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::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
273
274 case vmIntrinsics::_min:
275 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
276
277 case vmIntrinsics::_notify:
278 case vmIntrinsics::_notifyAll:
279 return inline_notify(intrinsic_id());
280
281 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
282 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
283 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
284 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
285 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
286 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
287 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
288 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
289 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
298 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
299 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
300 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
301
302 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
309 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
310
311 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
312 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
313
314 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
315 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
316 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
317 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
318
319 case vmIntrinsics::_compressStringC:
320 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
321 case vmIntrinsics::_inflateStringC:
322 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
323
324 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
325 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
326 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
327 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
328 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
329 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
330 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
331 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
332 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
333 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
334 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
335 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_INLINE_TYPE,Relaxed, false);
336
337 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
338 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
339 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
340 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
341 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
342 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
343 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
344 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
345 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
346 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_INLINE_TYPE,Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
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::_maxF:
642 case vmIntrinsics::_minF:
643 case vmIntrinsics::_maxD:
644 case vmIntrinsics::_minD:
645 return inline_fp_min_max(intrinsic_id());
646
647 case vmIntrinsics::_VectorUnaryOp:
648 return inline_vector_nary_operation(1);
649 case vmIntrinsics::_VectorBinaryOp:
650 return inline_vector_nary_operation(2);
651 case vmIntrinsics::_VectorTernaryOp:
652 return inline_vector_nary_operation(3);
653 case vmIntrinsics::_VectorBroadcastCoerced:
654 return inline_vector_broadcast_coerced();
655 case vmIntrinsics::_VectorShuffleIota:
656 return inline_vector_shuffle_iota();
657 case vmIntrinsics::_VectorMaskOp:
658 return inline_vector_mask_operation();
659 case vmIntrinsics::_VectorShuffleToVector:
660 return inline_vector_shuffle_to_vector();
661 case vmIntrinsics::_VectorLoadOp:
662 return inline_vector_mem_operation(/*is_store=*/false);
663 case vmIntrinsics::_VectorStoreOp:
664 return inline_vector_mem_operation(/*is_store=*/true);
665 case vmIntrinsics::_VectorGatherOp:
666 return inline_vector_gather_scatter(/*is_scatter*/ false);
667 case vmIntrinsics::_VectorScatterOp:
668 return inline_vector_gather_scatter(/*is_scatter*/ true);
669 case vmIntrinsics::_VectorReductionCoerced:
670 return inline_vector_reduction();
671 case vmIntrinsics::_VectorTest:
672 return inline_vector_test();
673 case vmIntrinsics::_VectorBlend:
674 return inline_vector_blend();
675 case vmIntrinsics::_VectorRearrange:
676 return inline_vector_rearrange();
677 case vmIntrinsics::_VectorCompare:
678 return inline_vector_compare();
679 case vmIntrinsics::_VectorBroadcastInt:
680 return inline_vector_broadcast_int();
681 case vmIntrinsics::_VectorConvert:
682 return inline_vector_convert();
683 case vmIntrinsics::_VectorInsert:
684 return inline_vector_insert();
685 case vmIntrinsics::_VectorExtract:
686 return inline_vector_extract();
687
688 case vmIntrinsics::_getObjectSize:
689 return inline_getObjectSize();
690
691 case vmIntrinsics::_blackhole:
692 return inline_blackhole();
693
694 default:
695 // If you get here, it may be that someone has added a new intrinsic
696 // to the list in vmIntrinsics.hpp without implementing it here.
697 #ifndef PRODUCT
698 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
699 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
700 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
701 }
702 #endif
703 return false;
704 }
705 }
706
707 Node* LibraryCallKit::try_to_predicate(int predicate) {
708 if (!jvms()->has_method()) {
709 // Root JVMState has a null method.
710 assert(map()->memory()->Opcode() == Op_Parm, "");
711 // Insert the memory aliasing node
712 set_all_memory(reset_memory());
713 }
714 assert(merged_memory(), "");
715
716 switch (intrinsic_id()) {
717 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
718 return inline_cipherBlockChaining_AESCrypt_predicate(false);
719 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
720 return inline_cipherBlockChaining_AESCrypt_predicate(true);
721 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
722 return inline_electronicCodeBook_AESCrypt_predicate(false);
723 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
724 return inline_electronicCodeBook_AESCrypt_predicate(true);
725 case vmIntrinsics::_counterMode_AESCrypt:
726 return inline_counterMode_AESCrypt_predicate();
727 case vmIntrinsics::_digestBase_implCompressMB:
728 return inline_digestBase_implCompressMB_predicate(predicate);
729 case vmIntrinsics::_galoisCounterMode_AESCrypt:
730 return inline_galoisCounterMode_AESCrypt_predicate();
731
732 default:
733 // If you get here, it may be that someone has added a new intrinsic
734 // to the list in vmIntrinsics.hpp without implementing it here.
735 #ifndef PRODUCT
736 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
737 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
738 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
739 }
740 #endif
741 Node* slow_ctl = control();
742 set_control(top()); // No fast path instrinsic
743 return slow_ctl;
744 }
745 }
746
747 //------------------------------set_result-------------------------------
748 // Helper function for finishing intrinsics.
