypt:
744 return inline_galoisCounterMode_AESCrypt_predicate();
745
746 default:
747 // If you get here, it may be that someone has added a new intrinsic
748 // to the list in vmIntrinsics.hpp without implementing it here.
749 #ifndef PRODUCT
750 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
751 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
752 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
753 }
754 #endif
755 Node* slow_ctl = control();
756 set_control(top()); // No fast path instrinsic
757 return slow_ctl;
758 }
759 }
760
761 //------------------------------set_result-------------------------------
762 // Helper function for finishing intrinsics.
763 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
764 record_for_igvn(region);
765 set_control(_gvn.transform(region));
766 set_result( _gvn.transform(value));
767 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
768 }
769
770 //------------------------------generate_guard---------------------------
771 // Helper function for generating guarded fast-slow graph structures.
772 // The given 'test', if true, guards a slow path. If the test fails
773 // then a fast path can be taken. (We generally hope it fails.)
774 // In all cases, GraphKit::control() is updated to the fast path.
775 // The returned value represents the control for the slow path.
776 // The return value is never 'top'; it is either a valid control
777 // or NULL if it is obvious that the slow path can never be taken.
778 // Also, if region and the slow control are not NULL, the slow edge
779 // is appended to the region.
780 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
781 if (stopped()) {
782 // Already short circuited.
783 return NULL;
784 }
785
786 // Build an if node and its projections.
787 // If test is true we take the slow path, which we assume is uncommon.
788 if (_gvn.type(test) == TypeInt::ZERO) {
789 // The slow branch is never taken. No need to build this guard.
790 return NULL;
791 }
792
793 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
794
795 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
796 if (if_slow == top()) {
797 // The slow branch is never taken. No need to build this guard.
798 return NULL;
799 }
800
801 if (region != NULL)
802 region->add_req(if_slow);
803
804 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
805 set_control(if_fast);
806
807 return if_slow;
808 }
809
810 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
811 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
812 }
813 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
814 return generate_guard(test, region, PROB_FAIR);
815 }
816
817 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
818 Node* *pos_index) {
819 if (stopped())
820 return NULL; // already stopped
821 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
822 return NULL; // index is already adequately typed
823 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
824 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
825 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
826 if (is_neg != NULL && pos_index != NULL) {
827 // Emulate effect of Parse::adjust_map_after_if.
828 Node* ccast = new CastIINode(index, TypeInt::POS);
829 ccast->set_req(0, control());
830 (*pos_index) = _gvn.transform(ccast);
831 }
832 return is_neg;
833 }
834
835 // Make sure that 'position' is a valid limit index, in [0..length].
836 // There are two equivalent plans for checking this:
837 // A. (offset + copyLength) unsigned<= arrayLength
838 // B. offset <= (arrayLength - copyLength)
839 // We require that all of the values above, except for the sum and
840 // difference, are already known to be non-negative.
841 // Plan A is robust in the face of overflow, if offset and copyLength
842 // are both hugely positive.
843 //
844 // Plan B is less direct and intuitive, but it does not overflow at
845 // all, since the difference of two non-negatives is always
846 // representable. Whenever Java methods must perform the equivalent
847 // check they generally use Plan B instead of Plan A.
848 // For the moment we use Plan A.
849 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
850 Node* subseq_length,
851 Node* array_length,
852 RegionNode* region) {
853 if (stopped())
854 return NULL; // already stopped
855 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
856 if (zero_offset && subseq_length->eqv_uncast(array_length))
857 return NULL; // common case of whole-array copy
858 Node* last = subseq_length;
859 if (!zero_offset) // last += offset
860 last = _gvn.transform(new AddINode(last, offset));
861 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
862 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
863 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
864 return is_over;
865 }
866
867 // Emit range checks for the given String.value byte array
868 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
869 if (stopped()) {
870 return; // already stopped
871 }
872 RegionNode* bailout = new RegionNode(1);
873 record_for_igvn(bailout);
874 if (char_count) {
875 // Convert char count to byte count
876 count = _gvn.transform(new LShiftINode(count, intcon(1)));
877 }
878
879 // Offset and count must not be negative
880 generate_negative_guard(offset, bailout);
881 generate_negative_guard(count, bailout);
882 // Offset + count must not exceed length of array
883 generate_limit_guard(offset, count, load_array_length(array), bailout);
884
885 if (bailout->req() > 1) {
886 PreserveJVMState pjvms(this);
887 set_control(_gvn.transform(bailout));
888 uncommon_trap(Deoptimization::Reason_intrinsic,
889 Deoptimization::Action_maybe_recompile);
890 }
891 }
892
893 //--------------------------generate_current_thread--------------------
894 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
895 ciKlass* thread_klass = env()->Thread_klass();
896 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
897 Node* thread = _gvn.transform(new ThreadLocalNode());
898 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
899 tls_output = thread;
900 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
901 thread_obj_handle = _gvn.transform(thread_obj_handle);
902 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
903 }
904
905
906 //------------------------------make_string_method_node------------------------
907 // Helper method for String intrinsic functions. This version is called with
908 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
909 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
910 // containing the lengths of str1 and str2.
911 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
912 Node* result = NULL;
913 switch (opcode) {
914 case Op_StrIndexOf:
915 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
916 str1_start, cnt1, str2_start, cnt2, ae);
917 break;
918 case Op_StrComp:
919 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
920 str1_start, cnt1, str2_start, cnt2, ae);
921 break;
922 case Op_StrEquals:
923 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
924 // Use the constant length if there is one because optimized match rule may exist.
925 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
926 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
927 break;
928 default:
929 ShouldNotReachHere();
930 return NULL;
931 }
932
933 // All these intrinsics have checks.
934 C->set_has_split_ifs(true); // Has chance for split-if optimization
935 clear_upper_avx();
936
937 return _gvn.transform(result);
938 }
939
940 //------------------------------inline_string_compareTo------------------------
941 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
942 Node* arg1 = argument(0);
943 Node* arg2 = argument(1);
944
945 arg1 = must_be_not_null(arg1, true);
946 arg2 = must_be_not_null(arg2, true);
947
948 // Get start addr and length of first argument
949 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
950 Node* arg1_cnt = load_array_length(arg1);
951
952 // Get start addr and length of second argument
953 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
954 Node* arg2_cnt = load_array_length(arg2);
955
956 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
957 set_result(result);
958 return true;
959 }
960
961 //------------------------------inline_string_equals------------------------
962 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
963 Node* arg1 = argument(0);
964 Node* arg2 = argument(1);
965
966 // paths (plus control) merge
967 RegionNode* region = new RegionNode(3);
968 Node* phi = new PhiNode(region, TypeInt::BOOL);
969
970 if (!stopped()) {
971
972 arg1 = must_be_not_null(arg1, true);
973 arg2 = must_be_not_null(arg2, true);
974
975 // Get start addr and length of first argument
976 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
977 Node* arg1_cnt = load_array_length(arg1);
978
979 // Get start addr and length of second argument
980 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
981 Node* arg2_cnt = load_array_length(arg2);
982
983 // Check for arg1_cnt != arg2_cnt
984 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
985 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
986 Node* if_ne = generate_slow_guard(bol, NULL);
987 if (if_ne != NULL) {
988 phi->init_req(2, intcon(0));
989 region->init_req(2, if_ne);
990 }
991
992 // Check for count == 0 is done by assembler code for StrEquals.
993
994 if (!stopped()) {
995 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
996 phi->init_req(1, equals);
997 region->init_req(1, control());
998 }
999 }
1000
1001 // post merge
1002 set_control(_gvn.transform(region));
1003 record_for_igvn(region);
1004
1005 set_result(_gvn.transform(phi));
1006 return true;
1007 }
1008
1009 //------------------------------inline_array_equals----------------------------
1010 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1011 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1012 Node* arg1 = argument(0);
1013 Node* arg2 = argument(1);
1014
1015 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1016 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1017 clear_upper_avx();
1018
1019 return true;
1020 }
1021
1022 //------------------------------inline_hasNegatives------------------------------
1023 bool LibraryCallKit::inline_hasNegatives() {
1024 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1025 return false;
1026 }
1027
1028 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1029 // no receiver since it is static method
1030 Node* ba = argument(0);
1031 Node* offset = argument(1);
1032 Node* len = argument(2);
1033
1034 ba = must_be_not_null(ba, true);
1035
1036 // Range checks
1037 generate_string_range_check(ba, offset, len, false);
1038 if (stopped()) {
1039 return true;
1040 }
1041 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1042 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1043 set_result(_gvn.transform(result));
1044 return true;
1045 }
1046
1047 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1048 Node* index = argument(0);
1049 Node* length = bt == T_INT ? argument(1) : argument(2);
1050 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1051 return false;
1052 }
1053
1054 // check that length is positive
1055 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1056 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1057
1058 {
1059 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1060 uncommon_trap(Deoptimization::Reason_intrinsic,
1061 Deoptimization::Action_make_not_entrant);
1062 }
1063
1064 if (stopped()) {
1065 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1066 return true;
1067 }
1068
1069 // length is now known postive, add a cast node to make this explicit
1070 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1071 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1072 casted_length = _gvn.transform(casted_length);
1073 replace_in_map(length, casted_length);
1074 length = casted_length;
1075
1076 // Use an unsigned comparison for the range check itself
1077 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1078 BoolTest::mask btest = BoolTest::lt;
1079 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1080 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1081 _gvn.set_type(rc, rc->Value(&_gvn));
1082 if (!rc_bool->is_Con()) {
1083 record_for_igvn(rc);
1084 }
1085 set_control(_gvn.transform(new IfTrueNode(rc)));
1086 {
1087 PreserveJVMState pjvms(this);
1088 set_control(_gvn.transform(new IfFalseNode(rc)));
1089 uncommon_trap(Deoptimization::Reason_range_check,
1090 Deoptimization::Action_make_not_entrant);
1091 }
1092
1093 if (stopped()) {
1094 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1095 return true;
1096 }
1097
1098 // index is now known to be >= 0 and < length, cast it
1099 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1100 result = _gvn.transform(result);
1101 set_result(result);
1102 replace_in_map(index, result);
1103 clear_upper_avx();
1104 return true;
1105 }
1106
1107 //------------------------------inline_string_indexOf------------------------
1108 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1109 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1110 return false;
1111 }
1112 Node* src = argument(0);
1113 Node* tgt = argument(1);
1114
1115 // Make the merge point
1116 RegionNode* result_rgn = new RegionNode(4);
1117 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1118
1119 src = must_be_not_null(src, true);
1120 tgt = must_be_not_null(tgt, true);
1121
1122 // Get start addr and length of source string
1123 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1124 Node* src_count = load_array_length(src);
1125
1126 // Get start addr and length of substring
1127 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1128 Node* tgt_count = load_array_length(tgt);
1129
1130 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1131 // Divide src size by 2 if String is UTF16 encoded
1132 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1133 }
1134 if (ae == StrIntrinsicNode::UU) {
1135 // Divide substring size by 2 if String is UTF16 encoded
1136 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1137 }
1138
1139 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1140 if (result != NULL) {
1141 result_phi->init_req(3, result);
1142 result_rgn->init_req(3, control());
1143 }
1144 set_control(_gvn.transform(result_rgn));
1145 record_for_igvn(result_rgn);
1146 set_result(_gvn.transform(result_phi));
1147
1148 return true;
1149 }
1150
1151 //-----------------------------inline_string_indexOf-----------------------
1152 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1153 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1154 return false;
1155 }
1156 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1157 return false;
1158 }
1159 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1160 Node* src = argument(0); // byte[]
1161 Node* src_count = argument(1); // char count
1162 Node* tgt = argument(2); // byte[]
1163 Node* tgt_count = argument(3); // char count
1164 Node* from_index = argument(4); // char index
1165
1166 src = must_be_not_null(src, true);
1167 tgt = must_be_not_null(tgt, true);
1168
1169 // Multiply byte array index by 2 if String is UTF16 encoded
1170 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1171 src_count = _gvn.transform(new SubINode(src_count, from_index));
1172 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1173 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1174
1175 // Range checks
1176 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1177 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1178 if (stopped()) {
1179 return true;
1180 }
1181
1182 RegionNode* region = new RegionNode(5);
1183 Node* phi = new PhiNode(region, TypeInt::INT);
1184
1185 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1186 if (result != NULL) {
1187 // The result is index relative to from_index if substring was found, -1 otherwise.
1188 // Generate code which will fold into cmove.
1189 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1190 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1191
1192 Node* if_lt = generate_slow_guard(bol, NULL);
1193 if (if_lt != NULL) {
1194 // result == -1
1195 phi->init_req(3, result);
1196 region->init_req(3, if_lt);
1197 }
1198 if (!stopped()) {
1199 result = _gvn.transform(new AddINode(result, from_index));
1200 phi->init_req(4, result);
1201 region->init_req(4, control());
1202 }
1203 }
1204
1205 set_control(_gvn.transform(region));
1206 record_for_igvn(region);
1207 set_result(_gvn.transform(phi));
1208 clear_upper_avx();
1209
1210 return true;
1211 }
1212
1213 // Create StrIndexOfNode with fast path checks
1214 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1215 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1216 // Check for substr count > string count
1217 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1218 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1219 Node* if_gt = generate_slow_guard(bol, NULL);
1220 if (if_gt != NULL) {
1221 phi->init_req(1, intcon(-1));
1222 region->init_req(1, if_gt);
1223 }
1224 if (!stopped()) {
1225 // Check for substr count == 0
1226 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1227 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1228 Node* if_zero = generate_slow_guard(bol, NULL);
1229 if (if_zero != NULL) {
1230 phi->init_req(2, intcon(0));
1231 region->init_req(2, if_zero);
1232 }
1233 }
1234 if (!stopped()) {
1235 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1236 }
1237 return NULL;
1238 }
1239
1240 //-----------------------------inline_string_indexOfChar-----------------------
1241 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1242 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1243 return false;
1244 }
1245 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1246 return false;
1247 }
1248 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1249 Node* src = argument(0); // byte[]
1250 Node* tgt = argument(1); // tgt is int ch
1251 Node* from_index = argument(2);
1252 Node* max = argument(3);
1253
1254 src = must_be_not_null(src, true);
1255
1256 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1257 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1258 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1259
1260 // Range checks
1261 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1262 if (stopped()) {
1263 return true;
1264 }
1265
1266 RegionNode* region = new RegionNode(3);
1267 Node* phi = new PhiNode(region, TypeInt::INT);
1268
1269 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1270 C->set_has_split_ifs(true); // Has chance for split-if optimization
1271 _gvn.transform(result);
1272
1273 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1274 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1275
1276 Node* if_lt = generate_slow_guard(bol, NULL);
1277 if (if_lt != NULL) {
1278 // result == -1
1279 phi->init_req(2, result);
1280 region->init_req(2, if_lt);
1281 }
1282 if (!stopped()) {
1283 result = _gvn.transform(new AddINode(result, from_index));
1284 phi->init_req(1, result);
1285 region->init_req(1, control());
1286 }
1287 set_control(_gvn.transform(region));
1288 record_for_igvn(region);
1289 set_result(_gvn.transform(phi));
1290
1291 return true;
1292 }
1293 //---------------------------inline_string_copy---------------------
1294 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1295 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1296 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1297 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1298 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1299 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1300 bool LibraryCallKit::inline_string_copy(bool compress) {
1301 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1302 return false;
1303 }
1304 int nargs = 5; // 2 oops, 3 ints
1305 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1306
1307 Node* src = argument(0);
1308 Node* src_offset = argument(1);
1309 Node* dst = argument(2);
1310 Node* dst_offset = argument(3);
1311 Node* length = argument(4);
1312
1313 // Check for allocation before we add nodes that would confuse
1314 // tightly_coupled_allocation()
1315 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1316
1317 // Figure out the size and type of the elements we will be copying.
1318 const Type* src_type = src->Value(&_gvn);
1319 const Type* dst_type = dst->Value(&_gvn);
1320 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1321 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1322 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1323 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1324 "Unsupported array types for inline_string_copy");
1325
1326 src = must_be_not_null(src, true);
1327 dst = must_be_not_null(dst, true);
1328
1329 // Convert char[] offsets to byte[] offsets
1330 bool convert_src = (compress && src_elem == T_BYTE);
1331 bool convert_dst = (!compress && dst_elem == T_BYTE);
1332 if (convert_src) {
1333 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1334 } else if (convert_dst) {
1335 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1336 }
1337
1338 // Range checks
1339 generate_string_range_check(src, src_offset, length, convert_src);
1340 generate_string_range_check(dst, dst_offset, length, convert_dst);
1341 if (stopped()) {
1342 return true;
1343 }
1344
1345 Node* src_start = array_element_address(src, src_offset, src_elem);
1346 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1347 // 'src_start' points to src array + scaled offset
1348 // 'dst_start' points to dst array + scaled offset
1349 Node* count = NULL;
1350 if (compress) {
1351 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1352 } else {
1353 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1354 }
1355
1356 if (alloc != NULL) {
1357 if (alloc->maybe_set_complete(&_gvn)) {
1358 // "You break it, you buy it."
1359 InitializeNode* init = alloc->initialization();
1360 assert(init->is_complete(), "we just did this");
1361 init->set_complete_with_arraycopy();
1362 assert(dst->is_CheckCastPP(), "sanity");
1363 assert(dst->in(0)->in(0) == init, "dest pinned");
1364 }
1365 // Do not let stores that initialize this object be reordered with
1366 // a subsequent store that would make this object accessible by
1367 // other threads.
1368 // Record what AllocateNode this StoreStore protects so that
1369 // escape analysis can go from the MemBarStoreStoreNode to the
1370 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1371 // based on the escape status of the AllocateNode.
1372 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1373 }
1374 if (compress) {
1375 set_result(_gvn.transform(count));
1376 }
1377 clear_upper_avx();
1378
1379 return true;
1380 }
1381
1382 #ifdef _LP64
1383 #define XTOP ,top() /*additional argument*/
1384 #else //_LP64
1385 #define XTOP /*no additional argument*/
1386 #endif //_LP64
1387
1388 //------------------------inline_string_toBytesU--------------------------
1389 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1390 bool LibraryCallKit::inline_string_toBytesU() {
1391 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1392 return false;
1393 }
1394 // Get the arguments.
1395 Node* value = argument(0);
1396 Node* offset = argument(1);
1397 Node* length = argument(2);
1398
1399 Node* newcopy = NULL;
1400
1401 // Set the original stack and the reexecute bit for the interpreter to reexecute
1402 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1403 { PreserveReexecuteState preexecs(this);
1404 jvms()->set_should_reexecute(true);
1405
1406 // Check if a null path was taken unconditionally.
1407 value = null_check(value);
1408
1409 RegionNode* bailout = new RegionNode(1);
1410 record_for_igvn(bailout);
1411
1412 // Range checks
1413 generate_negative_guard(offset, bailout);
1414 generate_negative_guard(length, bailout);
1415 generate_limit_guard(offset, length, load_array_length(value), bailout);
1416 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1417 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1418
1419 if (bailout->req() > 1) {
1420 PreserveJVMState pjvms(this);
1421 set_control(_gvn.transform(bailout));
1422 uncommon_trap(Deoptimization::Reason_intrinsic,
1423 Deoptimization::Action_maybe_recompile);
1424 }
1425 if (stopped()) {
1426 return true;
1427 }
1428
1429 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1430 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1431 newcopy = new_array(klass_node, size, 0); // no arguments to push
1432 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1433 guarantee(alloc != NULL, "created above");
1434
1435 // Calculate starting addresses.
1436 Node* src_start = array_element_address(value, offset, T_CHAR);
1437 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1438
1439 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1440 const TypeInt* toffset = gvn().type(offset)->is_int();
1441 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1442
1443 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1444 const char* copyfunc_name = "arraycopy";
1445 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1446 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1447 OptoRuntime::fast_arraycopy_Type(),
1448 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1449 src_start, dst_start, ConvI2X(length) XTOP);
1450 // Do not let reads from the cloned object float above the arraycopy.
1451 if (alloc->maybe_set_complete(&_gvn)) {
1452 // "You break it, you buy it."
1453 InitializeNode* init = alloc->initialization();
1454 assert(init->is_complete(), "we just did this");
1455 init->set_complete_with_arraycopy();
1456 assert(newcopy->is_CheckCastPP(), "sanity");
1457 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1458 }
1459 // Do not let stores that initialize this object be reordered with
1460 // a subsequent store that would make this object accessible by
1461 // other threads.
1462 // Record what AllocateNode this StoreStore protects so that
1463 // escape analysis can go from the MemBarStoreStoreNode to the
1464 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1465 // based on the escape status of the AllocateNode.
1466 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1467 } // original reexecute is set back here
1468
1469 C->set_has_split_ifs(true); // Has chance for split-if optimization
1470 if (!stopped()) {
1471 set_result(newcopy);
1472 }
1473 clear_upper_avx();
1474
1475 return true;
1476 }
1477
1478 //------------------------inline_string_getCharsU--------------------------
1479 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1480 bool LibraryCallKit::inline_string_getCharsU() {
1481 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1482 return false;
1483 }
1484
1485 // Get the arguments.
1486 Node* src = argument(0);
1487 Node* src_begin = argument(1);
1488 Node* src_end = argument(2); // exclusive offset (i < src_end)
1489 Node* dst = argument(3);
1490 Node* dst_begin = argument(4);
1491
1492 // Check for allocation before we add nodes that would confuse
1493 // tightly_coupled_allocation()
1494 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1495
1496 // Check if a null path was taken unconditionally.
1497 src = null_check(src);
1498 dst = null_check(dst);
1499 if (stopped()) {
1500 return true;
1501 }
1502
1503 // Get length and convert char[] offset to byte[] offset
1504 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1505 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1506
1507 // Range checks
1508 generate_string_range_check(src, src_begin, length, true);
1509 generate_string_range_check(dst, dst_begin, length, false);
1510 if (stopped()) {
1511 return true;
1512 }
1513
1514 if (!stopped()) {
1515 // Calculate starting addresses.
1516 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1517 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1518
1519 // Check if array addresses are aligned to HeapWordSize
1520 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1521 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1522 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1523 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1524
1525 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1526 const char* copyfunc_name = "arraycopy";
1527 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1528 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1529 OptoRuntime::fast_arraycopy_Type(),
1530 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1531 src_start, dst_start, ConvI2X(length) XTOP);
1532 // Do not let reads from the cloned object float above the arraycopy.
1533 if (alloc != NULL) {
1534 if (alloc->maybe_set_complete(&_gvn)) {
1535 // "You break it, you buy it."
1536 InitializeNode* init = alloc->initialization();
1537 assert(init->is_complete(), "we just did this");
1538 init->set_complete_with_arraycopy();
1539 assert(dst->is_CheckCastPP(), "sanity");
1540 assert(dst->in(0)->in(0) == init, "dest pinned");
1541 }
1542 // Do not let stores that initialize this object be reordered with
1543 // a subsequent store that would make this object accessible by
1544 // other threads.
1545 // Record what AllocateNode this StoreStore protects so that
1546 // escape analysis can go from the MemBarStoreStoreNode to the
1547 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1548 // based on the escape status of the AllocateNode.
1549 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1550 } else {
1551 insert_mem_bar(Op_MemBarCPUOrder);
1552 }
1553 }
1554
1555 C->set_has_split_ifs(true); // Has chance for split-if optimization
1556 return true;
1557 }
1558
1559 //----------------------inline_string_char_access----------------------------
1560 // Store/Load char to/from byte[] array.
1561 // static void StringUTF16.putChar(byte[] val, int index, int c)
1562 // static char StringUTF16.getChar(byte[] val, int index)
1563 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1564 Node* value = argument(0);
1565 Node* index = argument(1);
1566 Node* ch = is_store ? argument(2) : NULL;
1567
1568 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1569 // correctly requires matched array shapes.
1570 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1571 "sanity: byte[] and char[] bases agree");
1572 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1573 "sanity: byte[] and char[] scales agree");
1574
1575 // Bail when getChar over constants is requested: constant folding would
1576 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1577 // Java method would constant fold nicely instead.
1578 if (!is_store && value->is_Con() && index->is_Con()) {
1579 return false;
1580 }
1581
1582 value = must_be_not_null(value, true);
1583
1584 Node* adr = array_element_address(value, index, T_CHAR);
1585 if (adr->is_top()) {
1586 return false;
1587 }
1588 if (is_store) {
1589 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1590 } else {
1591 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1592 set_result(ch);
1593 }
1594 return true;
1595 }
1596
1597 //--------------------------round_double_node--------------------------------
1598 // Round a double node if necessary.
1599 Node* LibraryCallKit::round_double_node(Node* n) {
1600 if (Matcher::strict_fp_requires_explicit_rounding) {
1601 #ifdef IA32
1602 if (UseSSE < 2) {
1603 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1604 }
1605 #else
1606 Unimplemented();
1607 #endif // IA32
1608 }
1609 return n;
1610 }
1611
1612 //------------------------------inline_math-----------------------------------
1613 // public static double Math.abs(double)
1614 // public static double Math.sqrt(double)
1615 // public static double Math.log(double)
1616 // public static double Math.log10(double)
1617 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1618 Node* arg = round_double_node(argument(0));
1619 Node* n = NULL;
1620 switch (id) {
1621 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1622 case vmIntrinsics::_dsqrt:
1623 case vmIntrinsics::_dsqrt_strict:
1624 n = new SqrtDNode(C, control(), arg); break;
1625 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1626 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1627 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1628 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1629 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1630 default: fatal_unexpected_iid(id); break;
1631 }
1632 set_result(_gvn.transform(n));
1633 return true;
1634 }
1635
1636 //------------------------------inline_math-----------------------------------
1637 // public static float Math.abs(float)
1638 // public static int Math.abs(int)
1639 // public static long Math.abs(long)
1640 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1641 Node* arg = argument(0);
1642 Node* n = NULL;
1643 switch (id) {
1644 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1645 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1646 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1647 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1648 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1649 default: fatal_unexpected_iid(id); break;
1650 }
1651 set_result(_gvn.transform(n));
1652 return true;
1653 }
1654
1655 //------------------------------runtime_math-----------------------------
1656 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1657 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1658 "must be (DD)D or (D)D type");
1659
1660 // Inputs
1661 Node* a = round_double_node(argument(0));
1662 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1663
1664 const TypePtr* no_memory_effects = NULL;
1665 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1666 no_memory_effects,
1667 a, top(), b, b ? top() : NULL);
1668 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1669 #ifdef ASSERT
1670 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1671 assert(value_top == top(), "second value must be top");
1672 #endif
1673
1674 set_result(value);
1675 return true;
1676 }
1677
1678 //------------------------------inline_math_pow-----------------------------
1679 bool LibraryCallKit::inline_math_pow() {
1680 Node* exp = round_double_node(argument(2));
1681 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1682 if (d != NULL) {
1683 if (d->getd() == 2.0) {
1684 // Special case: pow(x, 2.0) => x * x
1685 Node* base = round_double_node(argument(0));
1686 set_result(_gvn.transform(new MulDNode(base, base)));
1687 return true;
1688 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1689 // Special case: pow(x, 0.5) => sqrt(x)
1690 Node* base = round_double_node(argument(0));
1691 Node* zero = _gvn.zerocon(T_DOUBLE);
1692
1693 RegionNode* region = new RegionNode(3);
1694 Node* phi = new PhiNode(region, Type::DOUBLE);
1695
1696 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1697 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1698 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1699 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1700 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1701
1702 Node* if_pow = generate_slow_guard(test, NULL);
1703 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1704 phi->init_req(1, value_sqrt);
1705 region->init_req(1, control());
1706
1707 if (if_pow != NULL) {
1708 set_control(if_pow);
1709 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1710 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1711 const TypePtr* no_memory_effects = NULL;
1712 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1713 no_memory_effects, base, top(), exp, top());
1714 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1715 #ifdef ASSERT
1716 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1717 assert(value_top == top(), "second value must be top");
1718 #endif
1719 phi->init_req(2, value_pow);
1720 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1721 }
1722
1723 C->set_has_split_ifs(true); // Has chance for split-if optimization
1724 set_control(_gvn.transform(region));
1725 record_for_igvn(region);
1726 set_result(_gvn.transform(phi));
1727
1728 return true;
1729 }
1730 }
1731
1732 return StubRoutines::dpow() != NULL ?
1733 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1734 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1735 }
1736
1737 //------------------------------inline_math_native-----------------------------
1738 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1739 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1740 switch (id) {
1741 // These intrinsics are not properly supported on all hardware
1742 case vmIntrinsics::_dsin:
1743 return StubRoutines::dsin() != NULL ?
1744 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1745 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1746 case vmIntrinsics::_dcos:
1747 return StubRoutines::dcos() != NULL ?
1748 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1749 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1750 case vmIntrinsics::_dtan:
1751 return StubRoutines::dtan() != NULL ?
1752 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1753 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1754 case vmIntrinsics::_dlog:
1755 return StubRoutines::dlog() != NULL ?
1756 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1757 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1758 case vmIntrinsics::_dlog10:
1759 return StubRoutines::dlog10() != NULL ?
1760 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1761 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1762
1763 // These intrinsics are supported on all hardware
1764 case vmIntrinsics::_ceil:
1765 case vmIntrinsics::_floor:
1766 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1767 case vmIntrinsics::_dsqrt:
1768 case vmIntrinsics::_dsqrt_strict:
1769 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1770 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1771 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1772 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1773 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1774
1775 case vmIntrinsics::_dexp:
1776 return StubRoutines::dexp() != NULL ?
1777 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1778 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1779 #undef FN_PTR
1780
1781 case vmIntrinsics::_dpow: return inline_math_pow();
1782 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1783 case vmIntrinsics::_fcopySign: return inline_math(id);
1784 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1785 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1786
1787 // These intrinsics are not yet correctly implemented
1788 case vmIntrinsics::_datan2:
1789 return false;
1790
1791 default:
1792 fatal_unexpected_iid(id);
1793 return false;
1794 }
1795 }
1796
1797 static bool is_simple_name(Node* n) {
1798 return (n->req() == 1 // constant
1799 || (n->is_Type() && n->as_Type()->type()->singleton())
1800 || n->is_Proj() // parameter or return value
1801 || n->is_Phi() // local of some sort
1802 );
1803 }
1804
1805 //----------------------------inline_notify-----------------------------------*
1806 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1807 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1808 address func;
1809 if (id == vmIntrinsics::_notify) {
1810 func = OptoRuntime::monitor_notify_Java();
1811 } else {
1812 func = OptoRuntime::monitor_notifyAll_Java();
1813 }
1814 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1815 make_slow_call_ex(call, env()->Throwable_klass(), false);
1816 return true;
1817 }
1818
1819
1820 //----------------------------inline_min_max-----------------------------------
1821 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1822 set_result(generate_min_max(id, argument(0), argument(1)));
1823 return true;
1824 }
1825
1826 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1827 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1828 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1829 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1830 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1831
1832 {
1833 PreserveJVMState pjvms(this);
1834 PreserveReexecuteState preexecs(this);
1835 jvms()->set_should_reexecute(true);
1836
1837 set_control(slow_path);
1838 set_i_o(i_o());
1839
1840 uncommon_trap(Deoptimization::Reason_intrinsic,
1841 Deoptimization::Action_none);
1842 }
1843
1844 set_control(fast_path);
1845 set_result(math);
1846 }
1847
1848 template <typename OverflowOp>
1849 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1850 typedef typename OverflowOp::MathOp MathOp;
1851
1852 MathOp* mathOp = new MathOp(arg1, arg2);
1853 Node* operation = _gvn.transform( mathOp );
1854 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1855 inline_math_mathExact(operation, ofcheck);
1856 return true;
1857 }
1858
1859 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1860 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1861 }
1862
1863 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1864 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1865 }
1866
1867 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1868 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1869 }
1870
1871 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1872 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1873 }
1874
1875 bool LibraryCallKit::inline_math_negateExactI() {
1876 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1877 }
1878
1879 bool LibraryCallKit::inline_math_negateExactL() {
1880 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1881 }
1882
1883 bool LibraryCallKit::inline_math_multiplyExactI() {
1884 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1885 }
1886
1887 bool LibraryCallKit::inline_math_multiplyExactL() {
1888 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1889 }
1890
1891 bool LibraryCallKit::inline_math_multiplyHigh() {
1892 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1893 return true;
1894 }
1895
1896 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1897 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1898 return true;
1899 }
1900
1901 Node*
1902 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1903 // These are the candidate return value:
1904 Node* xvalue = x0;
1905 Node* yvalue = y0;
1906
1907 if (xvalue == yvalue) {
1908 return xvalue;
1909 }
1910
1911 bool want_max = (id == vmIntrinsics::_max || id == vmIntrinsics::_max_strict);
1912
1913 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1914 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1915 if (txvalue == NULL || tyvalue == NULL) return top();
1916 // This is not really necessary, but it is consistent with a
1917 // hypothetical MaxINode::Value method:
1918 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1919
1920 // %%% This folding logic should (ideally) be in a different place.
1921 // Some should be inside IfNode, and there to be a more reliable
1922 // transformation of ?: style patterns into cmoves. We also want
1923 // more powerful optimizations around cmove and min/max.
1924
1925 // Try to find a dominating comparison of these guys.
1926 // It can simplify the index computation for Arrays.copyOf
1927 // and similar uses of System.arraycopy.
1928 // First, compute the normalized version of CmpI(x, y).
1929 int cmp_op = Op_CmpI;
1930 Node* xkey = xvalue;
1931 Node* ykey = yvalue;
1932 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1933 if (ideal_cmpxy->is_Cmp()) {
1934 // E.g., if we have CmpI(length - offset, count),
1935 // it might idealize to CmpI(length, count + offset)
1936 cmp_op = ideal_cmpxy->Opcode();
1937 xkey = ideal_cmpxy->in(1);
1938 ykey = ideal_cmpxy->in(2);
1939 }
1940
1941 // Start by locating any relevant comparisons.
1942 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1943 Node* cmpxy = NULL;
1944 Node* cmpyx = NULL;
1945 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1946 Node* cmp = start_from->fast_out(k);
1947 if (cmp->outcnt() > 0 && // must have prior uses
1948 cmp->in(0) == NULL && // must be context-independent
1949 cmp->Opcode() == cmp_op) { // right kind of compare
1950 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1951 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1952 }
1953 }
1954
1955 const int NCMPS = 2;
1956 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1957 int cmpn;
1958 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1959 if (cmps[cmpn] != NULL) break; // find a result
1960 }
1961 if (cmpn < NCMPS) {
1962 // Look for a dominating test that tells us the min and max.
1963 int depth = 0; // Limit search depth for speed
1964 Node* dom = control();
1965 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1966 if (++depth >= 100) break;
1967 Node* ifproj = dom;
1968 if (!ifproj->is_Proj()) continue;
1969 Node* iff = ifproj->in(0);
1970 if (!iff->is_If()) continue;
1971 Node* bol = iff->in(1);
1972 if (!bol->is_Bool()) continue;
1973 Node* cmp = bol->in(1);
1974 if (cmp == NULL) continue;
1975 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1976 if (cmps[cmpn] == cmp) break;
1977 if (cmpn == NCMPS) continue;
1978 BoolTest::mask btest = bol->as_Bool()->_test._test;
1979 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1980 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1981 // At this point, we know that 'x btest y' is true.
1982 switch (btest) {
1983 case BoolTest::eq:
1984 // They are proven equal, so we can collapse the min/max.
1985 // Either value is the answer. Choose the simpler.
1986 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1987 return yvalue;
1988 return xvalue;
1989 case BoolTest::lt: // x < y
1990 case BoolTest::le: // x <= y
1991 return (want_max ? yvalue : xvalue);
1992 case BoolTest::gt: // x > y
1993 case BoolTest::ge: // x >= y
1994 return (want_max ? xvalue : yvalue);
1995 default:
1996 break;
1997 }
1998 }
1999 }
2000
2001 // We failed to find a dominating test.
2002 // Let's pick a test that might GVN with prior tests.
2003 Node* best_bol = NULL;
2004 BoolTest::mask best_btest = BoolTest::illegal;
2005 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2006 Node* cmp = cmps[cmpn];
2007 if (cmp == NULL) continue;
2008 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2009 Node* bol = cmp->fast_out(j);
2010 if (!bol->is_Bool()) continue;
2011 BoolTest::mask btest = bol->as_Bool()->_test._test;
2012 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2013 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2014 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2015 best_bol = bol->as_Bool();
2016 best_btest = btest;
2017 }
2018 }
2019 }
2020
2021 Node* answer_if_true = NULL;
2022 Node* answer_if_false = NULL;
2023 switch (best_btest) {
2024 default:
2025 if (cmpxy == NULL)
2026 cmpxy = ideal_cmpxy;
2027 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2028 // and fall through:
2029 case BoolTest::lt: // x < y
2030 case BoolTest::le: // x <= y
2031 answer_if_true = (want_max ? yvalue : xvalue);
2032 answer_if_false = (want_max ? xvalue : yvalue);
2033 break;
2034 case BoolTest::gt: // x > y
2035 case BoolTest::ge: // x >= y
2036 answer_if_true = (want_max ? xvalue : yvalue);
2037 answer_if_false = (want_max ? yvalue : xvalue);
2038 break;
2039 }
2040
2041 jint hi, lo;
2042 if (want_max) {
2043 // We can sharpen the minimum.
2044 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2045 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2046 } else {
2047 // We can sharpen the maximum.
2048 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2049 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2050 }
2051
2052 // Use a flow-free graph structure, to avoid creating excess control edges
2053 // which could hinder other optimizations.
2054 // Since Math.min/max is often used with arraycopy, we want
2055 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2056 Node* cmov = CMoveNode::make(NULL, best_bol,
2057 answer_if_false, answer_if_true,
2058 TypeInt::make(lo, hi, widen));
2059
2060 return _gvn.transform(cmov);
2061
2062 /*
2063 // This is not as desirable as it may seem, since Min and Max
2064 // nodes do not have a full set of optimizations.
2065 // And they would interfere, anyway, with 'if' optimizations
2066 // and with CMoveI canonical forms.
2067 switch (id) {
2068 case vmIntrinsics::_min:
2069 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2070 case vmIntrinsics::_max:
2071 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2072 default:
2073 ShouldNotReachHere();
2074 }
2075 */
2076 }
2077
2078 inline int
2079 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2080 const TypePtr* base_type = TypePtr::NULL_PTR;
2081 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2082 if (base_type == NULL) {
2083 // Unknown type.
2084 return Type::AnyPtr;
2085 } else if (base_type == TypePtr::NULL_PTR) {
2086 // Since this is a NULL+long form, we have to switch to a rawptr.
2087 base = _gvn.transform(new CastX2PNode(offset));
2088 offset = MakeConX(0);
2089 return Type::RawPtr;
2090 } else if (base_type->base() == Type::RawPtr) {
2091 return Type::RawPtr;
2092 } else if (base_type->isa_oopptr()) {
2093 // Base is never null => always a heap address.
2094 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2095 return Type::OopPtr;
2096 }
2097 // Offset is small => always a heap address.
2098 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2099 if (offset_type != NULL &&
2100 base_type->offset() == 0 && // (should always be?)
2101 offset_type->_lo >= 0 &&
2102 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2103 return Type::OopPtr;
2104 } else if (type == T_OBJECT) {
2105 // off heap access to an oop doesn't make any sense. Has to be on
2106 // heap.
2107 return Type::OopPtr;
2108 }
2109 // Otherwise, it might either be oop+off or NULL+addr.
2110 return Type::AnyPtr;
2111 } else {
2112 // No information:
2113 return Type::AnyPtr;
2114 }
2115 }
2116
2117 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2118 Node* uncasted_base = base;
2119 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2120 if (kind == Type::RawPtr) {
2121 return basic_plus_adr(top(), uncasted_base, offset);
2122 } else if (kind == Type::AnyPtr) {
2123 assert(base == uncasted_base, "unexpected base change");
2124 if (can_cast) {
2125 if (!_gvn.type(base)->speculative_maybe_null() &&
2126 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2127 // According to profiling, this access is always on
2128 // heap. Casting the base to not null and thus avoiding membars
2129 // around the access should allow better optimizations
2130 Node* null_ctl = top();
2131 base = null_check_oop(base, &null_ctl, true, true, true);
2132 assert(null_ctl->is_top(), "no null control here");
2133 return basic_plus_adr(base, offset);
2134 } else if (_gvn.type(base)->speculative_always_null() &&
2135 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2136 // According to profiling, this access is always off
2137 // heap.
2138 base = null_assert(base);
2139 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2140 offset = MakeConX(0);
2141 return basic_plus_adr(top(), raw_base, offset);
2142 }
2143 }
2144 // We don't know if it's an on heap or off heap access. Fall back
2145 // to raw memory access.
2146 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2147 return basic_plus_adr(top(), raw, offset);
2148 } else {
2149 assert(base == uncasted_base, "unexpected base change");
2150 // We know it's an on heap access so base can't be null
2151 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2152 base = must_be_not_null(base, true);
2153 }
2154 return basic_plus_adr(base, offset);
2155 }
2156 }
2157
2158 //--------------------------inline_number_methods-----------------------------
2159 // inline int Integer.numberOfLeadingZeros(int)
2160 // inline int Long.numberOfLeadingZeros(long)
2161 //
2162 // inline int Integer.numberOfTrailingZeros(int)
2163 // inline int Long.numberOfTrailingZeros(long)
2164 //
2165 // inline int Integer.bitCount(int)
2166 // inline int Long.bitCount(long)
2167 //
2168 // inline char Character.reverseBytes(char)
2169 // inline short Short.reverseBytes(short)
2170 // inline int Integer.reverseBytes(int)
2171 // inline long Long.reverseBytes(long)
2172 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2173 Node* arg = argument(0);
2174 Node* n = NULL;
2175 switch (id) {
2176 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2177 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2178 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2179 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2180 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2181 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2182 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2183 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2184 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2185 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2186 default: fatal_unexpected_iid(id); break;
2187 }
2188 set_result(_gvn.transform(n));
2189 return true;
2190 }
2191
2192 //----------------------------inline_unsafe_access----------------------------
2193
2194 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2195 // Attempt to infer a sharper value type from the offset and base type.
2196 ciKlass* sharpened_klass = NULL;
2197 bool null_free = false;
2198
2199 // See if it is an instance field, with an object type.
2200 if (alias_type->field() != NULL) {
2201 if (alias_type->field()->type()->is_klass()) {
2202 sharpened_klass = alias_type->field()->type()->as_klass();
2203 null_free = alias_type->field()->is_null_free();
2204 }
2205 }
2206
2207 // See if it is a narrow oop array.
2208 if (adr_type->isa_aryptr()) {
2209 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2210 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2211 null_free = adr_type->is_aryptr()->is_null_free();
2212 if (elem_type != NULL) {
2213 sharpened_klass = elem_type->klass();
2214 }
2215 }
2216 }
2217
2218 // The sharpened class might be unloaded if there is no class loader
2219 // contraint in place.
2220 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2221 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2222 if (null_free) {
2223 tjp = tjp->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2224 }
2225
2226 #ifndef PRODUCT
2227 if (C->print_intrinsics() || C->print_inlining()) {
2228 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2229 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2230 }
2231 #endif
2232 // Sharpen the value type.
2233 return tjp;
2234 }
2235 return NULL;
2236 }
2237
2238 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2239 switch (kind) {
2240 case Relaxed:
2241 return MO_UNORDERED;
2242 case Opaque:
2243 return MO_RELAXED;
2244 case Acquire:
2245 return MO_ACQUIRE;
2246 case Release:
2247 return MO_RELEASE;
2248 case Volatile:
2249 return MO_SEQ_CST;
2250 default:
2251 ShouldNotReachHere();
2252 return 0;
2253 }
2254 }
2255
2256 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2257 if (callee()->is_static()) return false; // caller must have the capability!
2258 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2259 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2260 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2261 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2262
2263 if (is_reference_type(type)) {
2264 decorators |= ON_UNKNOWN_OOP_REF;
2265 }
2266
2267 if (unaligned) {
2268 decorators |= C2_UNALIGNED;
2269 }
2270
2271 #ifndef PRODUCT
2272 {
2273 ResourceMark rm;
2274 // Check the signatures.
2275 ciSignature* sig = callee()->signature();
2276 #ifdef ASSERT
2277 if (!is_store) {
2278 // Object getReference(Object base, int/long offset), etc.
2279 BasicType rtype = sig->return_type()->basic_type();
2280 assert(rtype == type || (rtype == T_OBJECT && type == T_INLINE_TYPE), "getter must return the expected value");
2281 assert(sig->count() == 2 || (type == T_INLINE_TYPE && sig->count() == 3), "oop getter has 2 or 3 arguments");
2282 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2283 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2284 } else {
2285 // void putReference(Object base, int/long offset, Object x), etc.
2286 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2287 assert(sig->count() == 3 || (type == T_INLINE_TYPE && sig->count() == 4), "oop putter has 3 arguments");
2288 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2289 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2290 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2291 assert(vtype == type || (type == T_INLINE_TYPE && vtype == T_OBJECT), "putter must accept the expected value");
2292 }
2293 #endif // ASSERT
2294 }
2295 #endif //PRODUCT
2296
2297 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2298
2299 Node* receiver = argument(0); // type: oop
2300
2301 // Build address expression.
2302 Node* heap_base_oop = top();
2303
2304 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2305 Node* base = argument(1); // type: oop
2306 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2307 Node* offset = argument(2); // type: long
2308 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2309 // to be plain byte offsets, which are also the same as those accepted
2310 // by oopDesc::field_addr.
2311 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2312 "fieldOffset must be byte-scaled");
2313
2314 ciInlineKlass* inline_klass = NULL;
2315 if (type == T_INLINE_TYPE) {
2316 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2317 if (cls == NULL || cls->const_oop() == NULL) {
2318 return false;
2319 }
2320 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2321 if (!mirror_type->is_inlinetype()) {
2322 return false;
2323 }
2324 inline_klass = mirror_type->as_inline_klass();
2325 }
2326
2327 if (base->is_InlineTypeBase()) {
2328 InlineTypeBaseNode* vt = base->as_InlineTypeBase();
2329 if (is_store) {
2330 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->isa_inlinetype() || !_gvn.type(vt)->is_inlinetype()->larval()) {
2331 return false;
2332 }
2333 base = vt->get_oop();
2334 } else {
2335 if (offset->is_Con()) {
2336 long off = find_long_con(offset, 0);
2337 ciInlineKlass* vk = vt->type()->inline_klass();
2338 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2339 return false;
2340 }
2341
2342 ciField* field = vk->get_non_flattened_field_by_offset(off);
2343 if (field != NULL) {
2344 BasicType bt = field->layout_type();
2345 if (bt == T_ARRAY || bt == T_NARROWOOP || (bt == T_INLINE_TYPE && !field->is_flattened())) {
2346 bt = T_OBJECT;
2347 }
2348 if (bt == type && (bt != T_INLINE_TYPE || field->type() == inline_klass)) {
2349 set_result(vt->field_value_by_offset(off, false));
2350 return true;
2351 }
2352 }
2353 }
2354 if (vt->is_InlineType()) {
2355 // Re-execute the unsafe access if allocation triggers deoptimization.
2356 PreserveReexecuteState preexecs(this);
2357 jvms()->set_should_reexecute(true);
2358 vt = vt->buffer(this);
2359 }
2360 base = vt->get_oop();
2361 }
2362 }
2363
2364 // 32-bit machines ignore the high half!
2365 offset = ConvL2X(offset);
2366
2367 // Save state and restore on bailout
2368 uint old_sp = sp();
2369 SafePointNode* old_map = clone_map();
2370
2371 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2372
2373 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2374 if (type != T_OBJECT && (inline_klass == NULL || !inline_klass->has_object_fields())) {
2375 decorators |= IN_NATIVE; // off-heap primitive access
2376 } else {
2377 set_map(old_map);
2378 set_sp(old_sp);
2379 return false; // off-heap oop accesses are not supported
2380 }
2381 } else {
2382 heap_base_oop = base; // on-heap or mixed access
2383 }
2384
2385 // Can base be NULL? Otherwise, always on-heap access.
2386 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2387
2388 if (!can_access_non_heap) {
2389 decorators |= IN_HEAP;
2390 }
2391
2392 Node* val = is_store ? argument(4 + (type == T_INLINE_TYPE ? 1 : 0)) : NULL;
2393
2394 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2395 if (adr_type == TypePtr::NULL_PTR) {
2396 set_map(old_map);
2397 set_sp(old_sp);
2398 return false; // off-heap access with zero address
2399 }
2400
2401 // Try to categorize the address.
2402 Compile::AliasType* alias_type = C->alias_type(adr_type);
2403 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2404
2405 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2406 alias_type->adr_type() == TypeAryPtr::RANGE) {
2407 set_map(old_map);
2408 set_sp(old_sp);
2409 return false; // not supported
2410 }
2411
2412 bool mismatched = false;
2413 BasicType bt = T_ILLEGAL;
2414 ciField* field = NULL;
2415 if (adr_type->isa_instptr()) {
2416 const TypeInstPtr* instptr = adr_type->is_instptr();
2417 ciInstanceKlass* k = instptr->klass()->as_instance_klass();
2418 int off = instptr->offset();
2419 if (instptr->const_oop() != NULL &&
2420 instptr->klass() == ciEnv::current()->Class_klass() &&
2421 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) {
2422 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2423 field = k->get_field_by_offset(off, true);
2424 } else {
2425 field = k->get_non_flattened_field_by_offset(off);
2426 }
2427 if (field != NULL) {
2428 bt = field->layout_type();
2429 }
2430 assert(bt == alias_type->basic_type() || bt == T_INLINE_TYPE, "should match");
2431 if (field != NULL && bt == T_INLINE_TYPE && !field->is_flattened()) {
2432 bt = T_OBJECT;
2433 }
2434 } else {
2435 bt = alias_type->basic_type();
2436 }
2437
2438 if (bt != T_ILLEGAL) {
2439 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2440 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2441 // Alias type doesn't differentiate between byte[] and boolean[]).
2442 // Use address type to get the element type.
2443 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2444 }
2445 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2446 // accessing an array field with getReference is not a mismatch
2447 bt = T_OBJECT;
2448 }
2449 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2450 // Don't intrinsify mismatched object accesses
2451 set_map(old_map);
2452 set_sp(old_sp);
2453 return false;
2454 }
2455 mismatched = (bt != type);
2456 } else if (alias_type->adr_type()->isa_oopptr()) {
2457 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2458 }
2459
2460 if (type == T_INLINE_TYPE) {
2461 if (adr_type->isa_instptr()) {
2462 if (field == NULL || field->type() != inline_klass) {
2463 mismatched = true;
2464 }
2465 } else if (adr_type->isa_aryptr()) {
2466 const Type* elem = adr_type->is_aryptr()->elem();
2467 if (!elem->isa_inlinetype()) {
2468 mismatched = true;
2469 } else if (elem->inline_klass() != inline_klass) {
2470 mismatched = true;
2471 }
2472 } else {
2473 mismatched = true;
2474 }
2475 if (is_store) {
2476 const Type* val_t = _gvn.type(val);
2477 if (!val_t->isa_inlinetype() || val_t->inline_klass() != inline_klass) {
2478 set_map(old_map);
2479 set_sp(old_sp);
2480 return false;
2481 }
2482 }
2483 }
2484
2485 old_map->destruct(&_gvn);
2486 assert(!mismatched || type == T_INLINE_TYPE || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2487
2488 if (mismatched) {
2489 decorators |= C2_MISMATCHED;
2490 }
2491
2492 // First guess at the value type.
2493 const Type *value_type = Type::get_const_basic_type(type);
2494
2495 // Figure out the memory ordering.
2496 decorators |= mo_decorator_for_access_kind(kind);
2497
2498 if (!is_store) {
2499 if (type == T_OBJECT) {
2500 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2501 if (tjp != NULL) {
2502 value_type = tjp;
2503 }
2504 } else if (type == T_INLINE_TYPE) {
2505 value_type = NULL;
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
2522 if (heap_base_oop != top() && field != NULL && field->is_constant() && !field->is_flattened() && !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 if (type == T_INLINE_TYPE) {
2529 if (adr_type->isa_instptr() && !mismatched) {
2530 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2531 int offset = adr_type->is_instptr()->offset();
2532 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, base, holder, offset, decorators);
2533 } else {
2534 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, adr, NULL, 0, decorators);
2535 }
2536 } else {
2537 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2538 const TypeOopPtr* ptr = value_type->make_oopptr();
2539 if (ptr != NULL && ptr->is_inlinetypeptr()) {
2540 // Load a non-flattened inline type from memory
2541 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2542 }
2543 }
2544 // Normalize the value returned by getBoolean in the following cases
2545 if (type == T_BOOLEAN &&
2546 (mismatched ||
2547 heap_base_oop == top() || // - heap_base_oop is NULL or
2548 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2549 // and the unsafe access is made to large offset
2550 // (i.e., larger than the maximum offset necessary for any
2551 // field access)
2552 ) {
2553 IdealKit ideal = IdealKit(this);
2554 #define __ ideal.
2555 IdealVariable normalized_result(ideal);
2556 __ declarations_done();
2557 __ set(normalized_result, p);
2558 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2559 __ set(normalized_result, ideal.ConI(1));
2560 ideal.end_if();
2561 final_sync(ideal);
2562 p = __ value(normalized_result);
2563 #undef __
2564 }
2565 }
2566 if (type == T_ADDRESS) {
2567 p = gvn().transform(new CastP2XNode(NULL, p));
2568 p = ConvX2UL(p);
2569 }
2570 // The load node has the control of the preceding MemBarCPUOrder. All
2571 // following nodes will have the control of the MemBarCPUOrder inserted at
2572 // the end of this method. So, pushing the load onto the stack at a later
2573 // point is fine.