749 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
750 record_for_igvn(region);
751 set_control(_gvn.transform(region));
752 set_result( _gvn.transform(value));
753 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
754 }
755
756 //------------------------------generate_guard---------------------------
757 // Helper function for generating guarded fast-slow graph structures.
758 // The given 'test', if true, guards a slow path. If the test fails
759 // then a fast path can be taken. (We generally hope it fails.)
760 // In all cases, GraphKit::control() is updated to the fast path.
761 // The returned value represents the control for the slow path.
762 // The return value is never 'top'; it is either a valid control
763 // or NULL if it is obvious that the slow path can never be taken.
764 // Also, if region and the slow control are not NULL, the slow edge
765 // is appended to the region.
766 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
767 if (stopped()) {
768 // Already short circuited.
769 return NULL;
770 }
771
772 // Build an if node and its projections.
773 // If test is true we take the slow path, which we assume is uncommon.
774 if (_gvn.type(test) == TypeInt::ZERO) {
775 // The slow branch is never taken. No need to build this guard.
776 return NULL;
777 }
778
779 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
780
781 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
782 if (if_slow == top()) {
783 // The slow branch is never taken. No need to build this guard.
784 return NULL;
785 }
786
787 if (region != NULL)
788 region->add_req(if_slow);
789
790 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
791 set_control(if_fast);
792
793 return if_slow;
794 }
795
796 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
797 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
798 }
799 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
800 return generate_guard(test, region, PROB_FAIR);
801 }
802
803 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
804 Node* *pos_index) {
805 if (stopped())
806 return NULL; // already stopped
807 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
808 return NULL; // index is already adequately typed
809 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
810 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
811 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
812 if (is_neg != NULL && pos_index != NULL) {
813 // Emulate effect of Parse::adjust_map_after_if.
814 Node* ccast = new CastIINode(index, TypeInt::POS);
815 ccast->set_req(0, control());
816 (*pos_index) = _gvn.transform(ccast);
817 }
818 return is_neg;
819 }
820
821 // Make sure that 'position' is a valid limit index, in [0..length].
822 // There are two equivalent plans for checking this:
823 // A. (offset + copyLength) unsigned<= arrayLength
824 // B. offset <= (arrayLength - copyLength)
825 // We require that all of the values above, except for the sum and
826 // difference, are already known to be non-negative.
827 // Plan A is robust in the face of overflow, if offset and copyLength
828 // are both hugely positive.
829 //
830 // Plan B is less direct and intuitive, but it does not overflow at
831 // all, since the difference of two non-negatives is always
832 // representable. Whenever Java methods must perform the equivalent
833 // check they generally use Plan B instead of Plan A.
834 // For the moment we use Plan A.
835 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
836 Node* subseq_length,
837 Node* array_length,
838 RegionNode* region) {
839 if (stopped())
840 return NULL; // already stopped
841 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
842 if (zero_offset && subseq_length->eqv_uncast(array_length))
843 return NULL; // common case of whole-array copy
844 Node* last = subseq_length;
845 if (!zero_offset) // last += offset
846 last = _gvn.transform(new AddINode(last, offset));
847 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
848 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
849 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
850 return is_over;
851 }
852
853 // Emit range checks for the given String.value byte array
854 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
855 if (stopped()) {
856 return; // already stopped
857 }
858 RegionNode* bailout = new RegionNode(1);
859 record_for_igvn(bailout);
860 if (char_count) {
861 // Convert char count to byte count
862 count = _gvn.transform(new LShiftINode(count, intcon(1)));
863 }
864
865 // Offset and count must not be negative
866 generate_negative_guard(offset, bailout);
867 generate_negative_guard(count, bailout);
868 // Offset + count must not exceed length of array
869 generate_limit_guard(offset, count, load_array_length(array), bailout);
870
871 if (bailout->req() > 1) {
872 PreserveJVMState pjvms(this);
873 set_control(_gvn.transform(bailout));
874 uncommon_trap(Deoptimization::Reason_intrinsic,
875 Deoptimization::Action_maybe_recompile);
876 }
877 }
878
879 //--------------------------generate_current_thread--------------------
880 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
881 ciKlass* thread_klass = env()->Thread_klass();
882 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
883 Node* thread = _gvn.transform(new ThreadLocalNode());
884 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
885 tls_output = thread;
886 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
887 thread_obj_handle = _gvn.transform(thread_obj_handle);
888 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
889 }
890
891
892 //------------------------------make_string_method_node------------------------
893 // Helper method for String intrinsic functions. This version is called with
894 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
895 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
896 // containing the lengths of str1 and str2.
897 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
898 Node* result = NULL;
899 switch (opcode) {
900 case Op_StrIndexOf:
901 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
902 str1_start, cnt1, str2_start, cnt2, ae);
903 break;
904 case Op_StrComp:
905 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
906 str1_start, cnt1, str2_start, cnt2, ae);
907 break;
908 case Op_StrEquals:
909 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
910 // Use the constant length if there is one because optimized match rule may exist.