2574 set_result(p);
2575 } else {
2576 if (bt == T_ADDRESS) {
2577 // Repackage the long as a pointer.
2578 val = ConvL2X(val);
2579 val = gvn().transform(new CastX2PNode(val));
2580 }
2581 if (type == T_INLINE_TYPE) {
2582 if (adr_type->isa_instptr() && !mismatched) {
2583 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2584 int offset = adr_type->is_instptr()->offset();
2585 val->as_InlineTypeBase()->store_flattened(this, base, base, holder, offset, decorators);
2586 } else {
2587 val->as_InlineTypeBase()->store_flattened(this, base, adr, NULL, 0, decorators);
2588 }
2589 } else {
2590 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2591 }
2592 }
2593
2594 if (argument(1)->is_InlineType() && is_store) {
2595 Node* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(base)->inline_klass());
2596 value = value->as_InlineType()->make_larval(this, false);
2597 replace_in_map(argument(1), value);
2598 }
2599
2600 return true;
2601 }
2602
2603 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2604 Node* receiver = argument(0);
2605 Node* value = argument(1);
2606 if (!value->is_InlineType()) {
2607 return false;
2608 }
2609
2610 receiver = null_check(receiver);
2611 if (stopped()) {
2612 return true;
2613 }
2614
2615 set_result(value->as_InlineType()->make_larval(this, true));
2616 return true;
2617 }
2618
2619 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2620 Node* receiver = argument(0);
2621 Node* buffer = argument(1);
2622 if (!buffer->is_InlineType()) {
2623 return false;
2624 }
2625 InlineTypeNode* vt = buffer->as_InlineType();
2626 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_inlinetype()->larval()) {
2627 return false;
2628 }
2629
2630 receiver = null_check(receiver);
2631 if (stopped()) {
2632 return true;
2633 }
2634
2635 set_result(vt->finish_larval(this));
2636 return true;
2637 }
2638
2639 //----------------------------inline_unsafe_load_store----------------------------
2640 // This method serves a couple of different customers (depending on LoadStoreKind):
2641 //
2642 // LS_cmp_swap:
2643 //
2644 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2645 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2646 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2647 //
2648 // LS_cmp_swap_weak:
2649 //
2650 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2651 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2652 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2653 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2654 //
2655 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2656 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2657 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2658 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2659 //
2660 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2661 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2662 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2663 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2664 //
2665 // LS_cmp_exchange:
2666 //
2667 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2668 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2669 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2670 //
2671 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2672 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2673 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2674 //
2675 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2676 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2677 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2678 //
2679 // LS_get_add:
2680 //
2681 // int getAndAddInt( Object o, long offset, int delta)
2682 // long getAndAddLong(Object o, long offset, long delta)
2683 //
2684 // LS_get_set:
2685 //
2686 // int getAndSet(Object o, long offset, int newValue)
2687 // long getAndSet(Object o, long offset, long newValue)
2688 // Object getAndSet(Object o, long offset, Object newValue)
2689 //
2690 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2691 // This basic scheme here is the same as inline_unsafe_access, but
2692 // differs in enough details that combining them would make the code
2693 // overly confusing. (This is a true fact! I originally combined
2694 // them, but even I was confused by it!) As much code/comments as
2695 // possible are retained from inline_unsafe_access though to make
2696 // the correspondences clearer. - dl
2697
2698 if (callee()->is_static()) return false; // caller must have the capability!
2699
2700 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2701 decorators |= mo_decorator_for_access_kind(access_kind);
2702
2703 #ifndef PRODUCT
2704 BasicType rtype;
2705 {
2706 ResourceMark rm;
2707 // Check the signatures.
2708 ciSignature* sig = callee()->signature();
2709 rtype = sig->return_type()->basic_type();
2710 switch(kind) {
2711 case LS_get_add:
2712 case LS_get_set: {
2713 // Check the signatures.
2714 #ifdef ASSERT
2715 assert(rtype == type, "get and set must return the expected type");
2716 assert(sig->count() == 3, "get and set has 3 arguments");
2717 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2718 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2719 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2720 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2721 #endif // ASSERT
2722 break;
2723 }
2724 case LS_cmp_swap:
2725 case LS_cmp_swap_weak: {
2726 // Check the signatures.
2727 #ifdef ASSERT
2728 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2729 assert(sig->count() == 4, "CAS has 4 arguments");
2730 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2731 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2732 #endif // ASSERT
2733 break;
2734 }
2735 case LS_cmp_exchange: {
2736 // Check the signatures.
2737 #ifdef ASSERT
2738 assert(rtype == type, "CAS must return the expected type");
2739 assert(sig->count() == 4, "CAS has 4 arguments");
2740 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2741 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2742 #endif // ASSERT
2743 break;
2744 }
2745 default:
2746 ShouldNotReachHere();
2747 }
2748 }
2749 #endif //PRODUCT
2750
2751 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2752
2753 // Get arguments:
2754 Node* receiver = NULL;
2755 Node* base = NULL;
2756 Node* offset = NULL;
2757 Node* oldval = NULL;
2758 Node* newval = NULL;
2759 switch(kind) {
2760 case LS_cmp_swap:
2761 case LS_cmp_swap_weak:
2762 case LS_cmp_exchange: {
2763 const bool two_slot_type = type2size[type] == 2;
2764 receiver = argument(0); // type: oop
2765 base = argument(1); // type: oop
2766 offset = argument(2); // type: long
2767 oldval = argument(4); // type: oop, int, or long
2768 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2769 break;
2770 }
2771 case LS_get_add:
2772 case LS_get_set: {
2773 receiver = argument(0); // type: oop
2774 base = argument(1); // type: oop
2775 offset = argument(2); // type: long
2776 oldval = NULL;
2777 newval = argument(4); // type: oop, int, or long
2778 break;
2779 }
2780 default:
2781 ShouldNotReachHere();
2782 }
2783
2784 // Build field offset expression.
2785 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2786 // to be plain byte offsets, which are also the same as those accepted
2787 // by oopDesc::field_addr.
2788 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2789 // 32-bit machines ignore the high half of long offsets
2790 offset = ConvL2X(offset);
2791 // Save state and restore on bailout
2792 uint old_sp = sp();
2793 SafePointNode* old_map = clone_map();
2794 Node* adr = make_unsafe_address(base, offset,type, false);
2795 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2796
2797 Compile::AliasType* alias_type = C->alias_type(adr_type);
2798 BasicType bt = alias_type->basic_type();
2799 if (bt != T_ILLEGAL &&
2800 (is_reference_type(bt) != (type == T_OBJECT))) {
2801 // Don't intrinsify mismatched object accesses.
2802 set_map(old_map);
2803 set_sp(old_sp);
2804 return false;
2805 }
2806
2807 old_map->destruct(&_gvn);
2808
2809 // For CAS, unlike inline_unsafe_access, there seems no point in
2810 // trying to refine types. Just use the coarse types here.
2811 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2812 const Type *value_type = Type::get_const_basic_type(type);
2813
2814 switch (kind) {
2815 case LS_get_set:
2816 case LS_cmp_exchange: {
2817 if (type == T_OBJECT) {
2818 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2819 if (tjp != NULL) {
2820 value_type = tjp;
2821 }
2822 }
2823 break;
2824 }
2825 case LS_cmp_swap:
2826 case LS_cmp_swap_weak:
2827 case LS_get_add:
2828 break;
2829 default:
2830 ShouldNotReachHere();
2831 }
2832
2833 // Null check receiver.
2834 receiver = null_check(receiver);
2835 if (stopped()) {
2836 return true;
2837 }
2838
2839 int alias_idx = C->get_alias_index(adr_type);
2840
2841 if (is_reference_type(type)) {
2842 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2843
2844 if (oldval != NULL && oldval->is_InlineType()) {
2845 // Re-execute the unsafe access if allocation triggers deoptimization.
2846 PreserveReexecuteState preexecs(this);
2847 jvms()->set_should_reexecute(true);
2848 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2849 }
2850 if (newval != NULL && newval->is_InlineType()) {
2851 // Re-execute the unsafe access if allocation triggers deoptimization.
2852 PreserveReexecuteState preexecs(this);
2853 jvms()->set_should_reexecute(true);
2854 newval = newval->as_InlineType()->buffer(this)->get_oop();
2855 }
2856
2857 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2858 // could be delayed during Parse (for example, in adjust_map_after_if()).
2859 // Execute transformation here to avoid barrier generation in such case.
2860 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2861 newval = _gvn.makecon(TypePtr::NULL_PTR);
2862
2863 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2864 // Refine the value to a null constant, when it is known to be null
2865 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2866 }
2867 }
2868
2869 Node* result = NULL;
2870 switch (kind) {
2871 case LS_cmp_exchange: {
2872 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2873 oldval, newval, value_type, type, decorators);
2874 break;
2875 }
2876 case LS_cmp_swap_weak:
2877 decorators |= C2_WEAK_CMPXCHG;
2878 case LS_cmp_swap: {
2879 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2880 oldval, newval, value_type, type, decorators);
2881 break;
2882 }
2883 case LS_get_set: {
2884 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2885 newval, value_type, type, decorators);
2886 break;
2887 }
2888 case LS_get_add: {
2889 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2890 newval, value_type, type, decorators);
2891 break;
2892 }
2893 default:
2894 ShouldNotReachHere();
2895 }
2896
2897 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2898 set_result(result);
2899 return true;
2900 }
2901
2902 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2903 // Regardless of form, don't allow previous ld/st to move down,
2904 // then issue acquire, release, or volatile mem_bar.
2905 insert_mem_bar(Op_MemBarCPUOrder);
2906 switch(id) {
2907 case vmIntrinsics::_loadFence:
2908 insert_mem_bar(Op_LoadFence);
2909 return true;
2910 case vmIntrinsics::_storeFence:
2911 insert_mem_bar(Op_StoreFence);
2912 return true;
2913 case vmIntrinsics::_storeStoreFence:
2914 insert_mem_bar(Op_StoreStoreFence);
2915 return true;
2916 case vmIntrinsics::_fullFence:
2917 insert_mem_bar(Op_MemBarVolatile);
2918 return true;
2919 default:
2920 fatal_unexpected_iid(id);
2921 return false;
2922 }
2923 }
2924
2925 bool LibraryCallKit::inline_onspinwait() {
2926 insert_mem_bar(Op_OnSpinWait);
2927 return true;
2928 }
2929
2930 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2931 if (!kls->is_Con()) {
2932 return true;
2933 }
2934 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2935 if (klsptr == NULL) {
2936 return true;
2937 }
2938 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2939 // don't need a guard for a klass that is already initialized
2940 return !ik->is_initialized();
2941 }
2942
2943 //----------------------------inline_unsafe_writeback0-------------------------
2944 // public native void Unsafe.writeback0(long address)
2945 bool LibraryCallKit::inline_unsafe_writeback0() {
2946 if (!Matcher::has_match_rule(Op_CacheWB)) {
2947 return false;
2948 }
2949 #ifndef PRODUCT
2950 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2951 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2952 ciSignature* sig = callee()->signature();
2953 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2954 #endif
2955 null_check_receiver(); // null-check, then ignore
2956 Node *addr = argument(1);
2957 addr = new CastX2PNode(addr);
2958 addr = _gvn.transform(addr);
2959 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2960 flush = _gvn.transform(flush);
2961 set_memory(flush, TypeRawPtr::BOTTOM);
2962 return true;
2963 }
2964
2965 //----------------------------inline_unsafe_writeback0-------------------------
2966 // public native void Unsafe.writeback0(long address)
2967 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2968 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2969 return false;
2970 }
2971 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2972 return false;
2973 }
2974 #ifndef PRODUCT
2975 assert(Matcher::has_match_rule(Op_CacheWB),
2976 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2977 : "found match rule for CacheWBPostSync but not CacheWB"));
2978
2979 #endif
2980 null_check_receiver(); // null-check, then ignore
2981 Node *sync;
2982 if (is_pre) {
2983 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2984 } else {
2985 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2986 }
2987 sync = _gvn.transform(sync);
2988 set_memory(sync, TypeRawPtr::BOTTOM);
2989 return true;
2990 }
2991
2992 //----------------------------inline_unsafe_allocate---------------------------
2993 // public native Object Unsafe.allocateInstance(Class<?> cls);
2994 bool LibraryCallKit::inline_unsafe_allocate() {
2995 if (callee()->is_static()) return false; // caller must have the capability!
2996
2997 null_check_receiver(); // null-check, then ignore
2998 Node* cls = null_check(argument(1));
2999 if (stopped()) return true;
3000
3001 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3002 kls = null_check(kls);
3003 if (stopped()) return true; // argument was like int.class
3004
3005 Node* test = NULL;
3006 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3007 // Note: The argument might still be an illegal value like
3008 // Serializable.class or Object[].class. The runtime will handle it.
3009 // But we must make an explicit check for initialization.
3010 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3011 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3012 // can generate code to load it as unsigned byte.
3013 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3014 Node* bits = intcon(InstanceKlass::fully_initialized);
3015 test = _gvn.transform(new SubINode(inst, bits));
3016 // The 'test' is non-zero if we need to take a slow path.
3017 }
3018 Node* obj = NULL;
3019 ciKlass* klass = _gvn.type(kls)->is_klassptr()->klass();
3020 if (klass->is_inlinetype()) {
3021 obj = InlineTypeNode::make_default(_gvn, klass->as_inline_klass());
3022 } else {
3023 obj = new_instance(kls, test);
3024 }
3025 set_result(obj);
3026 return true;
3027 }
3028
3029 //------------------------inline_native_time_funcs--------------
3030 // inline code for System.currentTimeMillis() and System.nanoTime()
3031 // these have the same type and signature
3032 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3033 const TypeFunc* tf = OptoRuntime::void_long_Type();
3034 const TypePtr* no_memory_effects = NULL;
3035 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3036 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3037 #ifdef ASSERT
3038 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3039 assert(value_top == top(), "second value must be top");
3040 #endif
3041 set_result(value);
3042 return true;
3043 }
3044
3045 #ifdef JFR_HAVE_INTRINSICS
3046
3047 /**
3048 * if oop->klass != null
3049 * // normal class
3050 * epoch = _epoch_state ? 2 : 1
3051 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3052 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3053 * }
3054 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3055 * else
3056 * // primitive class
3057 * if oop->array_klass != null
3058 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3059 * else
3060 * id = LAST_TYPE_ID + 1 // void class path
3061 * if (!signaled)
3062 * signaled = true
3063 */
3064 bool LibraryCallKit::inline_native_classID() {
3065 Node* cls = argument(0);
3066
3067 IdealKit ideal(this);
3068 #define __ ideal.
3069 IdealVariable result(ideal); __ declarations_done();
3070 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3071 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3072 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3073
3074
3075 __ if_then(kls, BoolTest::ne, null()); {
3076 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3077 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3078
3079 Node* epoch_address = makecon(TypeRawPtr::make(Jfr::epoch_address()));
3080 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3081 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3082 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3083 mask = _gvn.transform(new OrLNode(mask, epoch));
3084 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3085
3086 float unlikely = PROB_UNLIKELY(0.999);
3087 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3088 sync_kit(ideal);
3089 make_runtime_call(RC_LEAF,
3090 OptoRuntime::get_class_id_intrinsic_Type(),
3091 CAST_FROM_FN_PTR(address, Jfr::get_class_id_intrinsic),
3092 "get_class_id_intrinsic",
3093 TypePtr::BOTTOM,
3094 kls);
3095 ideal.sync_kit(this);
3096 } __ end_if();
3097
3098 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3099 } __ else_(); {
3100 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3101 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3102 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3103 __ if_then(array_kls, BoolTest::ne, null()); {
3104 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3105 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3106 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3107 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3108 } __ else_(); {
3109 // void class case
3110 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3111 } __ end_if();
3112
3113 Node* signaled_flag_address = makecon(TypeRawPtr::make(Jfr::signal_address()));
3114 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3115 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3116 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3117 } __ end_if();
3118 } __ end_if();
3119
3120 final_sync(ideal);
3121 set_result(ideal.value(result));
3122 #undef __
3123 return true;
3124 }
3125
3126 bool LibraryCallKit::inline_native_getEventWriter() {
3127 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3128
3129 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3130 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3131
3132 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3133
3134 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3135 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3136
3137 IfNode* iff_jobj_null =
3138 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3139
3140 enum { _normal_path = 1,
3141 _null_path = 2,
3142 PATH_LIMIT };
3143
3144 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3145 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3146
3147 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3148 result_rgn->init_req(_null_path, jobj_is_null);
3149 result_val->init_req(_null_path, null());
3150
3151 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3152 set_control(jobj_is_not_null);
3153 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3154 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3155 result_rgn->init_req(_normal_path, control());
3156 result_val->init_req(_normal_path, res);
3157
3158 set_result(result_rgn, result_val);
3159
3160 return true;
3161 }
3162
3163 #endif // JFR_HAVE_INTRINSICS
3164
3165 //------------------------inline_native_currentThread------------------
3166 bool LibraryCallKit::inline_native_currentThread() {
3167 Node* junk = NULL;
3168 set_result(generate_current_thread(junk));
3169 return true;
3170 }
3171
3172 //-----------------------load_klass_from_mirror_common-------------------------
3173 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3174 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3175 // and branch to the given path on the region.
3176 // If never_see_null, take an uncommon trap on null, so we can optimistically
3177 // compile for the non-null case.
3178 // If the region is NULL, force never_see_null = true.
3179 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3180 bool never_see_null,
3181 RegionNode* region,
3182 int null_path,
3183 int offset) {
3184 if (region == NULL) never_see_null = true;
3185 Node* p = basic_plus_adr(mirror, offset);
3186 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3187 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3188 Node* null_ctl = top();
3189 kls = null_check_oop(kls, &null_ctl, never_see_null);
3190 if (region != NULL) {
3191 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3192 region->init_req(null_path, null_ctl);
3193 } else {
3194 assert(null_ctl == top(), "no loose ends");
3195 }
3196 return kls;
3197 }
3198
3199 //--------------------(inline_native_Class_query helpers)---------------------
3200 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3201 // Fall through if (mods & mask) == bits, take the guard otherwise.
3202 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3203 // Branch around if the given klass has the given modifier bit set.
3204 // Like generate_guard, adds a new path onto the region.
3205 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3206 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3207 Node* mask = intcon(modifier_mask);
3208 Node* bits = intcon(modifier_bits);
3209 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3210 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3211 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3212 return generate_fair_guard(bol, region);
3213 }
3214
3215 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3216 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3217 }
3218 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3219 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3220 }
3221
3222 //-------------------------inline_native_Class_query-------------------
3223 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3224 const Type* return_type = TypeInt::BOOL;
3225 Node* prim_return_value = top(); // what happens if it's a primitive class?
3226 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3227 bool expect_prim = false; // most of these guys expect to work on refs
3228
3229 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3230
3231 Node* mirror = argument(0);
3232 Node* obj = top();
3233
3234 switch (id) {
3235 case vmIntrinsics::_isInstance:
3236 // nothing is an instance of a primitive type
3237 prim_return_value = intcon(0);
3238 obj = argument(1);
3239 break;
3240 case vmIntrinsics::_getModifiers:
3241 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3242 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3243 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3244 break;
3245 case vmIntrinsics::_isInterface:
3246 prim_return_value = intcon(0);
3247 break;
3248 case vmIntrinsics::_isArray:
3249 prim_return_value = intcon(0);
3250 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3251 break;
3252 case vmIntrinsics::_isPrimitive:
3253 prim_return_value = intcon(1);
3254 expect_prim = true; // obviously
3255 break;
3256 case vmIntrinsics::_isHidden:
3257 prim_return_value = intcon(0);
3258 break;
3259 case vmIntrinsics::_getSuperclass:
3260 prim_return_value = null();
3261 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3262 break;
3263 case vmIntrinsics::_getClassAccessFlags:
3264 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3265 return_type = TypeInt::INT; // not bool! 6297094
3266 break;
3267 default:
3268 fatal_unexpected_iid(id);
3269 break;
3270 }
3271
3272 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3273 if (mirror_con == NULL) return false; // cannot happen?
3274
3275 #ifndef PRODUCT
3276 if (C->print_intrinsics() || C->print_inlining()) {
3277 ciType* k = mirror_con->java_mirror_type();
3278 if (k) {
3279 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3280 k->print_name();
3281 tty->cr();
3282 }
3283 }
3284 #endif
3285
3286 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3287 RegionNode* region = new RegionNode(PATH_LIMIT);
3288 record_for_igvn(region);
3289 PhiNode* phi = new PhiNode(region, return_type);
3290
3291 // The mirror will never be null of Reflection.getClassAccessFlags, however
3292 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3293 // if it is. See bug 4774291.
3294
3295 // For Reflection.getClassAccessFlags(), the null check occurs in
3296 // the wrong place; see inline_unsafe_access(), above, for a similar
3297 // situation.
3298 mirror = null_check(mirror);
3299 // If mirror or obj is dead, only null-path is taken.
3300 if (stopped()) return true;
3301
3302 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3303
3304 // Now load the mirror's klass metaobject, and null-check it.
3305 // Side-effects region with the control path if the klass is null.
3306 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3307 // If kls is null, we have a primitive mirror.
3308 phi->init_req(_prim_path, prim_return_value);
3309 if (stopped()) { set_result(region, phi); return true; }
3310 bool safe_for_replace = (region->in(_prim_path) == top());
3311
3312 Node* p; // handy temp
3313 Node* null_ctl;
3314
3315 // Now that we have the non-null klass, we can perform the real query.
3316 // For constant classes, the query will constant-fold in LoadNode::Value.
3317 Node* query_value = top();
3318 switch (id) {
3319 case vmIntrinsics::_isInstance:
3320 // nothing is an instance of a primitive type
3321 query_value = gen_instanceof(obj, kls, safe_for_replace);
3322 break;
3323
3324 case vmIntrinsics::_getModifiers:
3325 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3326 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3327 break;
3328
3329 case vmIntrinsics::_isInterface:
3330 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3331 if (generate_interface_guard(kls, region) != NULL)
3332 // A guard was added. If the guard is taken, it was an interface.
3333 phi->add_req(intcon(1));
3334 // If we fall through, it's a plain class.
3335 query_value = intcon(0);
3336 break;
3337
3338 case vmIntrinsics::_isArray:
3339 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3340 if (generate_array_guard(kls, region) != NULL)
3341 // A guard was added. If the guard is taken, it was an array.
3342 phi->add_req(intcon(1));
3343 // If we fall through, it's a plain class.
3344 query_value = intcon(0);
3345 break;
3346
3347 case vmIntrinsics::_isPrimitive:
3348 query_value = intcon(0); // "normal" path produces false
3349 break;
3350
3351 case vmIntrinsics::_isHidden:
3352 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3353 if (generate_hidden_class_guard(kls, region) != NULL)
3354 // A guard was added. If the guard is taken, it was an hidden class.
3355 phi->add_req(intcon(1));
3356 // If we fall through, it's a plain class.
3357 query_value = intcon(0);
3358 break;
3359
3360
3361 case vmIntrinsics::_getSuperclass:
3362 // The rules here are somewhat unfortunate, but we can still do better
3363 // with random logic than with a JNI call.
3364 // Interfaces store null or Object as _super, but must report null.
3365 // Arrays store an intermediate super as _super, but must report Object.
3366 // Other types can report the actual _super.
3367 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3368 if (generate_interface_guard(kls, region) != NULL)
3369 // A guard was added. If the guard is taken, it was an interface.
3370 phi->add_req(null());
3371 if (generate_array_guard(kls, region) != NULL)
3372 // A guard was added. If the guard is taken, it was an array.
3373 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3374 // If we fall through, it's a plain class. Get its _super.
3375 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3376 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3377 null_ctl = top();
3378 kls = null_check_oop(kls, &null_ctl);
3379 if (null_ctl != top()) {
3380 // If the guard is taken, Object.superClass is null (both klass and mirror).
3381 region->add_req(null_ctl);
3382 phi ->add_req(null());
3383 }
3384 if (!stopped()) {
3385 query_value = load_mirror_from_klass(kls);
3386 }
3387 break;
3388
3389 case vmIntrinsics::_getClassAccessFlags:
3390 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3391 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3392 break;
3393
3394 default:
3395 fatal_unexpected_iid(id);
3396 break;
3397 }
3398
3399 // Fall-through is the normal case of a query to a real class.
3400 phi->init_req(1, query_value);
3401 region->init_req(1, control());
3402
3403 C->set_has_split_ifs(true); // Has chance for split-if optimization
3404 set_result(region, phi);
3405 return true;
3406 }
3407
3408 //-------------------------inline_primitive_Class_conversion-------------------
3409 // public Class<T> java.lang.Class.asPrimaryType();
3410 // public Class<T> java.lang.Class.asValueType()
3411 bool LibraryCallKit::inline_primitive_Class_conversion(vmIntrinsics::ID id) {
3412 Node* mirror = argument(0); // Receiver Class
3413 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3414 if (mirror_con == NULL) {
3415 return false;
3416 }
3417
3418 bool is_val_mirror = true;
3419 ciType* tm = mirror_con->java_mirror_type(&is_val_mirror);
3420 if (tm != NULL) {
3421 Node* result = mirror;
3422 if (id == vmIntrinsics::_asPrimaryType && is_val_mirror) {
3423 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->ref_mirror()));
3424 } else if (id == vmIntrinsics::_asValueType) {
3425 if (!tm->is_inlinetype()) {
3426 return false; // Throw UnsupportedOperationException
3427 } else if (!is_val_mirror) {
3428 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->val_mirror()));
3429 }
3430 }
3431 set_result(result);
3432 return true;
3433 }
3434 return false;
3435 }
3436
3437 //-------------------------inline_Class_cast-------------------
3438 bool LibraryCallKit::inline_Class_cast() {
3439 Node* mirror = argument(0); // Class
3440 Node* obj = argument(1);
3441 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3442 if (mirror_con == NULL) {
3443 return false; // dead path (mirror->is_top()).
3444 }
3445 if (obj == NULL || obj->is_top()) {
3446 return false; // dead path
3447 }
3448 ciKlass* obj_klass = NULL;
3449 const Type* obj_t = _gvn.type(obj);
3450 if (obj->is_InlineType()) {
3451 obj_klass = obj_t->inline_klass();
3452 } else if (obj_t->isa_oopptr()) {
3453 obj_klass = obj_t->is_oopptr()->klass();
3454 }
3455
3456 // First, see if Class.cast() can be folded statically.
3457 // java_mirror_type() returns non-null for compile-time Class constants.
3458 bool requires_null_check = false;
3459 ciType* tm = mirror_con->java_mirror_type(&requires_null_check);
3460 // Check for null if casting to QMyValue
3461 requires_null_check &= !obj->is_InlineType();
3462 if (tm != NULL && tm->is_klass() && obj_klass != NULL) {
3463 if (!obj_klass->is_loaded()) {
3464 // Don't use intrinsic when class is not loaded.
3465 return false;
3466 } else {
3467 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass);
3468 if (static_res == Compile::SSC_always_true) {
3469 // isInstance() is true - fold the code.
3470 if (requires_null_check) {
3471 obj = null_check(obj);
3472 }
3473 set_result(obj);
3474 return true;
3475 } else if (static_res == Compile::SSC_always_false) {
3476 // Don't use intrinsic, have to throw ClassCastException.
3477 // If the reference is null, the non-intrinsic bytecode will
3478 // be optimized appropriately.
3479 return false;
3480 }
3481 }
3482 }
3483
3484 // Bailout intrinsic and do normal inlining if exception path is frequent.
3485 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3486 return false;
3487 }
3488
3489 // Generate dynamic checks.
3490 // Class.cast() is java implementation of _checkcast bytecode.
3491 // Do checkcast (Parse::do_checkcast()) optimizations here.
3492
3493 if (requires_null_check) {
3494 obj = null_check(obj);
3495 }
3496 mirror = null_check(mirror);
3497 // If mirror is dead, only null-path is taken.
3498 if (stopped()) {
3499 return true;
3500 }
3501
3502 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3503 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
3504 RegionNode* region = new RegionNode(PATH_LIMIT);
3505 record_for_igvn(region);
3506
3507 // Now load the mirror's klass metaobject, and null-check it.
3508 // If kls is null, we have a primitive mirror and
3509 // nothing is an instance of a primitive type.
3510 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3511
3512 Node* res = top();
3513 if (!stopped()) {
3514 if (EnableValhalla && !obj->is_InlineType() && !requires_null_check) {
3515 // Check if we are casting to QMyValue
3516 Node* ctrl_val_mirror = generate_fair_guard(is_val_mirror(mirror), NULL);
3517 if (ctrl_val_mirror != NULL) {
3518 RegionNode* r = new RegionNode(3);
3519 record_for_igvn(r);
3520 r->init_req(1, control());
3521
3522 // Casting to QMyValue, check for null
3523 set_control(ctrl_val_mirror);
3524 { // PreserveJVMState because null check replaces obj in map
3525 PreserveJVMState pjvms(this);
3526 Node* null_ctr = top();
3527 null_check_oop(obj, &null_ctr);
3528 region->init_req(_npe_path, null_ctr);
3529 r->init_req(2, control());
3530 }
3531 set_control(_gvn.transform(r));
3532 }
3533 }
3534
3535 Node* bad_type_ctrl = top();
3536 // Do checkcast optimizations.
3537 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3538 region->init_req(_bad_type_path, bad_type_ctrl);
3539 }
3540 if (region->in(_prim_path) != top() ||
3541 region->in(_bad_type_path) != top() ||
3542 region->in(_npe_path) != top()) {
3543 // Let Interpreter throw ClassCastException.
3544 PreserveJVMState pjvms(this);
3545 set_control(_gvn.transform(region));
3546 uncommon_trap(Deoptimization::Reason_intrinsic,
3547 Deoptimization::Action_maybe_recompile);
3548 }
3549 if (!stopped()) {
3550 set_result(res);
3551 }
3552 return true;
3553 }
3554
3555
3556 //--------------------------inline_native_subtype_check------------------------
3557 // This intrinsic takes the JNI calls out of the heart of
3558 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3559 bool LibraryCallKit::inline_native_subtype_check() {
3560 // Pull both arguments off the stack.
3561 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3562 args[0] = argument(0);
3563 args[1] = argument(1);
3564 Node* klasses[2]; // corresponding Klasses: superk, subk
3565 klasses[0] = klasses[1] = top();
3566
3567 enum {
3568 // A full decision tree on {superc is prim, subc is prim}:
3569 _prim_0_path = 1, // {P,N} => false
3570 // {P,P} & superc!=subc => false
3571 _prim_same_path, // {P,P} & superc==subc => true
3572 _prim_1_path, // {N,P} => false
3573 _ref_subtype_path, // {N,N} & subtype check wins => true
3574 _both_ref_path, // {N,N} & subtype check loses => false
3575 PATH_LIMIT
3576 };
3577
3578 RegionNode* region = new RegionNode(PATH_LIMIT);
3579 RegionNode* prim_region = new RegionNode(2);
3580 Node* phi = new PhiNode(region, TypeInt::BOOL);
3581 record_for_igvn(region);
3582 record_for_igvn(prim_region);
3583
3584 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3585 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3586 int class_klass_offset = java_lang_Class::klass_offset();
3587
3588 // First null-check both mirrors and load each mirror's klass metaobject.
3589 int which_arg;
3590 for (which_arg = 0; which_arg <= 1; which_arg++) {
3591 Node* arg = args[which_arg];
3592 arg = null_check(arg);
3593 if (stopped()) break;
3594 args[which_arg] = arg;
3595
3596 Node* p = basic_plus_adr(arg, class_klass_offset);
3597 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3598 klasses[which_arg] = _gvn.transform(kls);
3599 }
3600
3601 // Having loaded both klasses, test each for null.
3602 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3603 for (which_arg = 0; which_arg <= 1; which_arg++) {
3604 Node* kls = klasses[which_arg];
3605 Node* null_ctl = top();
3606 kls = null_check_oop(kls, &null_ctl, never_see_null);
3607 if (which_arg == 0) {
3608 prim_region->init_req(1, null_ctl);
3609 } else {
3610 region->init_req(_prim_1_path, null_ctl);
3611 }
3612 if (stopped()) break;
3613 klasses[which_arg] = kls;
3614 }
3615
3616 if (!stopped()) {
3617 // now we have two reference types, in klasses[0..1]
3618 Node* subk = klasses[1]; // the argument to isAssignableFrom
3619 Node* superk = klasses[0]; // the receiver
3620 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3621 // If superc is an inline mirror, we also need to check if superc == subc because LMyValue
3622 // is not a subtype of QMyValue but due to subk == superk the subtype check will pass.
3623 generate_fair_guard(is_val_mirror(args[0]), prim_region);
3624 // now we have a successful reference subtype check
3625 region->set_req(_ref_subtype_path, control());
3626 }
3627
3628 // If both operands are primitive (both klasses null), then
3629 // we must return true when they are identical primitives.
3630 // It is convenient to test this after the first null klass check.
3631 // This path is also used if superc is a value mirror.
3632 set_control(_gvn.transform(prim_region));
3633 if (!stopped()) {
3634 // Since superc is primitive, make a guard for the superc==subc case.
3635 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3636 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3637 generate_fair_guard(bol_eq, region);
3638 if (region->req() == PATH_LIMIT+1) {
3639 // A guard was added. If the added guard is taken, superc==subc.
3640 region->swap_edges(PATH_LIMIT, _prim_same_path);
3641 region->del_req(PATH_LIMIT);
3642 }
3643 region->set_req(_prim_0_path, control()); // Not equal after all.
3644 }
3645
3646 // these are the only paths that produce 'true':
3647 phi->set_req(_prim_same_path, intcon(1));
3648 phi->set_req(_ref_subtype_path, intcon(1));
3649
3650 // pull together the cases:
3651 assert(region->req() == PATH_LIMIT, "sane region");
3652 for (uint i = 1; i < region->req(); i++) {
3653 Node* ctl = region->in(i);
3654 if (ctl == NULL || ctl == top()) {
3655 region->set_req(i, top());
3656 phi ->set_req(i, top());
3657 } else if (phi->in(i) == NULL) {
3658 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3659 }
3660 }
3661
3662 set_control(_gvn.transform(region));
3663 set_result(_gvn.transform(phi));
3664 return true;
3665 }
3666
3667 //---------------------generate_array_guard_common------------------------
3668 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
3669
3670 if (stopped()) {
3671 return NULL;
3672 }
3673
3674 // Like generate_guard, adds a new path onto the region.
3675 jint layout_con = 0;
3676 Node* layout_val = get_layout_helper(kls, layout_con);
3677 if (layout_val == NULL) {
3678 bool query = 0;
3679 switch(kind) {
3680 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
3681 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
3682 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
3683 case FlatArray: query = Klass::layout_helper_is_flatArray(layout_con); break;
3684 case NonFlatArray: query = !Klass::layout_helper_is_flatArray(layout_con); break;
3685 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
3686 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
3687 default:
3688 ShouldNotReachHere();
3689 }
3690 if (!query) {
3691 return NULL; // never a branch
3692 } else { // always a branch
3693 Node* always_branch = control();
3694 if (region != NULL)
3695 region->add_req(always_branch);
3696 set_control(top());
3697 return always_branch;
3698 }
3699 }
3700 unsigned int value = 0;
3701 BoolTest::mask btest = BoolTest::illegal;
3702 switch(kind) {
3703 case ObjectArray:
3704 case NonObjectArray: {
3705 value = Klass::_lh_array_tag_obj_value;
3706 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3707 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
3708 break;
3709 }
3710 case TypeArray: {
3711 value = Klass::_lh_array_tag_type_value;
3712 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3713 btest = BoolTest::eq;
3714 break;
3715 }
3716 case FlatArray:
3717 case NonFlatArray: {
3718 value = 0;
3719 layout_val = _gvn.transform(new AndINode(layout_val, intcon(Klass::_lh_array_tag_flat_value_bit_inplace)));
3720 btest = (kind == FlatArray) ? BoolTest::ne : BoolTest::eq;
3721 break;
3722 }
3723 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
3724 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
3725 default:
3726 ShouldNotReachHere();
3727 }
3728 // Now test the correct condition.
3729 jint nval = (jint)value;
3730 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3731 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3732 return generate_fair_guard(bol, region);
3733 }
3734
3735
3736 //-----------------------inline_native_newArray--------------------------
3737 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
3738 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3739 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3740 Node* mirror;
3741 Node* count_val;
3742 if (uninitialized) {
3743 mirror = argument(1);
3744 count_val = argument(2);
3745 } else {
3746 mirror = argument(0);
3747 count_val = argument(1);
3748 }
3749
3750 mirror = null_check(mirror);
3751 // If mirror or obj is dead, only null-path is taken.
3752 if (stopped()) return true;
3753
3754 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3755 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3756 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3757 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3758 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3759
3760 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3761 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3762 result_reg, _slow_path);
3763 Node* normal_ctl = control();
3764 Node* no_array_ctl = result_reg->in(_slow_path);
3765
3766 // Generate code for the slow case. We make a call to newArray().
3767 set_control(no_array_ctl);
3768 if (!stopped()) {
3769 // Either the input type is void.class, or else the
3770 // array klass has not yet been cached. Either the
3771 // ensuing call will throw an exception, or else it
3772 // will cache the array klass for next time.
3773 PreserveJVMState pjvms(this);
3774 CallJavaNode* slow_call = NULL;
3775 if (uninitialized) {
3776 // Generate optimized virtual call (holder class 'Unsafe' is final)
3777 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3778 } else {
3779 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3780 }
3781 Node* slow_result = set_results_for_java_call(slow_call);
3782 // this->control() comes from set_results_for_java_call
3783 result_reg->set_req(_slow_path, control());
3784 result_val->set_req(_slow_path, slow_result);
3785 result_io ->set_req(_slow_path, i_o());
3786 result_mem->set_req(_slow_path, reset_memory());
3787 }
3788
3789 set_control(normal_ctl);
3790 if (!stopped()) {
3791 // Normal case: The array type has been cached in the java.lang.Class.
3792 // The following call works fine even if the array type is polymorphic.
3793 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3794 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3795 result_reg->init_req(_normal_path, control());
3796 result_val->init_req(_normal_path, obj);
3797 result_io ->init_req(_normal_path, i_o());
3798 result_mem->init_req(_normal_path, reset_memory());
3799
3800 if (uninitialized) {
3801 // Mark the allocation so that zeroing is skipped
3802 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3803 alloc->maybe_set_complete(&_gvn);
3804 }
3805 }
3806
3807 // Return the combined state.
3808 set_i_o( _gvn.transform(result_io) );
3809 set_all_memory( _gvn.transform(result_mem));
3810
3811 C->set_has_split_ifs(true); // Has chance for split-if optimization
3812 set_result(result_reg, result_val);
3813 return true;
3814 }
3815
3816 //----------------------inline_native_getLength--------------------------
3817 // public static native int java.lang.reflect.Array.getLength(Object array);
3818 bool LibraryCallKit::inline_native_getLength() {
3819 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3820
3821 Node* array = null_check(argument(0));
3822 // If array is dead, only null-path is taken.
3823 if (stopped()) return true;
3824
3825 // Deoptimize if it is a non-array.
3826 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3827
3828 if (non_array != NULL) {
3829 PreserveJVMState pjvms(this);
3830 set_control(non_array);
3831 uncommon_trap(Deoptimization::Reason_intrinsic,
3832 Deoptimization::Action_maybe_recompile);
3833 }
3834
3835 // If control is dead, only non-array-path is taken.
3836 if (stopped()) return true;
3837
3838 // The works fine even if the array type is polymorphic.
3839 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3840 Node* result = load_array_length(array);
3841
3842 C->set_has_split_ifs(true); // Has chance for split-if optimization
3843 set_result(result);
3844 return true;
3845 }
3846
3847 //------------------------inline_array_copyOf----------------------------
3848 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3849 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3850 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3851 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3852
3853 // Get the arguments.
3854 Node* original = argument(0);
3855 Node* start = is_copyOfRange? argument(1): intcon(0);
3856 Node* end = is_copyOfRange? argument(2): argument(1);
3857 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3858
3859 Node* newcopy = NULL;
3860
3861 // Set the original stack and the reexecute bit for the interpreter to reexecute
3862 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3863 { PreserveReexecuteState preexecs(this);
3864 jvms()->set_should_reexecute(true);
3865
3866 array_type_mirror = null_check(array_type_mirror);
3867 original = null_check(original);
3868
3869 // Check if a null path was taken unconditionally.
3870 if (stopped()) return true;
3871
3872 Node* orig_length = load_array_length(original);
3873
3874 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3875 klass_node = null_check(klass_node);
3876
3877 RegionNode* bailout = new RegionNode(1);
3878 record_for_igvn(bailout);
3879
3880 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3881 // Bail out if that is so.
3882 // Inline type array may have object field that would require a
3883 // write barrier. Conservatively, go to slow path.
3884 // TODO 8251971: Optimize for the case when flat src/dst are later found
3885 // to not contain oops (i.e., move this check to the macro expansion phase).
3886 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
3887 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
3888 ciKlass* klass = _gvn.type(klass_node)->is_klassptr()->klass();
3889 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
3890 // Can src array be flat and contain oops?
3891 (orig_t == NULL || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
3892 // Can dest array be flat and contain oops?
3893 klass->can_be_inline_array_klass() && (!klass->is_flat_array_klass() || klass->as_flat_array_klass()->element_klass()->as_inline_klass()->contains_oops());
3894 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
3895 if (not_objArray != NULL) {
3896 // Improve the klass node's type from the new optimistic assumption:
3897 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3898 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
3899 Node* cast = new CastPPNode(klass_node, akls);
3900 cast->init_req(0, control());
3901 klass_node = _gvn.transform(cast);
3902 }
3903
3904 Node* original_kls = load_object_klass(original);
3905 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3906 // loads/stores but it is legal only if we're sure the
3907 // Arrays.copyOf would succeed. So we need all input arguments
3908 // to the copyOf to be validated, including that the copy to the
3909 // new array won't trigger an ArrayStoreException. That subtype
3910 // check can be optimized if we know something on the type of
3911 // the input array from type speculation.
3912 if (_gvn.type(klass_node)->singleton() && !stopped()) {
3913 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass();
3914 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3915
3916 int test = C->static_subtype_check(superk, subk);
3917 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3918 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3919 if (t_original->speculative_type() != NULL) {
3920 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3921 original_kls = load_object_klass(original);
3922 }
3923 }
3924 }
3925
3926 // Bail out if either start or end is negative.
3927 generate_negative_guard(start, bailout, &start);
3928 generate_negative_guard(end, bailout, &end);
3929
3930 Node* length = end;
3931 if (_gvn.type(start) != TypeInt::ZERO) {
3932 length = _gvn.transform(new SubINode(end, start));
3933 }
3934
3935 // Bail out if length is negative.
3936 // Without this the new_array would throw
3937 // NegativeArraySizeException but IllegalArgumentException is what
3938 // should be thrown
3939 generate_negative_guard(length, bailout, &length);
3940
3941 // Handle inline type arrays
3942 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
3943 if (!stopped()) {
3944 orig_t = _gvn.type(original)->isa_aryptr();
3945 if (orig_t != NULL && orig_t->is_flat()) {
3946 // Src is flat, check that dest is flat as well
3947 if (exclude_flat) {
3948 // Dest can't be flat, bail out
3949 bailout->add_req(control());
3950 set_control(top());
3951 } else {
3952 generate_non_flatArray_guard(klass_node, bailout);
3953 }
3954 } else if (UseFlatArray && (orig_t == NULL || !orig_t->is_not_flat()) &&
3955 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
3956 ((!klass->is_flat_array_klass() && klass->can_be_inline_array_klass()) || !can_validate)) {
3957 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
3958 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
3959 generate_flatArray_guard(original_kls, bailout);
3960 if (orig_t != NULL) {
3961 orig_t = orig_t->cast_to_not_flat();
3962 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
3963 }
3964 }
3965 if (!can_validate) {
3966 // No validation. The subtype check emitted at macro expansion time will not go to the slow
3967 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
3968 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
3969 generate_fair_guard(null_free_array_test(klass_node), bailout);
3970 }
3971 }
3972
3973 if (bailout->req() > 1) {
3974 PreserveJVMState pjvms(this);
3975 set_control(_gvn.transform(bailout));
3976 uncommon_trap(Deoptimization::Reason_intrinsic,
3977 Deoptimization::Action_maybe_recompile);
3978 }
3979
3980 if (!stopped()) {
3981 // How many elements will we copy from the original?
3982 // The answer is MinI(orig_length - start, length).
3983 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3984 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3985
3986 // Generate a direct call to the right arraycopy function(s).
3987 // We know the copy is disjoint but we might not know if the
3988 // oop stores need checking.
3989 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3990 // This will fail a store-check if x contains any non-nulls.
3991
3992 bool validated = false;
3993 // Reason_class_check rather than Reason_intrinsic because we
3994 // want to intrinsify even if this traps.
3995 if (can_validate) {
3996 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3997
3998 if (not_subtype_ctrl != top()) {
3999 PreserveJVMState pjvms(this);
4000 set_control(not_subtype_ctrl);
4001 uncommon_trap(Deoptimization::Reason_class_check,
4002 Deoptimization::Action_make_not_entrant);
4003 assert(stopped(), "Should be stopped");
4004 }
4005 validated = true;
4006 }
4007
4008 if (!stopped()) {
4009 newcopy = new_array(klass_node, length, 0); // no arguments to push
4010
4011 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4012 original_kls, klass_node);
4013 if (!is_copyOfRange) {
4014 ac->set_copyof(validated);
4015 } else {
4016 ac->set_copyofrange(validated);
4017 }
4018 Node* n = _gvn.transform(ac);
4019 if (n == ac) {
4020 ac->connect_outputs(this);
4021 } else {
4022 assert(validated, "shouldn't transform if all arguments not validated");
4023 set_all_memory(n);
4024 }
4025 }
4026 }
4027 } // original reexecute is set back here
4028
4029 C->set_has_split_ifs(true); // Has chance for split-if optimization
4030 if (!stopped()) {
4031 set_result(newcopy);
4032 }
4033 return true;
4034 }
4035
4036
4037 //----------------------generate_virtual_guard---------------------------
4038 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4039 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4040 RegionNode* slow_region) {
4041 ciMethod* method = callee();
4042 int vtable_index = method->vtable_index();
4043 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4044 "bad index %d", vtable_index);
4045 // Get the Method* out of the appropriate vtable entry.
4046 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4047 vtable_index*vtableEntry::size_in_bytes() +
4048 vtableEntry::method_offset_in_bytes();
4049 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4050 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4051
4052 // Compare the target method with the expected method (e.g., Object.hashCode).
4053 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4054
4055 Node* native_call = makecon(native_call_addr);
4056 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4057 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4058
4059 return generate_slow_guard(test_native, slow_region);
4060 }
4061
4062 //-----------------------generate_method_call----------------------------
4063 // Use generate_method_call to make a slow-call to the real
4064 // method if the fast path fails. An alternative would be to
4065 // use a stub like OptoRuntime::slow_arraycopy_Java.
4066 // This only works for expanding the current library call,
4067 // not another intrinsic. (E.g., don't use this for making an
4068 // arraycopy call inside of the copyOf intrinsic.)
4069 CallJavaNode*
4070 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4071 // When compiling the intrinsic method itself, do not use this technique.
4072 guarantee(callee() != C->method(), "cannot make slow-call to self");
4073
4074 ciMethod* method = callee();
4075 // ensure the JVMS we have will be correct for this call
4076 guarantee(method_id == method->intrinsic_id(), "must match");
4077
4078 const TypeFunc* tf = TypeFunc::make(method);
4079 CallJavaNode* slow_call;
4080 if (is_static) {
4081 assert(!is_virtual, "");
4082 slow_call = new CallStaticJavaNode(C, tf,
4083 SharedRuntime::get_resolve_static_call_stub(), method);
4084 } else if (is_virtual) {
4085 null_check_receiver();
4086 int vtable_index = Method::invalid_vtable_index;
4087 if (UseInlineCaches) {
4088 // Suppress the vtable call
4089 } else {
4090 // hashCode and clone are not a miranda methods,
4091 // so the vtable index is fixed.
4092 // No need to use the linkResolver to get it.
4093 vtable_index = method->vtable_index();
4094 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4095 "bad index %d", vtable_index);
4096 }
4097 slow_call = new CallDynamicJavaNode(tf,
4098 SharedRuntime::get_resolve_virtual_call_stub(),
4099 method, vtable_index);
4100 } else { // neither virtual nor static: opt_virtual
4101 null_check_receiver();
4102 slow_call = new CallStaticJavaNode(C, tf,
4103 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4104 slow_call->set_optimized_virtual(true);
4105 }
4106 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4107 // To be able to issue a direct call (optimized virtual or virtual)
4108 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4109 // about the method being invoked should be attached to the call site to
4110 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4111 slow_call->set_override_symbolic_info(true);
4112 }
4113 set_arguments_for_java_call(slow_call);
4114 set_edges_for_java_call(slow_call);
4115 return slow_call;
4116 }
4117
4118
4119 /**
4120 * Build special case code for calls to hashCode on an object. This call may
4121 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4122 * slightly different code.
4123 */
4124 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4125 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4126 assert(!(is_virtual && is_static), "either virtual, special, or static");
4127
4128 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4129
4130 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4131 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4132 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4133 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4134 Node* obj = argument(0);
4135
4136 if (obj->is_InlineType() || gvn().type(obj)->is_inlinetypeptr()) {
4137 return false;
4138 }
4139
4140 if (!is_static) {
4141 // Check for hashing null object
4142 obj = null_check_receiver();
4143 if (stopped()) return true; // unconditionally null
4144 result_reg->init_req(_null_path, top());
4145 result_val->init_req(_null_path, top());
4146 } else {
4147 // Do a null check, and return zero if null.
4148 // System.identityHashCode(null) == 0
4149 Node* null_ctl = top();
4150 obj = null_check_oop(obj, &null_ctl);
4151 result_reg->init_req(_null_path, null_ctl);
4152 result_val->init_req(_null_path, _gvn.intcon(0));
4153 }
4154
4155 // Unconditionally null? Then return right away.
4156 if (stopped()) {
4157 set_control( result_reg->in(_null_path));
4158 if (!stopped())
4159 set_result(result_val->in(_null_path));
4160 return true;
4161 }
4162
4163 // We only go to the fast case code if we pass a number of guards. The
4164 // paths which do not pass are accumulated in the slow_region.
4165 RegionNode* slow_region = new RegionNode(1);
4166 record_for_igvn(slow_region);
4167
4168 // If this is a virtual call, we generate a funny guard. We pull out
4169 // the vtable entry corresponding to hashCode() from the target object.
4170 // If the target method which we are calling happens to be the native
4171 // Object hashCode() method, we pass the guard. We do not need this
4172 // guard for non-virtual calls -- the caller is known to be the native
4173 // Object hashCode().
4174 if (is_virtual) {
4175 // After null check, get the object's klass.
4176 Node* obj_klass = load_object_klass(obj);
4177 generate_virtual_guard(obj_klass, slow_region);
4178 }
4179
4180 // Get the header out of the object, use LoadMarkNode when available
4181 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4182 // The control of the load must be NULL. Otherwise, the load can move before
4183 // the null check after castPP removal.
4184 Node* no_ctrl = NULL;
4185 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4186
4187 // Test the header to see if it is unlocked.
4188 // This also serves as guard against inline types
4189 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4190 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4191 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4192 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4193 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4194
4195 generate_slow_guard(test_unlocked, slow_region);
4196
4197 // Get the hash value and check to see that it has been properly assigned.
4198 // We depend on hash_mask being at most 32 bits and avoid the use of
4199 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4200 // vm: see markWord.hpp.
4201 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4202 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4203 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4204 // This hack lets the hash bits live anywhere in the mark object now, as long
4205 // as the shift drops the relevant bits into the low 32 bits. Note that
4206 // Java spec says that HashCode is an int so there's no point in capturing
4207 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4208 hshifted_header = ConvX2I(hshifted_header);
4209 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4210
4211 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4212 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4213 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4214
4215 generate_slow_guard(test_assigned, slow_region);
4216
4217 Node* init_mem = reset_memory();
4218 // fill in the rest of the null path:
4219 result_io ->init_req(_null_path, i_o());
4220 result_mem->init_req(_null_path, init_mem);
4221
4222 result_val->init_req(_fast_path, hash_val);
4223 result_reg->init_req(_fast_path, control());
4224 result_io ->init_req(_fast_path, i_o());
4225 result_mem->init_req(_fast_path, init_mem);
4226
4227 // Generate code for the slow case. We make a call to hashCode().
4228 set_control(_gvn.transform(slow_region));
4229 if (!stopped()) {
4230 // No need for PreserveJVMState, because we're using up the present state.
4231 set_all_memory(init_mem);
4232 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4233 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4234 Node* slow_result = set_results_for_java_call(slow_call);
4235 // this->control() comes from set_results_for_java_call
4236 result_reg->init_req(_slow_path, control());
4237 result_val->init_req(_slow_path, slow_result);
4238 result_io ->set_req(_slow_path, i_o());
4239 result_mem ->set_req(_slow_path, reset_memory());
4240 }
4241
4242 // Return the combined state.
4243 set_i_o( _gvn.transform(result_io) );
4244 set_all_memory( _gvn.transform(result_mem));
4245
4246 set_result(result_reg, result_val);
4247 return true;
4248 }
4249
4250 //---------------------------inline_native_getClass----------------------------
4251 // public final native Class<?> java.lang.Object.getClass();
4252 //
4253 // Build special case code for calls to getClass on an object.