911 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
912 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
913 break;
914 default:
915 ShouldNotReachHere();
916 return NULL;
917 }
918
919 // All these intrinsics have checks.
920 C->set_has_split_ifs(true); // Has chance for split-if optimization
921 clear_upper_avx();
922
923 return _gvn.transform(result);
924 }
925
926 //------------------------------inline_string_compareTo------------------------
927 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
928 Node* arg1 = argument(0);
929 Node* arg2 = argument(1);
930
931 arg1 = must_be_not_null(arg1, true);
932 arg2 = must_be_not_null(arg2, true);
933
934 // Get start addr and length of first argument
935 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
936 Node* arg1_cnt = load_array_length(arg1);
937
938 // Get start addr and length of second argument
939 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
940 Node* arg2_cnt = load_array_length(arg2);
941
942 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
943 set_result(result);
944 return true;
945 }
946
947 //------------------------------inline_string_equals------------------------
948 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
949 Node* arg1 = argument(0);
950 Node* arg2 = argument(1);
951
952 // paths (plus control) merge
953 RegionNode* region = new RegionNode(3);
954 Node* phi = new PhiNode(region, TypeInt::BOOL);
955
956 if (!stopped()) {
957
958 arg1 = must_be_not_null(arg1, true);
959 arg2 = must_be_not_null(arg2, true);
960
961 // Get start addr and length of first argument
962 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
963 Node* arg1_cnt = load_array_length(arg1);
964
965 // Get start addr and length of second argument
966 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
967 Node* arg2_cnt = load_array_length(arg2);
968
969 // Check for arg1_cnt != arg2_cnt
970 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
971 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
972 Node* if_ne = generate_slow_guard(bol, NULL);
973 if (if_ne != NULL) {
974 phi->init_req(2, intcon(0));
975 region->init_req(2, if_ne);
976 }
977
978 // Check for count == 0 is done by assembler code for StrEquals.
979
980 if (!stopped()) {
981 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
982 phi->init_req(1, equals);
983 region->init_req(1, control());
984 }
985 }
986
987 // post merge
988 set_control(_gvn.transform(region));
989 record_for_igvn(region);
990
991 set_result(_gvn.transform(phi));
992 return true;
993 }
994
995 //------------------------------inline_array_equals----------------------------
996 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
997 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
998 Node* arg1 = argument(0);
999 Node* arg2 = argument(1);
1000
1001 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1002 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1003 clear_upper_avx();
1004
1005 return true;
1006 }
1007
1008 //------------------------------inline_hasNegatives------------------------------
1009 bool LibraryCallKit::inline_hasNegatives() {
1010 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1011 return false;
1012 }
1013
1014 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1015 // no receiver since it is static method
1016 Node* ba = argument(0);
1017 Node* offset = argument(1);
1018 Node* len = argument(2);
1019
1020 ba = must_be_not_null(ba, true);
1021
1022 // Range checks
1023 generate_string_range_check(ba, offset, len, false);
1024 if (stopped()) {
1025 return true;
1026 }
1027 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1028 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1029 set_result(_gvn.transform(result));
1030 return true;
1031 }
1032
1033 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1034 Node* index = argument(0);
1035 Node* length = bt == T_INT ? argument(1) : argument(2);
1036 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1037 return false;
1038 }
1039
1040 // check that length is positive
1041 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1042 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1043
1044 {
1045 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1046 uncommon_trap(Deoptimization::Reason_intrinsic,
1047 Deoptimization::Action_make_not_entrant);
1048 }
1049
1050 if (stopped()) {
1051 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1052 return true;
1053 }
1054
1055 // length is now known postive, add a cast node to make this explicit
1056 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1057 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1058 casted_length = _gvn.transform(casted_length);
1059 replace_in_map(length, casted_length);
1060 length = casted_length;
1061
1062 // Use an unsigned comparison for the range check itself
1063 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1064 BoolTest::mask btest = BoolTest::lt;
1065 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1066 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1067 _gvn.set_type(rc, rc->Value(&_gvn));
1068 if (!rc_bool->is_Con()) {
1069 record_for_igvn(rc);
1070 }
1071 set_control(_gvn.transform(new IfTrueNode(rc)));
1072 {
1073 PreserveJVMState pjvms(this);
1074 set_control(_gvn.transform(new IfFalseNode(rc)));
1075 uncommon_trap(Deoptimization::Reason_range_check,
1076 Deoptimization::Action_make_not_entrant);
1077 }
1078
1079 if (stopped()) {
1080 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1081 return true;
1082 }
1083
1084 // index is now known to be >= 0 and < length, cast it
1085 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1086 result = _gvn.transform(result);
1087 set_result(result);
1088 replace_in_map(index, result);
1089 clear_upper_avx();
1090 return true;
1091 }
1092
1093 //------------------------------inline_string_indexOf------------------------
1094 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1095 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1096 return false;
1097 }
1098 Node* src = argument(0);
1099 Node* tgt = argument(1);
1100
1101 // Make the merge point
1102 RegionNode* result_rgn = new RegionNode(4);
1103 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1104
1105 src = must_be_not_null(src, true);
1106 tgt = must_be_not_null(tgt, true);
1107
1108 // Get start addr and length of source string
1109 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1110 Node* src_count = load_array_length(src);
1111
1112 // Get start addr and length of substring
1113 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1114 Node* tgt_count = load_array_length(tgt);
1115
1116 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1117 // Divide src size by 2 if String is UTF16 encoded
1118 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1119 }
1120 if (ae == StrIntrinsicNode::UU) {
1121 // Divide substring size by 2 if String is UTF16 encoded
1122 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1123 }
1124
1125 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1126 if (result != NULL) {
1127 result_phi->init_req(3, result);
1128 result_rgn->init_req(3, control());
1129 }
1130 set_control(_gvn.transform(result_rgn));
1131 record_for_igvn(result_rgn);
1132 set_result(_gvn.transform(result_phi));
1133
1134 return true;
1135 }
1136
1137 //-----------------------------inline_string_indexOf-----------------------
1138 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1139 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1140 return false;
1141 }
1142 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1143 return false;
1144 }
1145 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1146 Node* src = argument(0); // byte[]
1147 Node* src_count = argument(1); // char count
1148 Node* tgt = argument(2); // byte[]
1149 Node* tgt_count = argument(3); // char count
1150 Node* from_index = argument(4); // char index
1151
1152 src = must_be_not_null(src, true);
1153 tgt = must_be_not_null(tgt, true);
1154
1155 // Multiply byte array index by 2 if String is UTF16 encoded
1156 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1157 src_count = _gvn.transform(new SubINode(src_count, from_index));
1158 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1159 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1160
1161 // Range checks
1162 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1163 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1164 if (stopped()) {
1165 return true;
1166 }
1167
1168 RegionNode* region = new RegionNode(5);
1169 Node* phi = new PhiNode(region, TypeInt::INT);
1170
1171 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1172 if (result != NULL) {
1173 // The result is index relative to from_index if substring was found, -1 otherwise.
1174 // Generate code which will fold into cmove.
1175 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1176 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1177
1178 Node* if_lt = generate_slow_guard(bol, NULL);
1179 if (if_lt != NULL) {
1180 // result == -1
1181 phi->init_req(3, result);
1182 region->init_req(3, if_lt);
1183 }
1184 if (!stopped()) {
1185 result = _gvn.transform(new AddINode(result, from_index));
1186 phi->init_req(4, result);
1187 region->init_req(4, control());
1188 }
1189 }
1190
1191 set_control(_gvn.transform(region));
1192 record_for_igvn(region);
1193 set_result(_gvn.transform(phi));
1194 clear_upper_avx();
1195
1196 return true;
1197 }
1198
1199 // Create StrIndexOfNode with fast path checks
1200 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1201 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1202 // Check for substr count > string count
1203 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1204 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1205 Node* if_gt = generate_slow_guard(bol, NULL);
1206 if (if_gt != NULL) {
1207 phi->init_req(1, intcon(-1));
1208 region->init_req(1, if_gt);
1209 }
1210 if (!stopped()) {
1211 // Check for substr count == 0
1212 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1213 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1214 Node* if_zero = generate_slow_guard(bol, NULL);
1215 if (if_zero != NULL) {
1216 phi->init_req(2, intcon(0));
1217 region->init_req(2, if_zero);
1218 }
1219 }
1220 if (!stopped()) {
1221 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1222 }
1223 return NULL;
1224 }
1225
1226 //-----------------------------inline_string_indexOfChar-----------------------
1227 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1228 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1229 return false;
1230 }
1231 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1232 return false;
1233 }
1234 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1235 Node* src = argument(0); // byte[]
1236 Node* tgt = argument(1); // tgt is int ch
1237 Node* from_index = argument(2);
1238 Node* max = argument(3);
1239
1240 src = must_be_not_null(src, true);
1241
1242 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1243 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1244 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1245
1246 // Range checks
1247 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1248 if (stopped()) {
1249 return true;
1250 }
1251
1252 RegionNode* region = new RegionNode(3);
1253 Node* phi = new PhiNode(region, TypeInt::INT);
1254
1255 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1256 C->set_has_split_ifs(true); // Has chance for split-if optimization
1257 _gvn.transform(result);
1258
1259 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1260 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1261
1262 Node* if_lt = generate_slow_guard(bol, NULL);
1263 if (if_lt != NULL) {
1264 // result == -1
1265 phi->init_req(2, result);
1266 region->init_req(2, if_lt);
1267 }
1268 if (!stopped()) {
1269 result = _gvn.transform(new AddINode(result, from_index));
1270 phi->init_req(1, result);
1271 region->init_req(1, control());
1272 }
1273 set_control(_gvn.transform(region));
1274 record_for_igvn(region);
1275 set_result(_gvn.transform(phi));
1276
1277 return true;
1278 }
1279 //---------------------------inline_string_copy---------------------
1280 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1281 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1282 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1283 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1284 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1285 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1286 bool LibraryCallKit::inline_string_copy(bool compress) {
1287 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1288 return false;
1289 }
1290 int nargs = 5; // 2 oops, 3 ints
1291 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1292
1293 Node* src = argument(0);
1294 Node* src_offset = argument(1);
1295 Node* dst = argument(2);
1296 Node* dst_offset = argument(3);
1297 Node* length = argument(4);
1298
1299 // Check for allocation before we add nodes that would confuse
1300 // tightly_coupled_allocation()
1301 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1302
1303 // Figure out the size and type of the elements we will be copying.