4254 bool LibraryCallKit::inline_native_getClass() {
4255 Node* obj = argument(0);
4256 if (obj->is_InlineTypeBase()) {
4257 const Type* t = _gvn.type(obj);
4258 if (t->maybe_null()) {
4259 null_check(obj);
4260 }
4261 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4262 return true;
4263 }
4264 obj = null_check_receiver();
4265 if (stopped()) return true;
4266 set_result(load_mirror_from_klass(load_object_klass(obj)));
4267 return true;
4268 }
4269
4270 //-----------------inline_native_Reflection_getCallerClass---------------------
4271 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4272 //
4273 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4274 //
4275 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4276 // in that it must skip particular security frames and checks for
4277 // caller sensitive methods.
4278 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4279 #ifndef PRODUCT
4280 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4281 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4282 }
4283 #endif
4284
4285 if (!jvms()->has_method()) {
4286 #ifndef PRODUCT
4287 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4288 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4289 }
4290 #endif
4291 return false;
4292 }
4293
4294 // Walk back up the JVM state to find the caller at the required
4295 // depth.
4296 JVMState* caller_jvms = jvms();
4297
4298 // Cf. JVM_GetCallerClass
4299 // NOTE: Start the loop at depth 1 because the current JVM state does
4300 // not include the Reflection.getCallerClass() frame.
4301 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4302 ciMethod* m = caller_jvms->method();
4303 switch (n) {
4304 case 0:
4305 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4306 break;
4307 case 1:
4308 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4309 if (!m->caller_sensitive()) {
4310 #ifndef PRODUCT
4311 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4312 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4313 }
4314 #endif
4315 return false; // bail-out; let JVM_GetCallerClass do the work
4316 }
4317 break;
4318 default:
4319 if (!m->is_ignored_by_security_stack_walk()) {
4320 // We have reached the desired frame; return the holder class.
4321 // Acquire method holder as java.lang.Class and push as constant.
4322 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4323 ciInstance* caller_mirror = caller_klass->java_mirror();
4324 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4325
4326 #ifndef PRODUCT
4327 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4328 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());
4329 tty->print_cr(" JVM state at this point:");
4330 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4331 ciMethod* m = jvms()->of_depth(i)->method();
4332 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4333 }
4334 }
4335 #endif
4336 return true;
4337 }
4338 break;
4339 }
4340 }
4341
4342 #ifndef PRODUCT
4343 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4344 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4345 tty->print_cr(" JVM state at this point:");
4346 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4347 ciMethod* m = jvms()->of_depth(i)->method();
4348 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4349 }
4350 }
4351 #endif
4352
4353 return false; // bail-out; let JVM_GetCallerClass do the work
4354 }
4355
4356 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4357 Node* arg = argument(0);
4358 Node* result = NULL;
4359
4360 switch (id) {
4361 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4362 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4363 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4364 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4365
4366 case vmIntrinsics::_doubleToLongBits: {
4367 // two paths (plus control) merge in a wood
4368 RegionNode *r = new RegionNode(3);
4369 Node *phi = new PhiNode(r, TypeLong::LONG);
4370
4371 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4372 // Build the boolean node
4373 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4374
4375 // Branch either way.
4376 // NaN case is less traveled, which makes all the difference.
4377 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4378 Node *opt_isnan = _gvn.transform(ifisnan);
4379 assert( opt_isnan->is_If(), "Expect an IfNode");
4380 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4381 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4382
4383 set_control(iftrue);
4384
4385 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4386 Node *slow_result = longcon(nan_bits); // return NaN
4387 phi->init_req(1, _gvn.transform( slow_result ));
4388 r->init_req(1, iftrue);
4389
4390 // Else fall through
4391 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4392 set_control(iffalse);
4393
4394 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4395 r->init_req(2, iffalse);
4396
4397 // Post merge
4398 set_control(_gvn.transform(r));
4399 record_for_igvn(r);
4400
4401 C->set_has_split_ifs(true); // Has chance for split-if optimization
4402 result = phi;
4403 assert(result->bottom_type()->isa_long(), "must be");
4404 break;
4405 }
4406
4407 case vmIntrinsics::_floatToIntBits: {
4408 // two paths (plus control) merge in a wood
4409 RegionNode *r = new RegionNode(3);
4410 Node *phi = new PhiNode(r, TypeInt::INT);
4411
4412 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4413 // Build the boolean node
4414 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4415
4416 // Branch either way.
4417 // NaN case is less traveled, which makes all the difference.
4418 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4419 Node *opt_isnan = _gvn.transform(ifisnan);
4420 assert( opt_isnan->is_If(), "Expect an IfNode");
4421 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4422 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4423
4424 set_control(iftrue);
4425
4426 static const jint nan_bits = 0x7fc00000;
4427 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4428 phi->init_req(1, _gvn.transform( slow_result ));
4429 r->init_req(1, iftrue);
4430
4431 // Else fall through
4432 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4433 set_control(iffalse);
4434
4435 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4436 r->init_req(2, iffalse);
4437
4438 // Post merge
4439 set_control(_gvn.transform(r));
4440 record_for_igvn(r);
4441
4442 C->set_has_split_ifs(true); // Has chance for split-if optimization
4443 result = phi;
4444 assert(result->bottom_type()->isa_int(), "must be");
4445 break;
4446 }
4447
4448 default:
4449 fatal_unexpected_iid(id);
4450 break;
4451 }
4452 set_result(_gvn.transform(result));
4453 return true;
4454 }
4455
4456 //----------------------inline_unsafe_copyMemory-------------------------
4457 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4458
4459 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4460 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4461 const Type* base_t = gvn.type(base);
4462
4463 bool in_native = (base_t == TypePtr::NULL_PTR);
4464 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4465 bool is_mixed = !in_heap && !in_native;
4466
4467 if (is_mixed) {
4468 return true; // mixed accesses can touch both on-heap and off-heap memory
4469 }
4470 if (in_heap) {
4471 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4472 if (!is_prim_array) {
4473 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4474 // there's not enough type information available to determine proper memory slice for it.
4475 return true;
4476 }
4477 }
4478 return false;
4479 }
4480
4481 bool LibraryCallKit::inline_unsafe_copyMemory() {
4482 if (callee()->is_static()) return false; // caller must have the capability!
4483 null_check_receiver(); // null-check receiver
4484 if (stopped()) return true;
4485
4486 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4487
4488 Node* src_base = argument(1); // type: oop
4489 Node* src_off = ConvL2X(argument(2)); // type: long
4490 Node* dst_base = argument(4); // type: oop
4491 Node* dst_off = ConvL2X(argument(5)); // type: long
4492 Node* size = ConvL2X(argument(7)); // type: long
4493
4494 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4495 "fieldOffset must be byte-scaled");
4496
4497 Node* src_addr = make_unsafe_address(src_base, src_off);
4498 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4499
4500 Node* thread = _gvn.transform(new ThreadLocalNode());
4501 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4502 BasicType doing_unsafe_access_bt = T_BYTE;
4503 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4504
4505 // update volatile field
4506 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4507
4508 int flags = RC_LEAF | RC_NO_FP;
4509
4510 const TypePtr* dst_type = TypePtr::BOTTOM;
4511
4512 // Adjust memory effects of the runtime call based on input values.
4513 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4514 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4515 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4516
4517 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4518 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4519 flags |= RC_NARROW_MEM; // narrow in memory
4520 }
4521 }
4522
4523 // Call it. Note that the length argument is not scaled.
4524 make_runtime_call(flags,
4525 OptoRuntime::fast_arraycopy_Type(),
4526 StubRoutines::unsafe_arraycopy(),
4527 "unsafe_arraycopy",
4528 dst_type,
4529 src_addr, dst_addr, size XTOP);
4530
4531 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4532
4533 return true;
4534 }
4535
4536 #undef XTOP
4537
4538 //------------------------clone_coping-----------------------------------
4539 // Helper function for inline_native_clone.
4540 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4541 assert(obj_size != NULL, "");
4542 Node* raw_obj = alloc_obj->in(1);
4543 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4544
4545 AllocateNode* alloc = NULL;
4546 if (ReduceBulkZeroing) {
4547 // We will be completely responsible for initializing this object -
4548 // mark Initialize node as complete.
4549 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4550 // The object was just allocated - there should be no any stores!
4551 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4552 // Mark as complete_with_arraycopy so that on AllocateNode
4553 // expansion, we know this AllocateNode is initialized by an array
4554 // copy and a StoreStore barrier exists after the array copy.
4555 alloc->initialization()->set_complete_with_arraycopy();
4556 }
4557
4558 Node* size = _gvn.transform(obj_size);
4559 access_clone(obj, alloc_obj, size, is_array);
4560
4561 // Do not let reads from the cloned object float above the arraycopy.
4562 if (alloc != NULL) {
4563 // Do not let stores that initialize this object be reordered with
4564 // a subsequent store that would make this object accessible by
4565 // other threads.
4566 // Record what AllocateNode this StoreStore protects so that
4567 // escape analysis can go from the MemBarStoreStoreNode to the
4568 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4569 // based on the escape status of the AllocateNode.
4570 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4571 } else {
4572 insert_mem_bar(Op_MemBarCPUOrder);
4573 }
4574 }
4575
4576 //------------------------inline_native_clone----------------------------
4577 // protected native Object java.lang.Object.clone();
4578 //
4579 // Here are the simple edge cases:
4580 // null receiver => normal trap
4581 // virtual and clone was overridden => slow path to out-of-line clone
4582 // not cloneable or finalizer => slow path to out-of-line Object.clone
4583 //
4584 // The general case has two steps, allocation and copying.
4585 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4586 //
4587 // Copying also has two cases, oop arrays and everything else.
4588 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4589 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4590 //
4591 // These steps fold up nicely if and when the cloned object's klass
4592 // can be sharply typed as an object array, a type array, or an instance.
4593 //
4594 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4595 PhiNode* result_val;
4596
4597 // Set the reexecute bit for the interpreter to reexecute
4598 // the bytecode that invokes Object.clone if deoptimization happens.
4599 { PreserveReexecuteState preexecs(this);
4600 jvms()->set_should_reexecute(true);
4601
4602 Node* obj = argument(0);
4603 if (obj->is_InlineType()) {
4604 return false;
4605 }
4606
4607 obj = null_check_receiver();
4608 if (stopped()) return true;
4609
4610 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4611
4612 // If we are going to clone an instance, we need its exact type to
4613 // know the number and types of fields to convert the clone to
4614 // loads/stores. Maybe a speculative type can help us.
4615 if (!obj_type->klass_is_exact() &&
4616 obj_type->speculative_type() != NULL &&
4617 obj_type->speculative_type()->is_instance_klass() &&
4618 !obj_type->speculative_type()->is_inlinetype()) {
4619 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4620 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4621 !spec_ik->has_injected_fields()) {
4622 ciKlass* k = obj_type->klass();
4623 if (!k->is_instance_klass() ||
4624 k->as_instance_klass()->is_interface() ||
4625 k->as_instance_klass()->has_subklass()) {
4626 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4627 }
4628 }
4629 }
4630
4631 // Conservatively insert a memory barrier on all memory slices.
4632 // Do not let writes into the original float below the clone.
4633 insert_mem_bar(Op_MemBarCPUOrder);
4634
4635 // paths into result_reg:
4636 enum {
4637 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4638 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4639 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4640 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4641 PATH_LIMIT
4642 };
4643 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4644 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4645 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4646 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4647 record_for_igvn(result_reg);
4648
4649 Node* obj_klass = load_object_klass(obj);
4650 // We only go to the fast case code if we pass a number of guards.
4651 // The paths which do not pass are accumulated in the slow_region.
4652 RegionNode* slow_region = new RegionNode(1);
4653 record_for_igvn(slow_region);
4654
4655 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4656 if (array_ctl != NULL) {
4657 // It's an array.
4658 PreserveJVMState pjvms(this);
4659 set_control(array_ctl);
4660
4661 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4662 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
4663 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
4664 obj_type->klass()->can_be_inline_array_klass() &&
4665 (ary_ptr == NULL || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
4666 // Flattened inline type array may have object field that would require a
4667 // write barrier. Conservatively, go to slow path.
4668 generate_flatArray_guard(obj_klass, slow_region);
4669 }
4670
4671 if (!stopped()) {
4672 Node* obj_length = load_array_length(obj);
4673 Node* obj_size = NULL;
4674 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4675
4676 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4677 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4678 // If it is an oop array, it requires very special treatment,
4679 // because gc barriers are required when accessing the array.
4680 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4681 if (is_obja != NULL) {
4682 PreserveJVMState pjvms2(this);
4683 set_control(is_obja);
4684 // Generate a direct call to the right arraycopy function(s).
4685 // Clones are always tightly coupled.
4686 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4687 ac->set_clone_oop_array();
4688 Node* n = _gvn.transform(ac);
4689 assert(n == ac, "cannot disappear");
4690 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4691
4692 result_reg->init_req(_objArray_path, control());
4693 result_val->init_req(_objArray_path, alloc_obj);
4694 result_i_o ->set_req(_objArray_path, i_o());
4695 result_mem ->set_req(_objArray_path, reset_memory());
4696 }
4697 }
4698 // Otherwise, there are no barriers to worry about.
4699 // (We can dispense with card marks if we know the allocation
4700 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4701 // causes the non-eden paths to take compensating steps to
4702 // simulate a fresh allocation, so that no further
4703 // card marks are required in compiled code to initialize
4704 // the object.)
4705
4706 if (!stopped()) {
4707 copy_to_clone(obj, alloc_obj, obj_size, true);
4708
4709 // Present the results of the copy.
4710 result_reg->init_req(_array_path, control());
4711 result_val->init_req(_array_path, alloc_obj);
4712 result_i_o ->set_req(_array_path, i_o());
4713 result_mem ->set_req(_array_path, reset_memory());
4714 }
4715 }
4716 }
4717
4718 if (!stopped()) {
4719 // It's an instance (we did array above). Make the slow-path tests.
4720 // If this is a virtual call, we generate a funny guard. We grab
4721 // the vtable entry corresponding to clone() from the target object.
4722 // If the target method which we are calling happens to be the
4723 // Object clone() method, we pass the guard. We do not need this
4724 // guard for non-virtual calls; the caller is known to be the native
4725 // Object clone().
4726 if (is_virtual) {
4727 generate_virtual_guard(obj_klass, slow_region);
4728 }
4729
4730 // The object must be easily cloneable and must not have a finalizer.
4731 // Both of these conditions may be checked in a single test.
4732 // We could optimize the test further, but we don't care.
4733 generate_access_flags_guard(obj_klass,
4734 // Test both conditions:
4735 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4736 // Must be cloneable but not finalizer:
4737 JVM_ACC_IS_CLONEABLE_FAST,
4738 slow_region);
4739 }
4740
4741 if (!stopped()) {
4742 // It's an instance, and it passed the slow-path tests.
4743 PreserveJVMState pjvms(this);
4744 Node* obj_size = NULL;
4745 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4746 // is reexecuted if deoptimization occurs and there could be problems when merging
4747 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4748 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4749
4750 copy_to_clone(obj, alloc_obj, obj_size, false);
4751
4752 // Present the results of the slow call.
4753 result_reg->init_req(_instance_path, control());
4754 result_val->init_req(_instance_path, alloc_obj);
4755 result_i_o ->set_req(_instance_path, i_o());
4756 result_mem ->set_req(_instance_path, reset_memory());
4757 }
4758
4759 // Generate code for the slow case. We make a call to clone().
4760 set_control(_gvn.transform(slow_region));
4761 if (!stopped()) {
4762 PreserveJVMState pjvms(this);
4763 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4764 // We need to deoptimize on exception (see comment above)
4765 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4766 // this->control() comes from set_results_for_java_call
4767 result_reg->init_req(_slow_path, control());
4768 result_val->init_req(_slow_path, slow_result);
4769 result_i_o ->set_req(_slow_path, i_o());
4770 result_mem ->set_req(_slow_path, reset_memory());
4771 }
4772
4773 // Return the combined state.
4774 set_control( _gvn.transform(result_reg));
4775 set_i_o( _gvn.transform(result_i_o));
4776 set_all_memory( _gvn.transform(result_mem));
4777 } // original reexecute is set back here
4778
4779 set_result(_gvn.transform(result_val));
4780 return true;
4781 }
4782
4783 // If we have a tightly coupled allocation, the arraycopy may take care
4784 // of the array initialization. If one of the guards we insert between
4785 // the allocation and the arraycopy causes a deoptimization, an
4786 // unitialized array will escape the compiled method. To prevent that
4787 // we set the JVM state for uncommon traps between the allocation and
4788 // the arraycopy to the state before the allocation so, in case of
4789 // deoptimization, we'll reexecute the allocation and the
4790 // initialization.
4791 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4792 if (alloc != NULL) {
4793 ciMethod* trap_method = alloc->jvms()->method();
4794 int trap_bci = alloc->jvms()->bci();
4795
4796 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4797 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4798 // Make sure there's no store between the allocation and the
4799 // arraycopy otherwise visible side effects could be rexecuted
4800 // in case of deoptimization and cause incorrect execution.
4801 bool no_interfering_store = true;
4802 Node* mem = alloc->in(TypeFunc::Memory);
4803 if (mem->is_MergeMem()) {
4804 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4805 Node* n = mms.memory();
4806 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4807 assert(n->is_Store(), "what else?");
4808 no_interfering_store = false;
4809 break;
4810 }
4811 }
4812 } else {
4813 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4814 Node* n = mms.memory();
4815 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4816 assert(n->is_Store(), "what else?");
4817 no_interfering_store = false;
4818 break;
4819 }
4820 }
4821 }
4822
4823 if (no_interfering_store) {
4824 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4825 uint size = alloc->req();
4826 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4827 old_jvms->set_map(sfpt);
4828 for (uint i = 0; i < size; i++) {
4829 sfpt->init_req(i, alloc->in(i));
4830 }
4831 // re-push array length for deoptimization
4832 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4833 old_jvms->set_sp(old_jvms->sp()+1);
4834 old_jvms->set_monoff(old_jvms->monoff()+1);
4835 old_jvms->set_scloff(old_jvms->scloff()+1);
4836 old_jvms->set_endoff(old_jvms->endoff()+1);
4837 old_jvms->set_should_reexecute(true);
4838
4839 sfpt->set_i_o(map()->i_o());
4840 sfpt->set_memory(map()->memory());
4841 sfpt->set_control(map()->control());
4842
4843 JVMState* saved_jvms = jvms();
4844 saved_reexecute_sp = _reexecute_sp;
4845
4846 set_jvms(sfpt->jvms());
4847 _reexecute_sp = jvms()->sp();
4848
4849 return saved_jvms;
4850 }
4851 }
4852 }
4853 return NULL;
4854 }
4855
4856 // In case of a deoptimization, we restart execution at the
4857 // allocation, allocating a new array. We would leave an uninitialized
4858 // array in the heap that GCs wouldn't expect. Move the allocation
4859 // after the traps so we don't allocate the array if we
4860 // deoptimize. This is possible because tightly_coupled_allocation()
4861 // guarantees there's no observer of the allocated array at this point
4862 // and the control flow is simple enough.
4863 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4864 int saved_reexecute_sp, uint new_idx) {
4865 if (saved_jvms != NULL && !stopped()) {
4866 assert(alloc != NULL, "only with a tightly coupled allocation");
4867 // restore JVM state to the state at the arraycopy
4868 saved_jvms->map()->set_control(map()->control());
4869 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4870 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4871 // If we've improved the types of some nodes (null check) while
4872 // emitting the guards, propagate them to the current state
4873 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4874 set_jvms(saved_jvms);
4875 _reexecute_sp = saved_reexecute_sp;
4876
4877 // Remove the allocation from above the guards
4878 CallProjections* callprojs = alloc->extract_projections(true);
4879 InitializeNode* init = alloc->initialization();
4880 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4881 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4882 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4883
4884 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4885 // the allocation (i.e. is only valid if the allocation succeeds):
4886 // 1) replace CastIINode with AllocateArrayNode's length here
4887 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4888 //
4889 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4890 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4891 Node* init_control = init->proj_out(TypeFunc::Control);
4892 Node* alloc_length = alloc->Ideal_length();
4893 #ifdef ASSERT
4894 Node* prev_cast = NULL;
4895 #endif
4896 for (uint i = 0; i < init_control->outcnt(); i++) {
4897 Node* init_out = init_control->raw_out(i);
4898 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4899 #ifdef ASSERT
4900 if (prev_cast == NULL) {
4901 prev_cast = init_out;
4902 } else {
4903 if (prev_cast->cmp(*init_out) == false) {
4904 prev_cast->dump();
4905 init_out->dump();
4906 assert(false, "not equal CastIINode");
4907 }
4908 }
4909 #endif
4910 C->gvn_replace_by(init_out, alloc_length);
4911 }
4912 }
4913 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4914
4915 // move the allocation here (after the guards)
4916 _gvn.hash_delete(alloc);
4917 alloc->set_req(TypeFunc::Control, control());
4918 alloc->set_req(TypeFunc::I_O, i_o());
4919 Node *mem = reset_memory();
4920 set_all_memory(mem);
4921 alloc->set_req(TypeFunc::Memory, mem);
4922 set_control(init->proj_out_or_null(TypeFunc::Control));
4923 set_i_o(callprojs->fallthrough_ioproj);
4924
4925 // Update memory as done in GraphKit::set_output_for_allocation()
4926 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4927 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4928 if (ary_type->isa_aryptr() && length_type != NULL) {
4929 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4930 }
4931 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4932 int elemidx = C->get_alias_index(telemref);
4933 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4934 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4935
4936 Node* allocx = _gvn.transform(alloc);
4937 assert(allocx == alloc, "where has the allocation gone?");
4938 assert(dest->is_CheckCastPP(), "not an allocation result?");
4939
4940 _gvn.hash_delete(dest);
4941 dest->set_req(0, control());
4942 Node* destx = _gvn.transform(dest);
4943 assert(destx == dest, "where has the allocation result gone?");
4944
4945 array_ideal_length(alloc, ary_type, true);
4946 }
4947 }
4948
4949
4950 //------------------------------inline_arraycopy-----------------------
4951 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4952 // Object dest, int destPos,
4953 // int length);
4954 bool LibraryCallKit::inline_arraycopy() {
4955 // Get the arguments.
4956 Node* src = argument(0); // type: oop
4957 Node* src_offset = argument(1); // type: int
4958 Node* dest = argument(2); // type: oop
4959 Node* dest_offset = argument(3); // type: int
4960 Node* length = argument(4); // type: int
4961
4962 uint new_idx = C->unique();
4963
4964 // Check for allocation before we add nodes that would confuse
4965 // tightly_coupled_allocation()
4966 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
4967
4968 int saved_reexecute_sp = -1;
4969 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4970 // See arraycopy_restore_alloc_state() comment
4971 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4972 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4973 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4974 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4975
4976 // The following tests must be performed
4977 // (1) src and dest are arrays.
4978 // (2) src and dest arrays must have elements of the same BasicType
4979 // (3) src and dest must not be null.
4980 // (4) src_offset must not be negative.
4981 // (5) dest_offset must not be negative.
4982 // (6) length must not be negative.
4983 // (7) src_offset + length must not exceed length of src.
4984 // (8) dest_offset + length must not exceed length of dest.
4985 // (9) each element of an oop array must be assignable
4986
4987 // (3) src and dest must not be null.
4988 // always do this here because we need the JVM state for uncommon traps
4989 Node* null_ctl = top();
4990 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4991 assert(null_ctl->is_top(), "no null control here");
4992 dest = null_check(dest, T_ARRAY);
4993
4994 if (!can_emit_guards) {
4995 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4996 // guards but the arraycopy node could still take advantage of a
4997 // tightly allocated allocation. tightly_coupled_allocation() is
4998 // called again to make sure it takes the null check above into
4999 // account: the null check is mandatory and if it caused an
5000 // uncommon trap to be emitted then the allocation can't be
5001 // considered tightly coupled in this context.
5002 alloc = tightly_coupled_allocation(dest);
5003 }
5004
5005 bool validated = false;
5006
5007 const Type* src_type = _gvn.type(src);
5008 const Type* dest_type = _gvn.type(dest);
5009 const TypeAryPtr* top_src = src_type->isa_aryptr();
5010 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5011
5012 // Do we have the type of src?
5013 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5014 // Do we have the type of dest?
5015 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5016 // Is the type for src from speculation?
5017 bool src_spec = false;
5018 // Is the type for dest from speculation?
5019 bool dest_spec = false;
5020
5021 if ((!has_src || !has_dest) && can_emit_guards) {
5022 // We don't have sufficient type information, let's see if
5023 // speculative types can help. We need to have types for both src
5024 // and dest so that it pays off.
5025
5026 // Do we already have or could we have type information for src
5027 bool could_have_src = has_src;
5028 // Do we already have or could we have type information for dest
5029 bool could_have_dest = has_dest;
5030
5031 ciKlass* src_k = NULL;
5032 if (!has_src) {
5033 src_k = src_type->speculative_type_not_null();
5034 if (src_k != NULL && src_k->is_array_klass()) {
5035 could_have_src = true;
5036 }
5037 }
5038
5039 ciKlass* dest_k = NULL;
5040 if (!has_dest) {
5041 dest_k = dest_type->speculative_type_not_null();
5042 if (dest_k != NULL && dest_k->is_array_klass()) {
5043 could_have_dest = true;
5044 }
5045 }
5046
5047 if (could_have_src && could_have_dest) {
5048 // This is going to pay off so emit the required guards
5049 if (!has_src) {
5050 src = maybe_cast_profiled_obj(src, src_k, true);
5051 src_type = _gvn.type(src);
5052 top_src = src_type->isa_aryptr();
5053 has_src = (top_src != NULL && top_src->klass() != NULL);
5054 src_spec = true;
5055 }
5056 if (!has_dest) {
5057 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5058 dest_type = _gvn.type(dest);
5059 top_dest = dest_type->isa_aryptr();
5060 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5061 dest_spec = true;
5062 }
5063 }
5064 }
5065
5066 if (has_src && has_dest && can_emit_guards) {
5067 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5068 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5069 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
5070 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
5071
5072 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5073 // If both arrays are object arrays then having the exact types
5074 // for both will remove the need for a subtype check at runtime
5075 // before the call and may make it possible to pick a faster copy
5076 // routine (without a subtype check on every element)
5077 // Do we have the exact type of src?
5078 bool could_have_src = src_spec;
5079 // Do we have the exact type of dest?