1304 const Type* src_type = src->Value(&_gvn);
1305 const Type* dst_type = dst->Value(&_gvn);
1306 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1307 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1308 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1309 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1310 "Unsupported array types for inline_string_copy");
1311
1312 src = must_be_not_null(src, true);
1313 dst = must_be_not_null(dst, true);
1314
1315 // Convert char[] offsets to byte[] offsets
1316 bool convert_src = (compress && src_elem == T_BYTE);
1317 bool convert_dst = (!compress && dst_elem == T_BYTE);
1318 if (convert_src) {
1319 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1320 } else if (convert_dst) {
1321 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1322 }
1323
1324 // Range checks
1325 generate_string_range_check(src, src_offset, length, convert_src);
1326 generate_string_range_check(dst, dst_offset, length, convert_dst);
1327 if (stopped()) {
1328 return true;
1329 }
1330
1331 Node* src_start = array_element_address(src, src_offset, src_elem);
1332 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1333 // 'src_start' points to src array + scaled offset
1334 // 'dst_start' points to dst array + scaled offset
1335 Node* count = NULL;
1336 if (compress) {
1337 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1338 } else {
1339 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1340 }
1341
1342 if (alloc != NULL) {
1343 if (alloc->maybe_set_complete(&_gvn)) {
1344 // "You break it, you buy it."
1345 InitializeNode* init = alloc->initialization();
1346 assert(init->is_complete(), "we just did this");
1347 init->set_complete_with_arraycopy();
1348 assert(dst->is_CheckCastPP(), "sanity");
1349 assert(dst->in(0)->in(0) == init, "dest pinned");
1350 }
1351 // Do not let stores that initialize this object be reordered with
1352 // a subsequent store that would make this object accessible by
1353 // other threads.
1354 // Record what AllocateNode this StoreStore protects so that
1355 // escape analysis can go from the MemBarStoreStoreNode to the
1356 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1357 // based on the escape status of the AllocateNode.
1358 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1359 }
1360 if (compress) {
1361 set_result(_gvn.transform(count));
1362 }
1363 clear_upper_avx();
1364
1365 return true;
1366 }
1367
1368 #ifdef _LP64
1369 #define XTOP ,top() /*additional argument*/
1370 #else //_LP64
1371 #define XTOP /*no additional argument*/
1372 #endif //_LP64
1373
1374 //------------------------inline_string_toBytesU--------------------------
1375 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1376 bool LibraryCallKit::inline_string_toBytesU() {
1377 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1378 return false;
1379 }
1380 // Get the arguments.
1381 Node* value = argument(0);
1382 Node* offset = argument(1);
1383 Node* length = argument(2);
1384
1385 Node* newcopy = NULL;
1386
1387 // Set the original stack and the reexecute bit for the interpreter to reexecute
1388 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1389 { PreserveReexecuteState preexecs(this);
1390 jvms()->set_should_reexecute(true);
1391
1392 // Check if a null path was taken unconditionally.
1393 value = null_check(value);
1394
1395 RegionNode* bailout = new RegionNode(1);
1396 record_for_igvn(bailout);
1397
1398 // Range checks
1399 generate_negative_guard(offset, bailout);
1400 generate_negative_guard(length, bailout);
1401 generate_limit_guard(offset, length, load_array_length(value), bailout);
1402 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1403 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1404
1405 if (bailout->req() > 1) {
1406 PreserveJVMState pjvms(this);
1407 set_control(_gvn.transform(bailout));
1408 uncommon_trap(Deoptimization::Reason_intrinsic,
1409 Deoptimization::Action_maybe_recompile);
1410 }
1411 if (stopped()) {
1412 return true;
1413 }
1414
1415 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1416 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1417 newcopy = new_array(klass_node, size, 0); // no arguments to push
1418 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1419 guarantee(alloc != NULL, "created above");
1420
1421 // Calculate starting addresses.
1422 Node* src_start = array_element_address(value, offset, T_CHAR);
1423 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1424
1425 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1426 const TypeInt* toffset = gvn().type(offset)->is_int();
1427 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1428
1429 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1430 const char* copyfunc_name = "arraycopy";
1431 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1432 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1433 OptoRuntime::fast_arraycopy_Type(),
1434 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1435 src_start, dst_start, ConvI2X(length) XTOP);
1436 // Do not let reads from the cloned object float above the arraycopy.
1437 if (alloc->maybe_set_complete(&_gvn)) {
1438 // "You break it, you buy it."
1439 InitializeNode* init = alloc->initialization();
1440 assert(init->is_complete(), "we just did this");
1441 init->set_complete_with_arraycopy();
1442 assert(newcopy->is_CheckCastPP(), "sanity");
1443 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1444 }
1445 // Do not let stores that initialize this object be reordered with
1446 // a subsequent store that would make this object accessible by
1447 // other threads.