5080 bool could_have_dest = dest_spec;
5081 ciKlass* src_k = top_src->klass();
5082 ciKlass* dest_k = top_dest->klass();
5083 if (!src_spec) {
5084 src_k = src_type->speculative_type_not_null();
5085 if (src_k != NULL && src_k->is_array_klass()) {
5086 could_have_src = true;
5087 }
5088 }
5089 if (!dest_spec) {
5090 dest_k = dest_type->speculative_type_not_null();
5091 if (dest_k != NULL && dest_k->is_array_klass()) {
5092 could_have_dest = true;
5093 }
5094 }
5095 if (could_have_src && could_have_dest) {
5096 // If we can have both exact types, emit the missing guards
5097 if (could_have_src && !src_spec) {
5098 src = maybe_cast_profiled_obj(src, src_k, true);
5099 src_type = _gvn.type(src);
5100 top_src = src_type->isa_aryptr();
5101 }
5102 if (could_have_dest && !dest_spec) {
5103 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5104 dest_type = _gvn.type(dest);
5105 top_dest = dest_type->isa_aryptr();
5106 }
5107 }
5108 }
5109 }
5110
5111 ciMethod* trap_method = method();
5112 int trap_bci = bci();
5113 if (saved_jvms != NULL) {
5114 trap_method = alloc->jvms()->method();
5115 trap_bci = alloc->jvms()->bci();
5116 }
5117
5118 bool negative_length_guard_generated = false;
5119
5120 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5121 can_emit_guards && !src->is_top() && !dest->is_top()) {
5122 // validate arguments: enables transformation the ArrayCopyNode
5123 validated = true;
5124
5125 RegionNode* slow_region = new RegionNode(1);
5126 record_for_igvn(slow_region);
5127
5128 // (1) src and dest are arrays.
5129 generate_non_array_guard(load_object_klass(src), slow_region);
5130 generate_non_array_guard(load_object_klass(dest), slow_region);
5131
5132 // (2) src and dest arrays must have elements of the same BasicType
5133 // done at macro expansion or at Ideal transformation time
5134
5135 // (4) src_offset must not be negative.
5136 generate_negative_guard(src_offset, slow_region);
5137
5138 // (5) dest_offset must not be negative.
5139 generate_negative_guard(dest_offset, slow_region);
5140
5141 // (7) src_offset + length must not exceed length of src.
5142 generate_limit_guard(src_offset, length,
5143 load_array_length(src),
5144 slow_region);
5145
5146 // (8) dest_offset + length must not exceed length of dest.
5147 generate_limit_guard(dest_offset, length,
5148 load_array_length(dest),
5149 slow_region);
5150
5151 // (6) length must not be negative.
5152 // This is also checked in generate_arraycopy() during macro expansion, but
5153 // we also have to check it here for the case where the ArrayCopyNode will
5154 // be eliminated by Escape Analysis.
5155 if (EliminateAllocations) {
5156 generate_negative_guard(length, slow_region);
5157 negative_length_guard_generated = true;
5158 }
5159
5160 // (9) each element of an oop array must be assignable
5161 Node* dest_klass = load_object_klass(dest);
5162 if (src != dest) {
5163 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5164 slow_region->add_req(not_subtype_ctrl);
5165 }
5166
5167 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5168 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
5169 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5170 src_type = _gvn.type(src);
5171 top_src = src_type->isa_aryptr();
5172
5173 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
5174 if (!stopped() && UseFlatArray) {
5175 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
5176 assert(top_dest == NULL || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
5177 if (top_src != NULL && top_src->is_flat()) {
5178 // Src is flat, check that dest is flat as well
5179 if (top_dest != NULL && !top_dest->is_flat()) {
5180 generate_non_flatArray_guard(dest_klass, slow_region);
5181 // Since dest is flat and src <: dest, dest must have the same type as src.
5182 top_dest = TypeOopPtr::make_from_klass(top_src->klass())->isa_aryptr();
5183 assert(top_dest->is_flat(), "dest must be flat");
5184 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
5185 }
5186 } else if (top_src == NULL || !top_src->is_not_flat()) {
5187 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5188 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5189 assert(top_dest == NULL || !top_dest->is_flat(), "dest array must not be flat");
5190 generate_flatArray_guard(load_object_klass(src), slow_region);
5191 if (top_src != NULL) {
5192 top_src = top_src->cast_to_not_flat();
5193 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
5194 }
5195 }
5196 }
5197
5198 {
5199 PreserveJVMState pjvms(this);
5200 set_control(_gvn.transform(slow_region));
5201 uncommon_trap(Deoptimization::Reason_intrinsic,
5202 Deoptimization::Action_make_not_entrant);
5203 assert(stopped(), "Should be stopped");
5204 }
5205 }
5206
5207 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5208
5209 if (stopped()) {
5210 return true;
5211 }
5212
5213 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5214 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5215 // so the compiler has a chance to eliminate them: during macro expansion,
5216 // we have to set their control (CastPP nodes are eliminated).
5217 load_object_klass(src), load_object_klass(dest),
5218 load_array_length(src), load_array_length(dest));
5219
5220 ac->set_arraycopy(validated);
5221
5222 Node* n = _gvn.transform(ac);
5223 if (n == ac) {
5224 ac->connect_outputs(this);
5225 } else {
5226 assert(validated, "shouldn't transform if all arguments not validated");
5227 set_all_memory(n);
5228 }
5229 clear_upper_avx();
5230
5231
5232 return true;
5233 }
5234
5235
5236 // Helper function which determines if an arraycopy immediately follows
5237 // an allocation, with no intervening tests or other escapes for the object.
5238 AllocateArrayNode*
5239 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5240 if (stopped()) return NULL; // no fast path
5241 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5242
5243 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5244 if (alloc == NULL) return NULL;
5245
5246 Node* rawmem = memory(Compile::AliasIdxRaw);
5247 // Is the allocation's memory state untouched?
5248 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5249 // Bail out if there have been raw-memory effects since the allocation.
5250 // (Example: There might have been a call or safepoint.)
5251 return NULL;
5252 }
5253 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5254 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5255 return NULL;
5256 }
5257
5258 // There must be no unexpected observers of this allocation.
5259 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5260 Node* obs = ptr->fast_out(i);
5261 if (obs != this->map()) {
5262 return NULL;
5263 }
5264 }
5265
5266 // This arraycopy must unconditionally follow the allocation of the ptr.
5267 Node* alloc_ctl = ptr->in(0);
5268 Node* ctl = control();
5269 while (ctl != alloc_ctl) {
5270 // There may be guards which feed into the slow_region.
5271 // Any other control flow means that we might not get a chance
5272 // to finish initializing the allocated object.
5273 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5274 IfNode* iff = ctl->in(0)->as_If();
5275 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5276 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5277 // One more try: Various low-level checks bottom out in
5278 // uncommon traps. If the debug-info of the trap omits
5279 // any reference to the allocation, as we've already
5280 // observed, then there can be no objection to the trap.
5281 bool found_trap = false;
5282 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5283 Node* obs = not_ctl->fast_out(j);
5284 if (obs->in(0) == not_ctl && obs->is_Call() &&
5285 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5286 found_trap = true; break;
5287 }
5288 }
5289 if (found_trap) {
5290 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5291 continue;
5292 }
5293 }
5294 return NULL;
5295 }
5296
5297 // If we get this far, we have an allocation which immediately
5298 // precedes the arraycopy, and we can take over zeroing the new object.
5299 // The arraycopy will finish the initialization, and provide
5300 // a new control state to which we will anchor the destination pointer.
5301
5302 return alloc;
5303 }
5304
5305 //-------------inline_encodeISOArray-----------------------------------
5306 // encode char[] to byte[] in ISO_8859_1 or ASCII
5307 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5308 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5309 // no receiver since it is static method
5310 Node *src = argument(0);
5311 Node *src_offset = argument(1);
5312 Node *dst = argument(2);
5313 Node *dst_offset = argument(3);
5314 Node *length = argument(4);
5315
5316 src = must_be_not_null(src, true);
5317 dst = must_be_not_null(dst, true);
5318
5319 const Type* src_type = src->Value(&_gvn);
5320 const Type* dst_type = dst->Value(&_gvn);
5321 const TypeAryPtr* top_src = src_type->isa_aryptr();
5322 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5323 if (top_src == NULL || top_src->klass() == NULL ||
5324 top_dest == NULL || top_dest->klass() == NULL) {
5325 // failed array check
5326 return false;
5327 }
5328
5329 // Figure out the size and type of the elements we will be copying.
5330 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5331 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5332 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5333 return false;
5334 }
5335
5336 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5337 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5338 // 'src_start' points to src array + scaled offset
5339 // 'dst_start' points to dst array + scaled offset
5340
5341 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5342 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5343 enc = _gvn.transform(enc);
5344 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5345 set_memory(res_mem, mtype);
5346 set_result(enc);
5347 clear_upper_avx();
5348
5349 return true;
5350 }
5351
5352 //-------------inline_multiplyToLen-----------------------------------
5353 bool LibraryCallKit::inline_multiplyToLen() {
5354 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5355
5356 address stubAddr = StubRoutines::multiplyToLen();
5357 if (stubAddr == NULL) {
5358 return false; // Intrinsic's stub is not implemented on this platform
5359 }
5360 const char* stubName = "multiplyToLen";
5361
5362 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5363
5364 // no receiver because it is a static method
5365 Node* x = argument(0);
5366 Node* xlen = argument(1);
5367 Node* y = argument(2);
5368 Node* ylen = argument(3);
5369 Node* z = argument(4);
5370
5371 x = must_be_not_null(x, true);
5372 y = must_be_not_null(y, true);
5373
5374 const Type* x_type = x->Value(&_gvn);
5375 const Type* y_type = y->Value(&_gvn);
5376 const TypeAryPtr* top_x = x_type->isa_aryptr();
5377 const TypeAryPtr* top_y = y_type->isa_aryptr();
5378 if (top_x == NULL || top_x->klass() == NULL ||
5379 top_y == NULL || top_y->klass() == NULL) {
5380 // failed array check
5381 return false;
5382 }
5383
5384 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5385 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5386 if (x_elem != T_INT || y_elem != T_INT) {
5387 return false;
5388 }
5389
5390 // Set the original stack and the reexecute bit for the interpreter to reexecute
5391 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5392 // on the return from z array allocation in runtime.
5393 { PreserveReexecuteState preexecs(this);
5394 jvms()->set_should_reexecute(true);
5395
5396 Node* x_start = array_element_address(x, intcon(0), x_elem);
5397 Node* y_start = array_element_address(y, intcon(0), y_elem);
5398 // 'x_start' points to x array + scaled xlen
5399 // 'y_start' points to y array + scaled ylen
5400
5401 // Allocate the result array
5402 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5403 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5404 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5405
5406 IdealKit ideal(this);
5407
5408 #define __ ideal.
5409 Node* one = __ ConI(1);
5410 Node* zero = __ ConI(0);
5411 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5412 __ set(need_alloc, zero);
5413 __ set(z_alloc, z);
5414 __ if_then(z, BoolTest::eq, null()); {
5415 __ increment (need_alloc, one);
5416 } __ else_(); {
5417 // Update graphKit memory and control from IdealKit.
5418 sync_kit(ideal);
5419 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5420 cast->init_req(0, control());
5421 _gvn.set_type(cast, cast->bottom_type());
5422 C->record_for_igvn(cast);
5423
5424 Node* zlen_arg = load_array_length(cast);
5425 // Update IdealKit memory and control from graphKit.
5426 __ sync_kit(this);
5427 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5428 __ increment (need_alloc, one);
5429 } __ end_if();
5430 } __ end_if();
5431
5432 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5433 // Update graphKit memory and control from IdealKit.
5434 sync_kit(ideal);
5435 Node * narr = new_array(klass_node, zlen, 1);
5436 // Update IdealKit memory and control from graphKit.
5437 __ sync_kit(this);
5438 __ set(z_alloc, narr);
5439 } __ end_if();
5440
5441 sync_kit(ideal);
5442 z = __ value(z_alloc);
5443 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5444 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5445 // Final sync IdealKit and GraphKit.
5446 final_sync(ideal);
5447 #undef __
5448
5449 Node* z_start = array_element_address(z, intcon(0), T_INT);
5450
5451 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5452 OptoRuntime::multiplyToLen_Type(),
5453 stubAddr, stubName, TypePtr::BOTTOM,
5454 x_start, xlen, y_start, ylen, z_start, zlen);
5455 } // original reexecute is set back here
5456
5457 C->set_has_split_ifs(true); // Has chance for split-if optimization
5458 set_result(z);
5459 return true;
5460 }
5461
5462 //-------------inline_squareToLen------------------------------------
5463 bool LibraryCallKit::inline_squareToLen() {
5464 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5465
5466 address stubAddr = StubRoutines::squareToLen();
5467 if (stubAddr == NULL) {
5468 return false; // Intrinsic's stub is not implemented on this platform
5469 }
5470 const char* stubName = "squareToLen";
5471
5472 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5473
5474 Node* x = argument(0);
5475 Node* len = argument(1);
5476 Node* z = argument(2);
5477 Node* zlen = argument(3);
5478
5479 x = must_be_not_null(x, true);
5480 z = must_be_not_null(z, true);
5481
5482 const Type* x_type = x->Value(&_gvn);
5483 const Type* z_type = z->Value(&_gvn);
5484 const TypeAryPtr* top_x = x_type->isa_aryptr();
5485 const TypeAryPtr* top_z = z_type->isa_aryptr();
5486 if (top_x == NULL || top_x->klass() == NULL ||
5487 top_z == NULL || top_z->klass() == NULL) {
5488 // failed array check
5489 return false;
5490 }
5491
5492 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5493 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5494 if (x_elem != T_INT || z_elem != T_INT) {
5495 return false;
5496 }
5497
5498
5499 Node* x_start = array_element_address(x, intcon(0), x_elem);
5500 Node* z_start = array_element_address(z, intcon(0), z_elem);
5501
5502 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5503 OptoRuntime::squareToLen_Type(),
5504 stubAddr, stubName, TypePtr::BOTTOM,
5505 x_start, len, z_start, zlen);
5506
5507 set_result(z);
5508 return true;
5509 }
5510
5511 //-------------inline_mulAdd------------------------------------------
5512 bool LibraryCallKit::inline_mulAdd() {
5513 assert(UseMulAddIntrinsic, "not implemented on this platform");
5514
5515 address stubAddr = StubRoutines::mulAdd();
5516 if (stubAddr == NULL) {
5517 return false; // Intrinsic's stub is not implemented on this platform
5518 }
5519 const char* stubName = "mulAdd";
5520
5521 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5522
5523 Node* out = argument(0);
5524 Node* in = argument(1);
5525 Node* offset = argument(2);
5526 Node* len = argument(3);
5527 Node* k = argument(4);
5528
5529 out = must_be_not_null(out, true);
5530
5531 const Type* out_type = out->Value(&_gvn);
5532 const Type* in_type = in->Value(&_gvn);
5533 const TypeAryPtr* top_out = out_type->isa_aryptr();
5534 const TypeAryPtr* top_in = in_type->isa_aryptr();
5535 if (top_out == NULL || top_out->klass() == NULL ||
5536 top_in == NULL || top_in->klass() == NULL) {
5537 // failed array check
5538 return false;
5539 }
5540
5541 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5542 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5543 if (out_elem != T_INT || in_elem != T_INT) {
5544 return false;
5545 }
5546
5547 Node* outlen = load_array_length(out);
5548 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5549 Node* out_start = array_element_address(out, intcon(0), out_elem);
5550 Node* in_start = array_element_address(in, intcon(0), in_elem);
5551
5552 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5553 OptoRuntime::mulAdd_Type(),
5554 stubAddr, stubName, TypePtr::BOTTOM,
5555 out_start,in_start, new_offset, len, k);
5556 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5557 set_result(result);
5558 return true;
5559 }
5560
5561 //-------------inline_montgomeryMultiply-----------------------------------
5562 bool LibraryCallKit::inline_montgomeryMultiply() {
5563 address stubAddr = StubRoutines::montgomeryMultiply();
5564 if (stubAddr == NULL) {
5565 return false; // Intrinsic's stub is not implemented on this platform
5566 }
5567
5568 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5569 const char* stubName = "montgomery_multiply";
5570
5571 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5572
5573 Node* a = argument(0);
5574 Node* b = argument(1);
5575 Node* n = argument(2);
5576 Node* len = argument(3);
5577 Node* inv = argument(4);
5578 Node* m = argument(6);
5579
5580 const Type* a_type = a->Value(&_gvn);
5581 const TypeAryPtr* top_a = a_type->isa_aryptr();
5582 const Type* b_type = b->Value(&_gvn);
5583 const TypeAryPtr* top_b = b_type->isa_aryptr();
5584 const Type* n_type = a->Value(&_gvn);
5585 const TypeAryPtr* top_n = n_type->isa_aryptr();
5586 const Type* m_type = a->Value(&_gvn);
5587 const TypeAryPtr* top_m = m_type->isa_aryptr();
5588 if (top_a == NULL || top_a->klass() == NULL ||
5589 top_b == NULL || top_b->klass() == NULL ||
5590 top_n == NULL || top_n->klass() == NULL ||
5591 top_m == NULL || top_m->klass() == NULL) {
5592 // failed array check
5593 return false;
5594 }
5595
5596 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5597 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5598 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5599 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5600 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5601 return false;
5602 }
5603
5604 // Make the call
5605 {
5606 Node* a_start = array_element_address(a, intcon(0), a_elem);
5607 Node* b_start = array_element_address(b, intcon(0), b_elem);
5608 Node* n_start = array_element_address(n, intcon(0), n_elem);
5609 Node* m_start = array_element_address(m, intcon(0), m_elem);
5610
5611 Node* call = make_runtime_call(RC_LEAF,
5612 OptoRuntime::montgomeryMultiply_Type(),
5613 stubAddr, stubName, TypePtr::BOTTOM,
5614 a_start, b_start, n_start, len, inv, top(),
5615 m_start);
5616 set_result(m);
5617 }
5618
5619 return true;
5620 }
5621
5622 bool LibraryCallKit::inline_montgomerySquare() {
5623 address stubAddr = StubRoutines::montgomerySquare();
5624 if (stubAddr == NULL) {
5625 return false; // Intrinsic's stub is not implemented on this platform
5626 }
5627
5628 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5629 const char* stubName = "montgomery_square";
5630
5631 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5632
5633 Node* a = argument(0);
5634 Node* n = argument(1);
5635 Node* len = argument(2);
5636 Node* inv = argument(3);
5637 Node* m = argument(5);
5638
5639 const Type* a_type = a->Value(&_gvn);
5640 const TypeAryPtr* top_a = a_type->isa_aryptr();
5641 const Type* n_type = a->Value(&_gvn);
5642 const TypeAryPtr* top_n = n_type->isa_aryptr();
5643 const Type* m_type = a->Value(&_gvn);
5644 const TypeAryPtr* top_m = m_type->isa_aryptr();
5645 if (top_a == NULL || top_a->klass() == NULL ||
5646 top_n == NULL || top_n->klass() == NULL ||
5647 top_m == NULL || top_m->klass() == NULL) {
5648 // failed array check
5649 return false;
5650 }
5651
5652 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5653 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5654 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5655 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5656 return false;
5657 }
5658
5659 // Make the call
5660 {
5661 Node* a_start = array_element_address(a, intcon(0), a_elem);
5662 Node* n_start = array_element_address(n, intcon(0), n_elem);
5663 Node* m_start = array_element_address(m, intcon(0), m_elem);
5664
5665 Node* call = make_runtime_call(RC_LEAF,
5666 OptoRuntime::montgomerySquare_Type(),
5667 stubAddr, stubName, TypePtr::BOTTOM,
5668 a_start, n_start, len, inv, top(),
5669 m_start);
5670 set_result(m);
5671 }
5672
5673 return true;
5674 }
5675
5676 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5677 address stubAddr = NULL;
5678 const char* stubName = NULL;
5679
5680 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5681 if (stubAddr == NULL) {
5682 return false; // Intrinsic's stub is not implemented on this platform
5683 }
5684
5685 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5686
5687 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5688
5689 Node* newArr = argument(0);
5690 Node* oldArr = argument(1);
5691 Node* newIdx = argument(2);
5692 Node* shiftCount = argument(3);
5693 Node* numIter = argument(4);
5694
5695 const Type* newArr_type = newArr->Value(&_gvn);
5696 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5697 const Type* oldArr_type = oldArr->Value(&_gvn);
5698 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5699 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5700 || top_oldArr->klass() == NULL) {
5701 return false;
5702 }
5703
5704 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5705 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5706 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5707 return false;
5708 }
5709
5710 // Make the call
5711 {
5712 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5713 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5714
5715 Node* call = make_runtime_call(RC_LEAF,
5716 OptoRuntime::bigIntegerShift_Type(),
5717 stubAddr,
5718 stubName,
5719 TypePtr::BOTTOM,
5720 newArr_start,
5721 oldArr_start,
5722 newIdx,
5723 shiftCount,
5724 numIter);
5725 }
5726
5727 return true;
5728 }
5729
5730 //-------------inline_vectorizedMismatch------------------------------
5731 bool LibraryCallKit::inline_vectorizedMismatch() {
5732 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5733
5734 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5735 Node* obja = argument(0); // Object
5736 Node* aoffset = argument(1); // long
5737 Node* objb = argument(3); // Object
5738 Node* boffset = argument(4); // long
5739 Node* length = argument(6); // int
5740 Node* scale = argument(7); // int
5741
5742 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5743 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5744 if (obja_t == NULL || obja_t->klass() == NULL ||
5745 objb_t == NULL || objb_t->klass() == NULL ||
5746 scale == top()) {
5747 return false; // failed input validation
5748 }
5749
5750 Node* obja_adr = make_unsafe_address(obja, aoffset);
5751 Node* objb_adr = make_unsafe_address(objb, boffset);
5752
5753 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5754 //
5755 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5756 // if (length <= inline_limit) {
5757 // inline_path:
5758 // vmask = VectorMaskGen length
5759 // vload1 = LoadVectorMasked obja, vmask
5760 // vload2 = LoadVectorMasked objb, vmask
5761 // result1 = VectorCmpMasked vload1, vload2, vmask
5762 // } else {
5763 // call_stub_path:
5764 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5765 // }
5766 // exit_block:
5767 // return Phi(result1, result2);
5768 //
5769 enum { inline_path = 1, // input is small enough to process it all at once
5770 stub_path = 2, // input is too large; call into the VM
5771 PATH_LIMIT = 3
5772 };
5773
5774 Node* exit_block = new RegionNode(PATH_LIMIT);
5775 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5776 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5777
5778 Node* call_stub_path = control();
5779
5780 BasicType elem_bt = T_ILLEGAL;
5781
5782 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5783 if (scale_t->is_con()) {
5784 switch (scale_t->get_con()) {
5785 case 0: elem_bt = T_BYTE; break;
5786 case 1: elem_bt = T_SHORT; break;
5787 case 2: elem_bt = T_INT; break;
5788 case 3: elem_bt = T_LONG; break;
5789
5790 default: elem_bt = T_ILLEGAL; break; // not supported
5791 }
5792 }
5793
5794 int inline_limit = 0;
5795 bool do_partial_inline = false;
5796
5797 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5798 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5799 do_partial_inline = inline_limit >= 16;
5800 }
5801
5802 if (do_partial_inline) {
5803 assert(elem_bt != T_ILLEGAL, "sanity");
5804
5805 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5806 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5807 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5808
5809 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5810 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5811 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5812
5813 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5814
5815 if (!stopped()) {
5816 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5817
5818 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5819 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5820 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5821 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5822
5823 Node* vmask = _gvn.transform(new VectorMaskGenNode(ConvI2X(casted_length), TypeVect::VECTMASK, elem_bt));
5824 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5825 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5826 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5827
5828 exit_block->init_req(inline_path, control());
5829 memory_phi->init_req(inline_path, map()->memory());
5830 result_phi->init_req(inline_path, result);
5831
5832 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5833 clear_upper_avx();
5834 }
5835 }
5836 }
5837
5838 if (call_stub_path != NULL) {
5839 set_control(call_stub_path);
5840
5841 Node* call = make_runtime_call(RC_LEAF,
5842 OptoRuntime::vectorizedMismatch_Type(),
5843 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5844 obja_adr, objb_adr, length, scale);
5845
5846 exit_block->init_req(stub_path, control());
5847 memory_phi->init_req(stub_path, map()->memory());
5848 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5849 }
5850
5851 exit_block = _gvn.transform(exit_block);
5852 memory_phi = _gvn.transform(memory_phi);
5853 result_phi = _gvn.transform(result_phi);
5854
5855 set_control(exit_block);
5856 set_all_memory(memory_phi);
5857 set_result(result_phi);
5858
5859 return true;
5860 }
5861
5862 /**
5863 * Calculate CRC32 for byte.
5864 * int java.util.zip.CRC32.update(int crc, int b)
5865 */
5866 bool LibraryCallKit::inline_updateCRC32() {
5867 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5868 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5869 // no receiver since it is static method
5870 Node* crc = argument(0); // type: int
5871 Node* b = argument(1); // type: int
5872
5873 /*
5874 * int c = ~ crc;
5875 * b = timesXtoThe32[(b ^ c) & 0xFF];
5876 * b = b ^ (c >>> 8);
5877 * crc = ~b;
5878 */
5879
5880 Node* M1 = intcon(-1);
5881 crc = _gvn.transform(new XorINode(crc, M1));
5882 Node* result = _gvn.transform(new XorINode(crc, b));
5883 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5884
5885 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5886 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5887 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5888 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5889
5890 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5891 result = _gvn.transform(new XorINode(crc, result));
5892 result = _gvn.transform(new XorINode(result, M1));
5893 set_result(result);
5894 return true;
5895 }
5896
5897 /**
5898 * Calculate CRC32 for byte[] array.
5899 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5900 */
5901 bool LibraryCallKit::inline_updateBytesCRC32() {
5902 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5903 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5904 // no receiver since it is static method
5905 Node* crc = argument(0); // type: int
5906 Node* src = argument(1); // type: oop
5907 Node* offset = argument(2); // type: int
5908 Node* length = argument(3); // type: int
5909
5910 const Type* src_type = src->Value(&_gvn);
5911 const TypeAryPtr* top_src = src_type->isa_aryptr();
5912 if (top_src == NULL || top_src->klass() == NULL) {
5913 // failed array check
5914 return false;
5915 }
5916
5917 // Figure out the size and type of the elements we will be copying.
5918 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5919 if (src_elem != T_BYTE) {
5920 return false;
5921 }
5922
5923 // 'src_start' points to src array + scaled offset
5924 src = must_be_not_null(src, true);
5925 Node* src_start = array_element_address(src, offset, src_elem);
5926
5927 // We assume that range check is done by caller.
5928 // TODO: generate range check (offset+length < src.length) in debug VM.
5929
5930 // Call the stub.
5931 address stubAddr = StubRoutines::updateBytesCRC32();
5932 const char *stubName = "updateBytesCRC32";
5933
5934 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5935 stubAddr, stubName, TypePtr::BOTTOM,
5936 crc, src_start, length);
5937 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5938 set_result(result);
5939 return true;
5940 }
5941
5942 /**
5943 * Calculate CRC32 for ByteBuffer.