1448 // Record what AllocateNode this StoreStore protects so that
1449 // escape analysis can go from the MemBarStoreStoreNode to the
1450 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1451 // based on the escape status of the AllocateNode.
1452 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1453 } // original reexecute is set back here
1454
1455 C->set_has_split_ifs(true); // Has chance for split-if optimization
1456 if (!stopped()) {
1457 set_result(newcopy);
1458 }
1459 clear_upper_avx();
1460
1461 return true;
1462 }
1463
1464 //------------------------inline_string_getCharsU--------------------------
1465 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1466 bool LibraryCallKit::inline_string_getCharsU() {
1467 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1468 return false;
1469 }
1470
1471 // Get the arguments.
1472 Node* src = argument(0);
1473 Node* src_begin = argument(1);
1474 Node* src_end = argument(2); // exclusive offset (i < src_end)
1475 Node* dst = argument(3);
1476 Node* dst_begin = argument(4);
1477
1478 // Check for allocation before we add nodes that would confuse
1479 // tightly_coupled_allocation()
1480 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1481
1482 // Check if a null path was taken unconditionally.
1483 src = null_check(src);
1484 dst = null_check(dst);
1485 if (stopped()) {
1486 return true;
1487 }
1488
1489 // Get length and convert char[] offset to byte[] offset
1490 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1491 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1492
1493 // Range checks
1494 generate_string_range_check(src, src_begin, length, true);
1495 generate_string_range_check(dst, dst_begin, length, false);
1496 if (stopped()) {
1497 return true;
1498 }
1499
1500 if (!stopped()) {
1501 // Calculate starting addresses.
1502 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1503 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1504
1505 // Check if array addresses are aligned to HeapWordSize
1506 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1507 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1508 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1509 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1510
1511 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1512 const char* copyfunc_name = "arraycopy";
1513 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1514 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1515 OptoRuntime::fast_arraycopy_Type(),
1516 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1517 src_start, dst_start, ConvI2X(length) XTOP);
1518 // Do not let reads from the cloned object float above the arraycopy.
1519 if (alloc != NULL) {
1520 if (alloc->maybe_set_complete(&_gvn)) {
1521 // "You break it, you buy it."
1522 InitializeNode* init = alloc->initialization();
1523 assert(init->is_complete(), "we just did this");
1524 init->set_complete_with_arraycopy();
1525 assert(dst->is_CheckCastPP(), "sanity");
1526 assert(dst->in(0)->in(0) == init, "dest pinned");
1527 }
1528 // Do not let stores that initialize this object be reordered with
1529 // a subsequent store that would make this object accessible by
1530 // other threads.
1531 // Record what AllocateNode this StoreStore protects so that
1532 // escape analysis can go from the MemBarStoreStoreNode to the
1533 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1534 // based on the escape status of the AllocateNode.
1535 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1536 } else {
1537 insert_mem_bar(Op_MemBarCPUOrder);
1538 }
1539 }
1540
1541 C->set_has_split_ifs(true); // Has chance for split-if optimization
1542 return true;
1543 }
1544
1545 //----------------------inline_string_char_access----------------------------
1546 // Store/Load char to/from byte[] array.
1547 // static void StringUTF16.putChar(byte[] val, int index, int c)
1548 // static char StringUTF16.getChar(byte[] val, int index)
1549 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1550 Node* value = argument(0);
1551 Node* index = argument(1);
1552 Node* ch = is_store ? argument(2) : NULL;
1553
1554 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1555 // correctly requires matched array shapes.
1556 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1557 "sanity: byte[] and char[] bases agree");
1558 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1559 "sanity: byte[] and char[] scales agree");
1560
1561 // Bail when getChar over constants is requested: constant folding would
1562 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1563 // Java method would constant fold nicely instead.
1564 if (!is_store && value->is_Con() && index->is_Con()) {
1565 return false;
1566 }
1567
1568 value = must_be_not_null(value, true);
1569
1570 Node* adr = array_element_address(value, index, T_CHAR);
1571 if (adr->is_top()) {
1572 return false;
1573 }
1574 if (is_store) {
1575 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1576 } else {
1577 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1578 set_result(ch);
1579 }
1580 return true;
1581 }
1582
1583 //--------------------------round_double_node--------------------------------
1584 // Round a double node if necessary.