5944 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5945 */
5946 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5947 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5948 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5949 // no receiver since it is static method
5950 Node* crc = argument(0); // type: int
5951 Node* src = argument(1); // type: long
5952 Node* offset = argument(3); // type: int
5953 Node* length = argument(4); // type: int
5954
5955 src = ConvL2X(src); // adjust Java long to machine word
5956 Node* base = _gvn.transform(new CastX2PNode(src));
5957 offset = ConvI2X(offset);
5958
5959 // 'src_start' points to src array + scaled offset
5960 Node* src_start = basic_plus_adr(top(), base, offset);
5961
5962 // Call the stub.
5963 address stubAddr = StubRoutines::updateBytesCRC32();
5964 const char *stubName = "updateBytesCRC32";
5965
5966 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5967 stubAddr, stubName, TypePtr::BOTTOM,
5968 crc, src_start, length);
5969 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5970 set_result(result);
5971 return true;
5972 }
5973
5974 //------------------------------get_table_from_crc32c_class-----------------------
5975 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5976 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
5977 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5978
5979 return table;
5980 }
5981
5982 //------------------------------inline_updateBytesCRC32C-----------------------
5983 //
5984 // Calculate CRC32C for byte[] array.
5985 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5986 //
5987 bool LibraryCallKit::inline_updateBytesCRC32C() {
5988 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5989 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5990 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5991 // no receiver since it is a static method
5992 Node* crc = argument(0); // type: int
5993 Node* src = argument(1); // type: oop
5994 Node* offset = argument(2); // type: int
5995 Node* end = argument(3); // type: int
5996
5997 Node* length = _gvn.transform(new SubINode(end, offset));
5998
5999 const Type* src_type = src->Value(&_gvn);
6000 const TypeAryPtr* top_src = src_type->isa_aryptr();
6001 if (top_src == NULL || top_src->klass() == NULL) {
6002 // failed array check
6003 return false;
6004 }
6005
6006 // Figure out the size and type of the elements we will be copying.
6007 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6008 if (src_elem != T_BYTE) {
6009 return false;
6010 }
6011
6012 // 'src_start' points to src array + scaled offset
6013 src = must_be_not_null(src, true);
6014 Node* src_start = array_element_address(src, offset, src_elem);
6015
6016 // static final int[] byteTable in class CRC32C
6017 Node* table = get_table_from_crc32c_class(callee()->holder());
6018 table = must_be_not_null(table, true);
6019 Node* table_start = array_element_address(table, intcon(0), T_INT);
6020
6021 // We assume that range check is done by caller.
6022 // TODO: generate range check (offset+length < src.length) in debug VM.
6023
6024 // Call the stub.
6025 address stubAddr = StubRoutines::updateBytesCRC32C();
6026 const char *stubName = "updateBytesCRC32C";
6027
6028 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6029 stubAddr, stubName, TypePtr::BOTTOM,
6030 crc, src_start, length, table_start);
6031 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6032 set_result(result);
6033 return true;
6034 }
6035
6036 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6037 //
6038 // Calculate CRC32C for DirectByteBuffer.
6039 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6040 //
6041 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6042 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6043 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6044 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6045 // no receiver since it is a static method
6046 Node* crc = argument(0); // type: int
6047 Node* src = argument(1); // type: long
6048 Node* offset = argument(3); // type: int
6049 Node* end = argument(4); // type: int
6050
6051 Node* length = _gvn.transform(new SubINode(end, offset));
6052
6053 src = ConvL2X(src); // adjust Java long to machine word
6054 Node* base = _gvn.transform(new CastX2PNode(src));
6055 offset = ConvI2X(offset);
6056
6057 // 'src_start' points to src array + scaled offset
6058 Node* src_start = basic_plus_adr(top(), base, offset);
6059
6060 // static final int[] byteTable in class CRC32C
6061 Node* table = get_table_from_crc32c_class(callee()->holder());
6062 table = must_be_not_null(table, true);
6063 Node* table_start = array_element_address(table, intcon(0), T_INT);
6064
6065 // Call the stub.
6066 address stubAddr = StubRoutines::updateBytesCRC32C();
6067 const char *stubName = "updateBytesCRC32C";
6068
6069 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6070 stubAddr, stubName, TypePtr::BOTTOM,
6071 crc, src_start, length, table_start);
6072 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6073 set_result(result);
6074 return true;
6075 }
6076
6077 //------------------------------inline_updateBytesAdler32----------------------
6078 //
6079 // Calculate Adler32 checksum for byte[] array.
6080 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6081 //
6082 bool LibraryCallKit::inline_updateBytesAdler32() {
6083 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6084 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6085 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6086 // no receiver since it is static method
6087 Node* crc = argument(0); // type: int
6088 Node* src = argument(1); // type: oop
6089 Node* offset = argument(2); // type: int
6090 Node* length = argument(3); // type: int
6091
6092 const Type* src_type = src->Value(&_gvn);
6093 const TypeAryPtr* top_src = src_type->isa_aryptr();
6094 if (top_src == NULL || top_src->klass() == NULL) {
6095 // failed array check
6096 return false;
6097 }
6098
6099 // Figure out the size and type of the elements we will be copying.
6100 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6101 if (src_elem != T_BYTE) {
6102 return false;
6103 }
6104
6105 // 'src_start' points to src array + scaled offset
6106 Node* src_start = array_element_address(src, offset, src_elem);
6107
6108 // We assume that range check is done by caller.
6109 // TODO: generate range check (offset+length < src.length) in debug VM.
6110
6111 // Call the stub.
6112 address stubAddr = StubRoutines::updateBytesAdler32();
6113 const char *stubName = "updateBytesAdler32";
6114
6115 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6116 stubAddr, stubName, TypePtr::BOTTOM,
6117 crc, src_start, length);
6118 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6119 set_result(result);
6120 return true;
6121 }
6122
6123 //------------------------------inline_updateByteBufferAdler32---------------
6124 //
6125 // Calculate Adler32 checksum for DirectByteBuffer.
6126 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6127 //
6128 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6129 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6130 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6131 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6132 // no receiver since it is static method
6133 Node* crc = argument(0); // type: int
6134 Node* src = argument(1); // type: long
6135 Node* offset = argument(3); // type: int
6136 Node* length = argument(4); // type: int
6137
6138 src = ConvL2X(src); // adjust Java long to machine word
6139 Node* base = _gvn.transform(new CastX2PNode(src));
6140 offset = ConvI2X(offset);
6141
6142 // 'src_start' points to src array + scaled offset
6143 Node* src_start = basic_plus_adr(top(), base, offset);
6144
6145 // Call the stub.
6146 address stubAddr = StubRoutines::updateBytesAdler32();
6147 const char *stubName = "updateBytesAdler32";
6148
6149 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6150 stubAddr, stubName, TypePtr::BOTTOM,
6151 crc, src_start, length);
6152
6153 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6154 set_result(result);
6155 return true;
6156 }
6157
6158 //----------------------------inline_reference_get----------------------------
6159 // public T java.lang.ref.Reference.get();
6160 bool LibraryCallKit::inline_reference_get() {
6161 const int referent_offset = java_lang_ref_Reference::referent_offset();
6162
6163 // Get the argument:
6164 Node* reference_obj = null_check_receiver();
6165 if (stopped()) return true;
6166
6167 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6168 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6169 decorators, /*is_static*/ false, NULL);
6170 if (result == NULL) return false;
6171
6172 // Add memory barrier to prevent commoning reads from this field
6173 // across safepoint since GC can change its value.
6174 insert_mem_bar(Op_MemBarCPUOrder);
6175
6176 set_result(result);
6177 return true;
6178 }
6179
6180 //----------------------------inline_reference_refersTo0----------------------------
6181 // bool java.lang.ref.Reference.refersTo0();
6182 // bool java.lang.ref.PhantomReference.refersTo0();
6183 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6184 // Get arguments:
6185 Node* reference_obj = null_check_receiver();
6186 Node* other_obj = argument(1);
6187 if (stopped()) return true;
6188
6189 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6190 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6191 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6192 decorators, /*is_static*/ false, NULL);
6193 if (referent == NULL) return false;
6194
6195 // Add memory barrier to prevent commoning reads from this field
6196 // across safepoint since GC can change its value.
6197 insert_mem_bar(Op_MemBarCPUOrder);
6198
6199 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6200 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6201 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6202
6203 RegionNode* region = new RegionNode(3);
6204 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6205
6206 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6207 region->init_req(1, if_true);
6208 phi->init_req(1, intcon(1));
6209
6210 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6211 region->init_req(2, if_false);
6212 phi->init_req(2, intcon(0));
6213
6214 set_control(_gvn.transform(region));
6215 record_for_igvn(region);
6216 set_result(_gvn.transform(phi));
6217 return true;
6218 }
6219
6220
6221 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6222 DecoratorSet decorators = IN_HEAP, bool is_static = false,
6223 ciInstanceKlass* fromKls = NULL) {
6224 if (fromKls == NULL) {
6225 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6226 assert(tinst != NULL, "obj is null");
6227 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6228 fromKls = tinst->klass()->as_instance_klass();
6229 } else {
6230 assert(is_static, "only for static field access");
6231 }
6232 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6233 ciSymbol::make(fieldTypeString),
6234 is_static);
6235
6236 assert (field != NULL, "undefined field");
6237 if (field == NULL) return (Node *) NULL;
6238
6239 if (is_static) {
6240 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6241 fromObj = makecon(tip);
6242 }
6243
6244 // Next code copied from Parse::do_get_xxx():
6245
6246 // Compute address and memory type.
6247 int offset = field->offset_in_bytes();
6248 bool is_vol = field->is_volatile();
6249 ciType* field_klass = field->type();
6250 assert(field_klass->is_loaded(), "should be loaded");
6251 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6252 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6253 BasicType bt = field->layout_type();
6254
6255 // Build the resultant type of the load
6256 const Type *type;
6257 if (bt == T_OBJECT) {
6258 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6259 } else {
6260 type = Type::get_const_basic_type(bt);
6261 }
6262
6263 if (is_vol) {
6264 decorators |= MO_SEQ_CST;
6265 }
6266
6267 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6268 }
6269
6270 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6271 bool is_exact = true, bool is_static = false,
6272 ciInstanceKlass * fromKls = NULL) {
6273 if (fromKls == NULL) {
6274 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6275 assert(tinst != NULL, "obj is null");
6276 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6277 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6278 fromKls = tinst->klass()->as_instance_klass();
6279 }
6280 else {
6281 assert(is_static, "only for static field access");
6282 }
6283 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6284 ciSymbol::make(fieldTypeString),
6285 is_static);
6286
6287 assert(field != NULL, "undefined field");
6288 assert(!field->is_volatile(), "not defined for volatile fields");
6289
6290 if (is_static) {
6291 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6292 fromObj = makecon(tip);
6293 }
6294
6295 // Next code copied from Parse::do_get_xxx():
6296
6297 // Compute address and memory type.
6298 int offset = field->offset_in_bytes();
6299 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6300
6301 return adr;
6302 }
6303
6304 //------------------------------inline_aescrypt_Block-----------------------
6305 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6306 address stubAddr = NULL;
6307 const char *stubName;
6308 assert(UseAES, "need AES instruction support");
6309
6310 switch(id) {
6311 case vmIntrinsics::_aescrypt_encryptBlock:
6312 stubAddr = StubRoutines::aescrypt_encryptBlock();
6313 stubName = "aescrypt_encryptBlock";
6314 break;
6315 case vmIntrinsics::_aescrypt_decryptBlock:
6316 stubAddr = StubRoutines::aescrypt_decryptBlock();
6317 stubName = "aescrypt_decryptBlock";
6318 break;
6319 default:
6320 break;
6321 }
6322 if (stubAddr == NULL) return false;
6323
6324 Node* aescrypt_object = argument(0);
6325 Node* src = argument(1);
6326 Node* src_offset = argument(2);
6327 Node* dest = argument(3);
6328 Node* dest_offset = argument(4);
6329
6330 src = must_be_not_null(src, true);
6331 dest = must_be_not_null(dest, true);
6332
6333 // (1) src and dest are arrays.
6334 const Type* src_type = src->Value(&_gvn);
6335 const Type* dest_type = dest->Value(&_gvn);
6336 const TypeAryPtr* top_src = src_type->isa_aryptr();
6337 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6338 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6339
6340 // for the quick and dirty code we will skip all the checks.
6341 // we are just trying to get the call to be generated.
6342 Node* src_start = src;
6343 Node* dest_start = dest;
6344 if (src_offset != NULL || dest_offset != NULL) {
6345 assert(src_offset != NULL && dest_offset != NULL, "");
6346 src_start = array_element_address(src, src_offset, T_BYTE);
6347 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6348 }
6349
6350 // now need to get the start of its expanded key array
6351 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6352 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6353 if (k_start == NULL) return false;
6354
6355 // Call the stub.
6356 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6357 stubAddr, stubName, TypePtr::BOTTOM,
6358 src_start, dest_start, k_start);
6359
6360 return true;
6361 }
6362
6363 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6364 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6365 address stubAddr = NULL;
6366 const char *stubName = NULL;
6367
6368 assert(UseAES, "need AES instruction support");
6369
6370 switch(id) {
6371 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6372 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6373 stubName = "cipherBlockChaining_encryptAESCrypt";
6374 break;
6375 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6376 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6377 stubName = "cipherBlockChaining_decryptAESCrypt";
6378 break;
6379 default:
6380 break;
6381 }
6382 if (stubAddr == NULL) return false;
6383
6384 Node* cipherBlockChaining_object = argument(0);
6385 Node* src = argument(1);
6386 Node* src_offset = argument(2);
6387 Node* len = argument(3);
6388 Node* dest = argument(4);
6389 Node* dest_offset = argument(5);
6390
6391 src = must_be_not_null(src, false);
6392 dest = must_be_not_null(dest, false);
6393
6394 // (1) src and dest are arrays.
6395 const Type* src_type = src->Value(&_gvn);
6396 const Type* dest_type = dest->Value(&_gvn);
6397 const TypeAryPtr* top_src = src_type->isa_aryptr();
6398 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6399 assert (top_src != NULL && top_src->klass() != NULL
6400 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6401
6402 // checks are the responsibility of the caller
6403 Node* src_start = src;
6404 Node* dest_start = dest;
6405 if (src_offset != NULL || dest_offset != NULL) {
6406 assert(src_offset != NULL && dest_offset != NULL, "");
6407 src_start = array_element_address(src, src_offset, T_BYTE);
6408 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6409 }
6410
6411 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6412 // (because of the predicated logic executed earlier).
6413 // so we cast it here safely.
6414 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6415
6416 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6417 if (embeddedCipherObj == NULL) return false;
6418
6419 // cast it to what we know it will be at runtime
6420 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6421 assert(tinst != NULL, "CBC obj is null");
6422 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6423 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6424 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6425
6426 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6427 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6428 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6429 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6430 aescrypt_object = _gvn.transform(aescrypt_object);
6431
6432 // we need to get the start of the aescrypt_object's expanded key array
6433 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6434 if (k_start == NULL) return false;
6435
6436 // similarly, get the start address of the r vector
6437 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6438 if (objRvec == NULL) return false;
6439 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6440
6441 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6442 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6443 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6444 stubAddr, stubName, TypePtr::BOTTOM,
6445 src_start, dest_start, k_start, r_start, len);
6446
6447 // return cipher length (int)
6448 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6449 set_result(retvalue);
6450 return true;
6451 }
6452
6453 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6454 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6455 address stubAddr = NULL;
6456 const char *stubName = NULL;
6457
6458 assert(UseAES, "need AES instruction support");
6459
6460 switch (id) {
6461 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6462 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6463 stubName = "electronicCodeBook_encryptAESCrypt";
6464 break;
6465 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6466 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6467 stubName = "electronicCodeBook_decryptAESCrypt";
6468 break;
6469 default:
6470 break;
6471 }
6472
6473 if (stubAddr == NULL) return false;
6474
6475 Node* electronicCodeBook_object = argument(0);
6476 Node* src = argument(1);
6477 Node* src_offset = argument(2);
6478 Node* len = argument(3);
6479 Node* dest = argument(4);
6480 Node* dest_offset = argument(5);
6481
6482 // (1) src and dest are arrays.
6483 const Type* src_type = src->Value(&_gvn);
6484 const Type* dest_type = dest->Value(&_gvn);
6485 const TypeAryPtr* top_src = src_type->isa_aryptr();
6486 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6487 assert(top_src != NULL && top_src->klass() != NULL
6488 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6489
6490 // checks are the responsibility of the caller
6491 Node* src_start = src;
6492 Node* dest_start = dest;
6493 if (src_offset != NULL || dest_offset != NULL) {
6494 assert(src_offset != NULL && dest_offset != NULL, "");
6495 src_start = array_element_address(src, src_offset, T_BYTE);
6496 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6497 }
6498
6499 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6500 // (because of the predicated logic executed earlier).
6501 // so we cast it here safely.
6502 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6503
6504 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6505 if (embeddedCipherObj == NULL) return false;
6506
6507 // cast it to what we know it will be at runtime
6508 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6509 assert(tinst != NULL, "ECB obj is null");
6510 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6511 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6512 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6513
6514 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6515 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6516 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6517 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6518 aescrypt_object = _gvn.transform(aescrypt_object);
6519
6520 // we need to get the start of the aescrypt_object's expanded key array
6521 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6522 if (k_start == NULL) return false;
6523
6524 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6525 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6526 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6527 stubAddr, stubName, TypePtr::BOTTOM,
6528 src_start, dest_start, k_start, len);
6529
6530 // return cipher length (int)
6531 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6532 set_result(retvalue);
6533 return true;
6534 }
6535
6536 //------------------------------inline_counterMode_AESCrypt-----------------------
6537 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6538 assert(UseAES, "need AES instruction support");
6539 if (!UseAESCTRIntrinsics) return false;
6540
6541 address stubAddr = NULL;
6542 const char *stubName = NULL;
6543 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6544 stubAddr = StubRoutines::counterMode_AESCrypt();
6545 stubName = "counterMode_AESCrypt";
6546 }
6547 if (stubAddr == NULL) return false;
6548
6549 Node* counterMode_object = argument(0);
6550 Node* src = argument(1);
6551 Node* src_offset = argument(2);
6552 Node* len = argument(3);
6553 Node* dest = argument(4);
6554 Node* dest_offset = argument(5);
6555
6556 // (1) src and dest are arrays.
6557 const Type* src_type = src->Value(&_gvn);
6558 const Type* dest_type = dest->Value(&_gvn);
6559 const TypeAryPtr* top_src = src_type->isa_aryptr();
6560 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6561 assert(top_src != NULL && top_src->klass() != NULL &&
6562 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6563
6564 // checks are the responsibility of the caller
6565 Node* src_start = src;
6566 Node* dest_start = dest;
6567 if (src_offset != NULL || dest_offset != NULL) {
6568 assert(src_offset != NULL && dest_offset != NULL, "");
6569 src_start = array_element_address(src, src_offset, T_BYTE);
6570 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6571 }
6572
6573 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6574 // (because of the predicated logic executed earlier).
6575 // so we cast it here safely.
6576 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6577 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6578 if (embeddedCipherObj == NULL) return false;
6579 // cast it to what we know it will be at runtime
6580 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6581 assert(tinst != NULL, "CTR obj is null");
6582 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6583 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6584 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6585 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6586 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6587 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6588 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6589 aescrypt_object = _gvn.transform(aescrypt_object);
6590 // we need to get the start of the aescrypt_object's expanded key array
6591 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6592 if (k_start == NULL) return false;
6593 // similarly, get the start address of the r vector
6594 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6595 if (obj_counter == NULL) return false;
6596 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6597
6598 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6599 if (saved_encCounter == NULL) return false;
6600 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6601 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6602
6603 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6604 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6605 OptoRuntime::counterMode_aescrypt_Type(),
6606 stubAddr, stubName, TypePtr::BOTTOM,
6607 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6608
6609 // return cipher length (int)
6610 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6611 set_result(retvalue);
6612 return true;
6613 }
6614
6615 //------------------------------get_key_start_from_aescrypt_object-----------------------
6616 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6617 #if defined(PPC64) || defined(S390)
6618 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6619 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6620 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6621 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6622 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6623 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6624 if (objSessionK == NULL) {
6625 return (Node *) NULL;
6626 }
6627 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6628 #else
6629 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6630 #endif // PPC64
6631 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6632 if (objAESCryptKey == NULL) return (Node *) NULL;
6633
6634 // now have the array, need to get the start address of the K array
6635 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6636 return k_start;
6637 }
6638
6639 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6640 // Return node representing slow path of predicate check.
6641 // the pseudo code we want to emulate with this predicate is:
6642 // for encryption:
6643 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6644 // for decryption:
6645 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6646 // note cipher==plain is more conservative than the original java code but that's OK
6647 //
6648 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6649 // The receiver was checked for NULL already.
6650 Node* objCBC = argument(0);
6651
6652 Node* src = argument(1);
6653 Node* dest = argument(4);
6654
6655 // Load embeddedCipher field of CipherBlockChaining object.
6656 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6657
6658 // get AESCrypt klass for instanceOf check
6659 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6660 // will have same classloader as CipherBlockChaining object
6661 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6662 assert(tinst != NULL, "CBCobj is null");
6663 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6664
6665 // we want to do an instanceof comparison against the AESCrypt class
6666 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6667 if (!klass_AESCrypt->is_loaded()) {
6668 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6669 Node* ctrl = control();
6670 set_control(top()); // no regular fast path
6671 return ctrl;
6672 }
6673
6674 src = must_be_not_null(src, true);
6675 dest = must_be_not_null(dest, true);
6676
6677 // Resolve oops to stable for CmpP below.
6678 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6679
6680 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6681 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6682 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6683
6684 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6685
6686 // for encryption, we are done
6687 if (!decrypting)
6688 return instof_false; // even if it is NULL
6689
6690 // for decryption, we need to add a further check to avoid
6691 // taking the intrinsic path when cipher and plain are the same
6692 // see the original java code for why.
6693 RegionNode* region = new RegionNode(3);
6694 region->init_req(1, instof_false);
6695
6696 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6697 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6698 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6699 region->init_req(2, src_dest_conjoint);
6700
6701 record_for_igvn(region);
6702 return _gvn.transform(region);
6703 }
6704
6705 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6706 // Return node representing slow path of predicate check.
6707 // the pseudo code we want to emulate with this predicate is:
6708 // for encryption:
6709 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6710 // for decryption:
6711 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6712 // note cipher==plain is more conservative than the original java code but that's OK
6713 //
6714 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6715 // The receiver was checked for NULL already.
6716 Node* objECB = argument(0);
6717
6718 // Load embeddedCipher field of ElectronicCodeBook object.
6719 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6720
6721 // get AESCrypt klass for instanceOf check
6722 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6723 // will have same classloader as ElectronicCodeBook object
6724 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6725 assert(tinst != NULL, "ECBobj is null");
6726 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6727
6728 // we want to do an instanceof comparison against the AESCrypt class
6729 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6730 if (!klass_AESCrypt->is_loaded()) {
6731 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6732 Node* ctrl = control();
6733 set_control(top()); // no regular fast path
6734 return ctrl;
6735 }
6736 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6737
6738 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6739 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6740 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6741
6742 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6743
6744 // for encryption, we are done
6745 if (!decrypting)
6746 return instof_false; // even if it is NULL
6747
6748 // for decryption, we need to add a further check to avoid
6749 // taking the intrinsic path when cipher and plain are the same
6750 // see the original java code for why.
6751 RegionNode* region = new RegionNode(3);
6752 region->init_req(1, instof_false);
6753 Node* src = argument(1);
6754 Node* dest = argument(4);
6755 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6756 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6757 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6758 region->init_req(2, src_dest_conjoint);
6759
6760 record_for_igvn(region);
6761 return _gvn.transform(region);
6762 }
6763
6764 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6765 // Return node representing slow path of predicate check.
6766 // the pseudo code we want to emulate with this predicate is:
6767 // for encryption:
6768 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6769 // for decryption:
6770 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6771 // note cipher==plain is more conservative than the original java code but that's OK
6772 //
6773
6774 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6775 // The receiver was checked for NULL already.
6776 Node* objCTR = argument(0);
6777
6778 // Load embeddedCipher field of CipherBlockChaining object.
6779 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6780
6781 // get AESCrypt klass for instanceOf check
6782 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6783 // will have same classloader as CipherBlockChaining object
6784 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6785 assert(tinst != NULL, "CTRobj is null");
6786 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6787
6788 // we want to do an instanceof comparison against the AESCrypt class
6789 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6790 if (!klass_AESCrypt->is_loaded()) {
6791 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6792 Node* ctrl = control();
6793 set_control(top()); // no regular fast path
6794 return ctrl;
6795 }
6796
6797 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6798 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6799 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6800 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6801 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6802
6803 return instof_false; // even if it is NULL
6804 }
6805
6806 //------------------------------inline_ghash_processBlocks
6807 bool LibraryCallKit::inline_ghash_processBlocks() {
6808 address stubAddr;
6809 const char *stubName;
6810 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6811
6812 stubAddr = StubRoutines::ghash_processBlocks();
6813 stubName = "ghash_processBlocks";
6814
6815 Node* data = argument(0);
6816 Node* offset = argument(1);
6817 Node* len = argument(2);
6818 Node* state = argument(3);
6819 Node* subkeyH = argument(4);
6820
6821 state = must_be_not_null(state, true);
6822 subkeyH = must_be_not_null(subkeyH, true);
6823 data = must_be_not_null(data, true);
6824
6825 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6826 assert(state_start, "state is NULL");
6827 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6828 assert(subkeyH_start, "subkeyH is NULL");
6829 Node* data_start = array_element_address(data, offset, T_BYTE);
6830 assert(data_start, "data is NULL");
6831
6832 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6833 OptoRuntime::ghash_processBlocks_Type(),
6834 stubAddr, stubName, TypePtr::BOTTOM,
6835 state_start, subkeyH_start, data_start, len);
6836 return true;
6837 }
6838
6839 bool LibraryCallKit::inline_base64_encodeBlock() {
6840 address stubAddr;
6841 const char *stubName;
6842 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6843 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6844 stubAddr = StubRoutines::base64_encodeBlock();
6845 stubName = "encodeBlock";
6846
6847 if (!stubAddr) return false;
6848 Node* base64obj = argument(0);
6849 Node* src = argument(1);
6850 Node* offset = argument(2);
6851 Node* len = argument(3);
6852 Node* dest = argument(4);
6853 Node* dp = argument(5);
6854 Node* isURL = argument(6);
6855
6856 src = must_be_not_null(src, true);
6857 dest = must_be_not_null(dest, true);
6858
6859 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6860 assert(src_start, "source array is NULL");
6861 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6862 assert(dest_start, "destination array is NULL");
6863
6864 Node* base64 = make_runtime_call(RC_LEAF,
6865 OptoRuntime::base64_encodeBlock_Type(),
6866 stubAddr, stubName, TypePtr::BOTTOM,
6867 src_start, offset, len, dest_start, dp, isURL);
6868 return true;
6869 }
6870
6871 bool LibraryCallKit::inline_base64_decodeBlock() {
6872 address stubAddr;
6873 const char *stubName;
6874 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6875 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6876 stubAddr = StubRoutines::base64_decodeBlock();
6877 stubName = "decodeBlock";
6878
6879 if (!stubAddr) return false;
6880 Node* base64obj = argument(0);
6881 Node* src = argument(1);
6882 Node* src_offset = argument(2);
6883 Node* len = argument(3);
6884 Node* dest = argument(4);
6885 Node* dest_offset = argument(5);
6886 Node* isURL = argument(6);
6887 Node* isMIME = argument(7);
6888
6889 src = must_be_not_null(src, true);
6890 dest = must_be_not_null(dest, true);
6891
6892 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6893 assert(src_start, "source array is NULL");
6894 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6895 assert(dest_start, "destination array is NULL");
6896
6897 Node* call = make_runtime_call(RC_LEAF,
6898 OptoRuntime::base64_decodeBlock_Type(),
6899 stubAddr, stubName, TypePtr::BOTTOM,
6900 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6901 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6902 set_result(result);
6903 return true;
6904 }
6905
6906 //------------------------------inline_digestBase_implCompress-----------------------
6907 //
6908 // Calculate MD5 for single-block byte[] array.