1585 Node* LibraryCallKit::round_double_node(Node* n) {
1586 if (Matcher::strict_fp_requires_explicit_rounding) {
1587 #ifdef IA32
1588 if (UseSSE < 2) {
1589 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1590 }
1591 #else
1592 Unimplemented();
1593 #endif // IA32
1594 }
1595 return n;
1596 }
1597
1598 //------------------------------inline_math-----------------------------------
1599 // public static double Math.abs(double)
1600 // public static double Math.sqrt(double)
1601 // public static double Math.log(double)
1602 // public static double Math.log10(double)
1603 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1604 Node* arg = round_double_node(argument(0));
1605 Node* n = NULL;
1606 switch (id) {
1607 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1608 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1609 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1610 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1611 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1612 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1613 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1614 default: fatal_unexpected_iid(id); break;
1615 }
1616 set_result(_gvn.transform(n));
1617 return true;
1618 }
1619
1620 //------------------------------inline_math-----------------------------------
1621 // public static float Math.abs(float)
1622 // public static int Math.abs(int)
1623 // public static long Math.abs(long)
1624 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1625 Node* arg = argument(0);
1626 Node* n = NULL;
1627 switch (id) {
1628 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1629 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1630 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1631 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1632 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1633 default: fatal_unexpected_iid(id); break;
1634 }
1635 set_result(_gvn.transform(n));
1636 return true;
1637 }
1638
1639 //------------------------------runtime_math-----------------------------
1640 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1641 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1642 "must be (DD)D or (D)D type");
1643
1644 // Inputs
1645 Node* a = round_double_node(argument(0));
1646 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1647
1648 const TypePtr* no_memory_effects = NULL;
1649 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1650 no_memory_effects,
1651 a, top(), b, b ? top() : NULL);
1652 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1653 #ifdef ASSERT
1654 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1655 assert(value_top == top(), "second value must be top");
1656 #endif
1657
1658 set_result(value);
1659 return true;
1660 }
1661
1662 //------------------------------inline_math_pow-----------------------------
1663 bool LibraryCallKit::inline_math_pow() {
1664 Node* exp = round_double_node(argument(2));
1665 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1666 if (d != NULL) {
1667 if (d->getd() == 2.0) {
1668 // Special case: pow(x, 2.0) => x * x
1669 Node* base = round_double_node(argument(0));
1670 set_result(_gvn.transform(new MulDNode(base, base)));
1671 return true;
1672 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1673 // Special case: pow(x, 0.5) => sqrt(x)
1674 Node* base = round_double_node(argument(0));
1675 Node* zero = _gvn.zerocon(T_DOUBLE);
1676
1677 RegionNode* region = new RegionNode(3);
1678 Node* phi = new PhiNode(region, Type::DOUBLE);
1679
1680 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1681 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1682 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1683 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1684 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1685
1686 Node* if_pow = generate_slow_guard(test, NULL);
1687 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1688 phi->init_req(1, value_sqrt);
1689 region->init_req(1, control());
1690
1691 if (if_pow != NULL) {
1692 set_control(if_pow);
1693 address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1694 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1695 const TypePtr* no_memory_effects = NULL;
1696 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1697 no_memory_effects, base, top(), exp, top());
1698 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1699 #ifdef ASSERT
1700 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1701 assert(value_top == top(), "second value must be top");
1702 #endif
1703 phi->init_req(2, value_pow);
1704 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1705 }
1706
1707 C->set_has_split_ifs(true); // Has chance for split-if optimization
1708 set_control(_gvn.transform(region));
1709 record_for_igvn(region);
1710 set_result(_gvn.transform(phi));
1711
1712 return true;
1713 }
1714 }
1715
1716 return StubRoutines::dpow() != NULL ?
1717 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1718 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1719 }
1720
1721 //------------------------------inline_math_native-----------------------------
1722 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1723 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1724 switch (id) {
1725 // These intrinsics are not properly supported on all hardware
1726 case vmIntrinsics::_dsin:
1727 return StubRoutines::dsin() != NULL ?
1728 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1729 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1730 case vmIntrinsics::_dcos:
1731 return StubRoutines::dcos() != NULL ?
1732 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1733 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1734 case vmIntrinsics::_dtan:
1735 return StubRoutines::dtan() != NULL ?
1736 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1737 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1738 case vmIntrinsics::_dlog:
1739 return StubRoutines::dlog() != NULL ?
1740 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1741 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1742 case vmIntrinsics::_dlog10:
1743 return StubRoutines::dlog10() != NULL ?
1744 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1745 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1746
1747 // These intrinsics are supported on all hardware
1748 case vmIntrinsics::_ceil:
1749 case vmIntrinsics::_floor:
1750 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1751 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1752 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1753 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1754 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1755 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1756
1757 case vmIntrinsics::_dexp:
1758 return StubRoutines::dexp() != NULL ?