6909 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6910 //
6911 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6912 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6913 //
6914 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6915 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6916 //
6917 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6918 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6919 //
6920 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6921 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6922 //
6923 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6924 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6925
6926 Node* digestBase_obj = argument(0);
6927 Node* src = argument(1); // type oop
6928 Node* ofs = argument(2); // type int
6929
6930 const Type* src_type = src->Value(&_gvn);
6931 const TypeAryPtr* top_src = src_type->isa_aryptr();
6932 if (top_src == NULL || top_src->klass() == NULL) {
6933 // failed array check
6934 return false;
6935 }
6936 // Figure out the size and type of the elements we will be copying.
6937 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6938 if (src_elem != T_BYTE) {
6939 return false;
6940 }
6941 // 'src_start' points to src array + offset
6942 src = must_be_not_null(src, true);
6943 Node* src_start = array_element_address(src, ofs, src_elem);
6944 Node* state = NULL;
6945 Node* digest_length = NULL;
6946 address stubAddr;
6947 const char *stubName;
6948
6949 switch(id) {
6950 case vmIntrinsics::_md5_implCompress:
6951 assert(UseMD5Intrinsics, "need MD5 instruction support");
6952 state = get_state_from_digest_object(digestBase_obj, "[I");
6953 stubAddr = StubRoutines::md5_implCompress();
6954 stubName = "md5_implCompress";
6955 break;
6956 case vmIntrinsics::_sha_implCompress:
6957 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6958 state = get_state_from_digest_object(digestBase_obj, "[I");
6959 stubAddr = StubRoutines::sha1_implCompress();
6960 stubName = "sha1_implCompress";
6961 break;
6962 case vmIntrinsics::_sha2_implCompress:
6963 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6964 state = get_state_from_digest_object(digestBase_obj, "[I");
6965 stubAddr = StubRoutines::sha256_implCompress();
6966 stubName = "sha256_implCompress";
6967 break;
6968 case vmIntrinsics::_sha5_implCompress:
6969 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6970 state = get_state_from_digest_object(digestBase_obj, "[J");
6971 stubAddr = StubRoutines::sha512_implCompress();
6972 stubName = "sha512_implCompress";
6973 break;
6974 case vmIntrinsics::_sha3_implCompress:
6975 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6976 state = get_state_from_digest_object(digestBase_obj, "[B");
6977 stubAddr = StubRoutines::sha3_implCompress();
6978 stubName = "sha3_implCompress";
6979 digest_length = get_digest_length_from_digest_object(digestBase_obj);
6980 if (digest_length == NULL) return false;
6981 break;
6982 default:
6983 fatal_unexpected_iid(id);
6984 return false;
6985 }
6986 if (state == NULL) return false;
6987
6988 assert(stubAddr != NULL, "Stub is generated");
6989 if (stubAddr == NULL) return false;
6990
6991 // Call the stub.
6992 Node* call;
6993 if (digest_length == NULL) {
6994 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6995 stubAddr, stubName, TypePtr::BOTTOM,
6996 src_start, state);
6997 } else {
6998 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6999 stubAddr, stubName, TypePtr::BOTTOM,
7000 src_start, state, digest_length);
7001 }
7002
7003 return true;
7004 }
7005
7006 //------------------------------inline_digestBase_implCompressMB-----------------------
7007 //
7008 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7009 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7010 //
7011 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7012 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7013 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7014 assert((uint)predicate < 5, "sanity");
7015 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7016
7017 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7018 Node* src = argument(1); // byte[] array
7019 Node* ofs = argument(2); // type int
7020 Node* limit = argument(3); // type int
7021
7022 const Type* src_type = src->Value(&_gvn);
7023 const TypeAryPtr* top_src = src_type->isa_aryptr();
7024 if (top_src == NULL || top_src->klass() == NULL) {
7025 // failed array check
7026 return false;
7027 }
7028 // Figure out the size and type of the elements we will be copying.
7029 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
7030 if (src_elem != T_BYTE) {
7031 return false;
7032 }
7033 // 'src_start' points to src array + offset
7034 src = must_be_not_null(src, false);
7035 Node* src_start = array_element_address(src, ofs, src_elem);
7036
7037 const char* klass_digestBase_name = NULL;
7038 const char* stub_name = NULL;
7039 address stub_addr = NULL;
7040 const char* state_type = "[I";
7041
7042 switch (predicate) {
7043 case 0:
7044 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7045 klass_digestBase_name = "sun/security/provider/MD5";
7046 stub_name = "md5_implCompressMB";
7047 stub_addr = StubRoutines::md5_implCompressMB();
7048 }
7049 break;
7050 case 1:
7051 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7052 klass_digestBase_name = "sun/security/provider/SHA";
7053 stub_name = "sha1_implCompressMB";
7054 stub_addr = StubRoutines::sha1_implCompressMB();
7055 }
7056 break;
7057 case 2:
7058 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7059 klass_digestBase_name = "sun/security/provider/SHA2";
7060 stub_name = "sha256_implCompressMB";
7061 stub_addr = StubRoutines::sha256_implCompressMB();
7062 }
7063 break;
7064 case 3:
7065 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7066 klass_digestBase_name = "sun/security/provider/SHA5";
7067 stub_name = "sha512_implCompressMB";
7068 stub_addr = StubRoutines::sha512_implCompressMB();
7069 state_type = "[J";
7070 }
7071 break;
7072 case 4:
7073 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7074 klass_digestBase_name = "sun/security/provider/SHA3";
7075 stub_name = "sha3_implCompressMB";
7076 stub_addr = StubRoutines::sha3_implCompressMB();
7077 state_type = "[B";
7078 }
7079 break;
7080 default:
7081 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7082 }
7083 if (klass_digestBase_name != NULL) {
7084 assert(stub_addr != NULL, "Stub is generated");
7085 if (stub_addr == NULL) return false;
7086
7087 // get DigestBase klass to lookup for SHA klass
7088 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7089 assert(tinst != NULL, "digestBase_obj is not instance???");
7090 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7091
7092 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7093 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7094 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7095 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, state_type, stub_addr, stub_name, src_start, ofs, limit);
7096 }
7097 return false;
7098 }
7099
7100 //------------------------------inline_digestBase_implCompressMB-----------------------
7101 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7102 const char* state_type, address stubAddr, const char *stubName,
7103 Node* src_start, Node* ofs, Node* limit) {
7104 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7105 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7106 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7107 digest_obj = _gvn.transform(digest_obj);
7108
7109 Node* state = get_state_from_digest_object(digest_obj, state_type);
7110 if (state == NULL) return false;
7111
7112 Node* digest_length = NULL;
7113 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7114 digest_length = get_digest_length_from_digest_object(digest_obj);
7115 if (digest_length == NULL) return false;
7116 }
7117
7118 // Call the stub.
7119 Node* call;
7120 if (digest_length == NULL) {
7121 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7122 OptoRuntime::digestBase_implCompressMB_Type(false),
7123 stubAddr, stubName, TypePtr::BOTTOM,
7124 src_start, state, ofs, limit);
7125 } else {
7126 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7127 OptoRuntime::digestBase_implCompressMB_Type(true),
7128 stubAddr, stubName, TypePtr::BOTTOM,
7129 src_start, state, digest_length, ofs, limit);
7130 }
7131
7132 // return ofs (int)
7133 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7134 set_result(result);
7135
7136 return true;
7137 }
7138
7139 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7140 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7141 assert(UseAES, "need AES instruction support");
7142 address stubAddr = NULL;
7143 const char *stubName = NULL;
7144 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7145 stubName = "galoisCounterMode_AESCrypt";
7146
7147 if (stubAddr == NULL) return false;
7148
7149 Node* in = argument(0);
7150 Node* inOfs = argument(1);
7151 Node* len = argument(2);
7152 Node* ct = argument(3);
7153 Node* ctOfs = argument(4);
7154 Node* out = argument(5);
7155 Node* outOfs = argument(6);
7156 Node* gctr_object = argument(7);
7157 Node* ghash_object = argument(8);
7158
7159 // (1) in, ct and out are arrays.
7160 const Type* in_type = in->Value(&_gvn);
7161 const Type* ct_type = ct->Value(&_gvn);
7162 const Type* out_type = out->Value(&_gvn);
7163 const TypeAryPtr* top_in = in_type->isa_aryptr();
7164 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7165 const TypeAryPtr* top_out = out_type->isa_aryptr();
7166 assert(top_in != NULL && top_in->klass() != NULL &&
7167 top_ct != NULL && top_ct->klass() != NULL &&
7168 top_out != NULL && top_out->klass() != NULL, "args are strange");
7169
7170 // checks are the responsibility of the caller
7171 Node* in_start = in;
7172 Node* ct_start = ct;
7173 Node* out_start = out;
7174 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7175 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7176 in_start = array_element_address(in, inOfs, T_BYTE);
7177 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7178 out_start = array_element_address(out, outOfs, T_BYTE);
7179 }
7180
7181 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7182 // (because of the predicated logic executed earlier).
7183 // so we cast it here safely.
7184 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7185 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7186 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7187 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7188 Node* state = load_field_from_object(ghash_object, "state", "[J");
7189
7190 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7191 return false;
7192 }
7193 // cast it to what we know it will be at runtime
7194 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7195 assert(tinst != NULL, "GCTR obj is null");
7196 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
7197 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7198 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7199 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7200 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7201 const TypeOopPtr* xtype = aklass->as_instance_type();
7202 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7203 aescrypt_object = _gvn.transform(aescrypt_object);
7204 // we need to get the start of the aescrypt_object's expanded key array
7205 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7206 if (k_start == NULL) return false;
7207
7208 // similarly, get the start address of the r vector
7209 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7210 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7211 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7212
7213 ciKlass* klass = ciTypeArrayKlass::make(T_LONG);
7214 Node* klass_node = makecon(TypeKlassPtr::make(klass));
7215
7216 // Does this target support this intrinsic?
7217 if (Matcher::htbl_entries == -1) return false;
7218
7219 Node* subkeyHtbl_48_entries_start;
7220 if (Matcher::htbl_entries != 0) {
7221 // new array to hold 48 computed htbl entries
7222 Node* subkeyHtbl_48_entries = new_array(klass_node, intcon(Matcher::htbl_entries), 0);
7223 if (subkeyHtbl_48_entries == NULL) return false;
7224 subkeyHtbl_48_entries_start = array_element_address(subkeyHtbl_48_entries, intcon(0), T_LONG);
7225 } else {
7226 // This target doesn't need the extra-large Htbl.
7227 subkeyHtbl_48_entries_start = ConvL2X(intcon(0));
7228 }
7229
7230 // Call the stub, passing params
7231 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7232 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7233 stubAddr, stubName, TypePtr::BOTTOM,
7234 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, subkeyHtbl_48_entries_start, cnt_start);
7235
7236 // return cipher length (int)
7237 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7238 set_result(retvalue);
7239 return true;
7240 }
7241
7242 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7243 // Return node representing slow path of predicate check.
7244 // the pseudo code we want to emulate with this predicate is:
7245 // for encryption:
7246 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7247 // for decryption:
7248 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7249 // note cipher==plain is more conservative than the original java code but that's OK
7250 //
7251
7252 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7253 // The receiver was checked for NULL already.
7254 Node* objGCTR = argument(7);
7255 // Load embeddedCipher field of GCTR object.
7256 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7257 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7258
7259 // get AESCrypt klass for instanceOf check
7260 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7261 // will have same classloader as CipherBlockChaining object
7262 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7263 assert(tinst != NULL, "GCTR obj is null");
7264 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
7265
7266 // we want to do an instanceof comparison against the AESCrypt class
7267 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7268 if (!klass_AESCrypt->is_loaded()) {
7269 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7270 Node* ctrl = control();
7271 set_control(top()); // no regular fast path
7272 return ctrl;
7273 }
7274
7275 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7276 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7277 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7278 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7279 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7280
7281 return instof_false; // even if it is NULL
7282 }
7283
7284 //------------------------------get_state_from_digest_object-----------------------
7285 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, const char *state_type) {
7286 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7287 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7288 if (digest_state == NULL) return (Node *) NULL;
7289
7290 // now have the array, need to get the start address of the state array
7291 Node* state = array_element_address(digest_state, intcon(0), T_INT);
7292 return state;
7293 }
7294
7295 //------------------------------get_digest_length_from_sha3_object----------------------------------
7296 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7297 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7298 assert (digest_length != NULL, "sanity");
7299 return digest_length;
7300 }
7301
7302 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7303 // Return node representing slow path of predicate check.
7304 // the pseudo code we want to emulate with this predicate is:
7305 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7306 //
7307 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7308 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7309 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7310 assert((uint)predicate < 5, "sanity");
7311
7312 // The receiver was checked for NULL already.
7313 Node* digestBaseObj = argument(0);
7314
7315 // get DigestBase klass for instanceOf check
7316 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7317 assert(tinst != NULL, "digestBaseObj is null");
7318 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7319
7320 const char* klass_name = NULL;
7321 switch (predicate) {
7322 case 0:
7323 if (UseMD5Intrinsics) {
7324 // we want to do an instanceof comparison against the MD5 class
7325 klass_name = "sun/security/provider/MD5";
7326 }
7327 break;
7328 case 1:
7329 if (UseSHA1Intrinsics) {
7330 // we want to do an instanceof comparison against the SHA class
7331 klass_name = "sun/security/provider/SHA";
7332 }
7333 break;
7334 case 2:
7335 if (UseSHA256Intrinsics) {
7336 // we want to do an instanceof comparison against the SHA2 class
7337 klass_name = "sun/security/provider/SHA2";
7338 }
7339 break;
7340 case 3:
7341 if (UseSHA512Intrinsics) {
7342 // we want to do an instanceof comparison against the SHA5 class
7343 klass_name = "sun/security/provider/SHA5";
7344 }
7345 break;
7346 case 4:
7347 if (UseSHA3Intrinsics) {
7348 // we want to do an instanceof comparison against the SHA3 class
7349 klass_name = "sun/security/provider/SHA3";
7350 }
7351 break;
7352 default:
7353 fatal("unknown SHA intrinsic predicate: %d", predicate);
7354 }
7355
7356 ciKlass* klass = NULL;
7357 if (klass_name != NULL) {
7358 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
7359 }
7360 if ((klass == NULL) || !klass->is_loaded()) {
7361 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7362 Node* ctrl = control();
7363 set_control(top()); // no intrinsic path
7364 return ctrl;
7365 }
7366 ciInstanceKlass* instklass = klass->as_instance_klass();
7367
7368 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7369 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7370 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7371 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7372
7373 return instof_false; // even if it is NULL
7374 }
7375
7376 //-------------inline_fma-----------------------------------
7377 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7378 Node *a = NULL;
7379 Node *b = NULL;
7380 Node *c = NULL;
7381 Node* result = NULL;
7382 switch (id) {
7383 case vmIntrinsics::_fmaD:
7384 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7385 // no receiver since it is static method
7386 a = round_double_node(argument(0));
7387 b = round_double_node(argument(2));
7388 c = round_double_node(argument(4));
7389 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7390 break;
7391 case vmIntrinsics::_fmaF:
7392 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7393 a = argument(0);
7394 b = argument(1);
7395 c = argument(2);
7396 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7397 break;
7398 default:
7399 fatal_unexpected_iid(id); break;
7400 }
7401 set_result(result);
7402 return true;
7403 }
7404
7405 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7406 // argument(0) is receiver
7407 Node* codePoint = argument(1);
7408 Node* n = NULL;
7409
7410 switch (id) {
7411 case vmIntrinsics::_isDigit :
7412 n = new DigitNode(control(), codePoint);
7413 break;
7414 case vmIntrinsics::_isLowerCase :
7415 n = new LowerCaseNode(control(), codePoint);
7416 break;
7417 case vmIntrinsics::_isUpperCase :
7418 n = new UpperCaseNode(control(), codePoint);
7419 break;
7420 case vmIntrinsics::_isWhitespace :
7421 n = new WhitespaceNode(control(), codePoint);
7422 break;
7423 default:
7424 fatal_unexpected_iid(id);
7425 }
7426
7427 set_result(_gvn.transform(n));
7428 return true;
7429 }
7430
7431 //------------------------------inline_fp_min_max------------------------------
7432 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7433 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7434
7435 // The intrinsic should be used only when the API branches aren't predictable,
7436 // the last one performing the most important comparison. The following heuristic
7437 // uses the branch statistics to eventually bail out if necessary.
7438
7439 ciMethodData *md = callee()->method_data();
7440
7441 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7442 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7443
7444 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7445 // Bail out if the call-site didn't contribute enough to the statistics.
7446 return false;
7447 }
7448
7449 uint taken = 0, not_taken = 0;
7450
7451 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7452 if (p->is_BranchData()) {
7453 taken = ((ciBranchData*)p)->taken();
7454 not_taken = ((ciBranchData*)p)->not_taken();
7455 }
7456 }
7457
7458 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7459 balance = balance < 0 ? -balance : balance;
7460 if ( balance > 0.2 ) {
7461 // Bail out if the most important branch is predictable enough.
7462 return false;
7463 }
7464 }
7465 */
7466
7467 Node *a = NULL;
7468 Node *b = NULL;
7469 Node *n = NULL;
7470 switch (id) {
7471 case vmIntrinsics::_maxF:
7472 case vmIntrinsics::_minF:
7473 case vmIntrinsics::_maxF_strict:
7474 case vmIntrinsics::_minF_strict:
7475 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7476 a = argument(0);
7477 b = argument(1);
7478 break;
7479 case vmIntrinsics::_maxD:
7480 case vmIntrinsics::_minD:
7481 case vmIntrinsics::_maxD_strict:
7482 case vmIntrinsics::_minD_strict:
7483 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7484 a = round_double_node(argument(0));
7485 b = round_double_node(argument(2));
7486 break;
7487 default:
7488 fatal_unexpected_iid(id);
7489 break;
7490 }
7491 switch (id) {
7492 case vmIntrinsics::_maxF:
7493 case vmIntrinsics::_maxF_strict:
7494 n = new MaxFNode(a, b);
7495 break;
7496 case vmIntrinsics::_minF:
7497 case vmIntrinsics::_minF_strict:
7498 n = new MinFNode(a, b);
7499 break;
7500 case vmIntrinsics::_maxD:
7501 case vmIntrinsics::_maxD_strict:
7502 n = new MaxDNode(a, b);
7503 break;
7504 case vmIntrinsics::_minD:
7505 case vmIntrinsics::_minD_strict:
7506 n = new MinDNode(a, b);
7507 break;
7508 default:
7509 fatal_unexpected_iid(id);
7510 break;
7511 }
7512 set_result(_gvn.transform(n));
7513 return true;
7514 }
7515
7516 bool LibraryCallKit::inline_profileBoolean() {
7517 Node* counts = argument(1);
7518 const TypeAryPtr* ary = NULL;
7519 ciArray* aobj = NULL;
7520 if (counts->is_Con()
7521 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7522 && (aobj = ary->const_oop()->as_array()) != NULL
7523 && (aobj->length() == 2)) {
7524 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7525 jint false_cnt = aobj->element_value(0).as_int();
7526 jint true_cnt = aobj->element_value(1).as_int();
7527
7528 if (C->log() != NULL) {
7529 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7530 false_cnt, true_cnt);
7531 }
7532
7533 if (false_cnt + true_cnt == 0) {
7534 // According to profile, never executed.
7535 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7536 Deoptimization::Action_reinterpret);
7537 return true;
7538 }
7539
7540 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7541 // is a number of each value occurrences.
7542 Node* result = argument(0);
7543 if (false_cnt == 0 || true_cnt == 0) {
7544 // According to profile, one value has been never seen.
7545 int expected_val = (false_cnt == 0) ? 1 : 0;
7546
7547 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7548 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7549
7550 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7551 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7552 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7553
7554 { // Slow path: uncommon trap for never seen value and then reexecute
7555 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7556 // the value has been seen at least once.
7557 PreserveJVMState pjvms(this);
7558 PreserveReexecuteState preexecs(this);
7559 jvms()->set_should_reexecute(true);
7560
7561 set_control(slow_path);
7562 set_i_o(i_o());
7563
7564 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7565 Deoptimization::Action_reinterpret);
7566 }
7567 // The guard for never seen value enables sharpening of the result and
7568 // returning a constant. It allows to eliminate branches on the same value
7569 // later on.
7570 set_control(fast_path);
7571 result = intcon(expected_val);
7572 }
7573 // Stop profiling.
7574 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7575 // By replacing method body with profile data (represented as ProfileBooleanNode
7576 // on IR level) we effectively disable profiling.
7577 // It enables full speed execution once optimized code is generated.
7578 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7579 C->record_for_igvn(profile);
7580 set_result(profile);
7581 return true;
7582 } else {
7583 // Continue profiling.
7584 // Profile data isn't available at the moment. So, execute method's bytecode version.
7585 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7586 // is compiled and counters aren't available since corresponding MethodHandle
7587 // isn't a compile-time constant.
7588 return false;
7589 }
7590 }
7591
7592 bool LibraryCallKit::inline_isCompileConstant() {
7593 Node* n = argument(0);
7594 set_result(n->is_Con() ? intcon(1) : intcon(0));
7595 return true;
7596 }
7597
7598 //------------------------------- inline_getObjectSize --------------------------------------
7599 //
7600 // Calculate the runtime size of the object/array.
7601 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7602 //
7603 bool LibraryCallKit::inline_getObjectSize() {
7604 Node* obj = argument(3);
7605 Node* klass_node = load_object_klass(obj);
7606
7607 jint layout_con = Klass::_lh_neutral_value;
7608 Node* layout_val = get_layout_helper(klass_node, layout_con);
7609 int layout_is_con = (layout_val == NULL);
7610
7611 if (layout_is_con) {
7612 // Layout helper is constant, can figure out things at compile time.
7613
7614 if (Klass::layout_helper_is_instance(layout_con)) {
7615 // Instance case: layout_con contains the size itself.
7616 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7617 set_result(size);
7618 } else {
7619 // Array case: size is round(header + element_size*arraylength).
7620 // Since arraylength is different for every array instance, we have to
7621 // compute the whole thing at runtime.
7622
7623 Node* arr_length = load_array_length(obj);
7624
7625 int round_mask = MinObjAlignmentInBytes - 1;
7626 int hsize = Klass::layout_helper_header_size(layout_con);
7627 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7628
7629 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7630 round_mask = 0; // strength-reduce it if it goes away completely
7631 }
7632 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7633 Node* header_size = intcon(hsize + round_mask);
7634
7635 Node* lengthx = ConvI2X(arr_length);
7636 Node* headerx = ConvI2X(header_size);
7637
7638 Node* abody = lengthx;
7639 if (eshift != 0) {
7640 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7641 }
7642 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7643 if (round_mask != 0) {
7644 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7645 }
7646 size = ConvX2L(size);
7647 set_result(size);
7648 }
7649 } else {
7650 // Layout helper is not constant, need to test for array-ness at runtime.
7651
7652 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7653 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7654 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7655 record_for_igvn(result_reg);
7656
7657 Node* array_ctl = generate_array_guard(klass_node, NULL);
7658 if (array_ctl != NULL) {
7659 // Array case: size is round(header + element_size*arraylength).
7660 // Since arraylength is different for every array instance, we have to
7661 // compute the whole thing at runtime.
7662
7663 PreserveJVMState pjvms(this);
7664 set_control(array_ctl);
7665 Node* arr_length = load_array_length(obj);
7666
7667 int round_mask = MinObjAlignmentInBytes - 1;
7668 Node* mask = intcon(round_mask);
7669
7670 Node* hss = intcon(Klass::_lh_header_size_shift);
7671 Node* hsm = intcon(Klass::_lh_header_size_mask);
7672 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7673 header_size = _gvn.transform(new AndINode(header_size, hsm));
7674 header_size = _gvn.transform(new AddINode(header_size, mask));
7675
7676 // There is no need to mask or shift this value.
7677 // The semantics of LShiftINode include an implicit mask to 0x1F.
7678 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7679 Node* elem_shift = layout_val;
7680
7681 Node* lengthx = ConvI2X(arr_length);
7682 Node* headerx = ConvI2X(header_size);
7683
7684 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7685 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7686 if (round_mask != 0) {
7687 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7688 }
7689 size = ConvX2L(size);
7690
7691 result_reg->init_req(_array_path, control());
7692 result_val->init_req(_array_path, size);
7693 }
7694
7695 if (!stopped()) {
7696 // Instance case: the layout helper gives us instance size almost directly,
7697 // but we need to mask out the _lh_instance_slow_path_bit.
7698 Node* size = ConvI2X(layout_val);
7699 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7700 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7701 size = _gvn.transform(new AndXNode(size, mask));
7702 size = ConvX2L(size);
7703
7704 result_reg->init_req(_instance_path, control());
7705 result_val->init_req(_instance_path, size);
7706 }
7707
7708 set_result(result_reg, result_val);
7709 }
7710
7711 return true;
7712 }
7713
7714 //------------------------------- inline_blackhole --------------------------------------
7715 //
7716 // Make sure all arguments to this node are alive.
7717 // This matches methods that were requested to be blackholed through compile commands.
7718 //
7719 bool LibraryCallKit::inline_blackhole() {
7720 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7721 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7722 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7723
7724 // Bind call arguments as blackhole arguments to keep them alive
7725 Node* bh = insert_mem_bar(Op_Blackhole);
7726 uint nargs = callee()->arg_size();
7727 for (uint i = 0; i < nargs; i++) {
7728 bh->add_req(argument(i));
7729 }
7730
7731 return true;
7732 }