1759 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1760 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1761 #undef FN_PTR
1762
1763 case vmIntrinsics::_dpow: return inline_math_pow();
1764 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1765 case vmIntrinsics::_fcopySign: return inline_math(id);
1766 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1767 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1768
1769 // These intrinsics are not yet correctly implemented
1770 case vmIntrinsics::_datan2:
1771 return false;
1772
1773 default:
1774 fatal_unexpected_iid(id);
1775 return false;
1776 }
1777 }
1778
1779 static bool is_simple_name(Node* n) {
1780 return (n->req() == 1 // constant
1781 || (n->is_Type() && n->as_Type()->type()->singleton())
1782 || n->is_Proj() // parameter or return value
1783 || n->is_Phi() // local of some sort
1784 );
1785 }
1786
1787 //----------------------------inline_notify-----------------------------------*
1788 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1789 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1790 address func;
1791 if (id == vmIntrinsics::_notify) {
1792 func = OptoRuntime::monitor_notify_Java();
1793 } else {
1794 func = OptoRuntime::monitor_notifyAll_Java();
1795 }
1796 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1797 make_slow_call_ex(call, env()->Throwable_klass(), false);
1798 return true;
1799 }
1800
1801
1802 //----------------------------inline_min_max-----------------------------------
1803 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1804 set_result(generate_min_max(id, argument(0), argument(1)));
1805 return true;
1806 }
1807
1808 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1809 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1810 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1811 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1812 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1813
1814 {
1815 PreserveJVMState pjvms(this);
1816 PreserveReexecuteState preexecs(this);
1817 jvms()->set_should_reexecute(true);
1818
1819 set_control(slow_path);
1820 set_i_o(i_o());
1821
1822 uncommon_trap(Deoptimization::Reason_intrinsic,
1823 Deoptimization::Action_none);
1824 }
1825
1826 set_control(fast_path);
1827 set_result(math);
1828 }
1829
1830 template <typename OverflowOp>
1831 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1832 typedef typename OverflowOp::MathOp MathOp;
1833
1834 MathOp* mathOp = new MathOp(arg1, arg2);
1835 Node* operation = _gvn.transform( mathOp );
1836 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1837 inline_math_mathExact(operation, ofcheck);
1838 return true;
1839 }
1840
1841 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1842 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1843 }
1844
1845 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1846 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1847 }
1848
1849 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1850 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1851 }
1852
1853 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1854 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1855 }
1856
1857 bool LibraryCallKit::inline_math_negateExactI() {
1858 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1859 }
1860
1861 bool LibraryCallKit::inline_math_negateExactL() {
1862 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1863 }
1864
1865 bool LibraryCallKit::inline_math_multiplyExactI() {
1866 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1867 }
1868
1869 bool LibraryCallKit::inline_math_multiplyExactL() {
1870 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1871 }
1872
1873 bool LibraryCallKit::inline_math_multiplyHigh() {
1874 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1875 return true;
1876 }
1877
1878 Node*
1879 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1880 // These are the candidate return value:
1881 Node* xvalue = x0;
1882 Node* yvalue = y0;
1883
1884 if (xvalue == yvalue) {
1885 return xvalue;
1886 }
1887
1888 bool want_max = (id == vmIntrinsics::_max);
1889
1890 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1891 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1892 if (txvalue == NULL || tyvalue == NULL) return top();
1893 // This is not really necessary, but it is consistent with a
1894 // hypothetical MaxINode::Value method:
1895 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1896
1897 // %%% This folding logic should (ideally) be in a different place.
1898 // Some should be inside IfNode, and there to be a more reliable
1899 // transformation of ?: style patterns into cmoves. We also want
1900 // more powerful optimizations around cmove and min/max.
1901
1902 // Try to find a dominating comparison of these guys.
1903 // It can simplify the index computation for Arrays.copyOf
1904 // and similar uses of System.arraycopy.
1905 // First, compute the normalized version of CmpI(x, y).
1906 int cmp_op = Op_CmpI;
1907 Node* xkey = xvalue;
1908 Node* ykey = yvalue;
1909 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1910 if (ideal_cmpxy->is_Cmp()) {
1911 // E.g., if we have CmpI(length - offset, count),
1912 // it might idealize to CmpI(length, count + offset)
1913 cmp_op = ideal_cmpxy->Opcode();
1914 xkey = ideal_cmpxy->in(1);
1915 ykey = ideal_cmpxy->in(2);
1916 }
1917
1918 // Start by locating any relevant comparisons.
1919 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1920 Node* cmpxy = NULL;
1921 Node* cmpyx = NULL;
1922 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1923 Node* cmp = start_from->fast_out(k);
1924 if (cmp->outcnt() > 0 && // must have prior uses
1925 cmp->in(0) == NULL && // must be context-independent
1926 cmp->Opcode() == cmp_op) { // right kind of compare
1927 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1928 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1929 }
1930 }
1931
1932 const int NCMPS = 2;
1933 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1934 int cmpn;
1935 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1936 if (cmps[cmpn] != NULL) break; // find a result
1937 }
1938 if (cmpn < NCMPS) {
1939 // Look for a dominating test that tells us the min and max.
1940 int depth = 0; // Limit search depth for speed
1941 Node* dom = control();
1942 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1943 if (++depth >= 100) break;
1944 Node* ifproj = dom;
1945 if (!ifproj->is_Proj()) continue;
1946 Node* iff = ifproj->in(0);
1947 if (!iff->is_If()) continue;
1948 Node* bol = iff->in(1);
1949 if (!bol->is_Bool()) continue;
1950 Node* cmp = bol->in(1);
1951 if (cmp == NULL) continue;
1952 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1953 if (cmps[cmpn] == cmp) break;
1954 if (cmpn == NCMPS) continue;
1955 BoolTest::mask btest = bol->as_Bool()->_test._test;
1956 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1957 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1958 // At this point, we know that 'x btest y' is true.
1959 switch (btest) {
1960 case BoolTest::eq:
1961 // They are proven equal, so we can collapse the min/max.
1962 // Either value is the answer. Choose the simpler.
1963 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1964 return yvalue;
1965 return xvalue;
1966 case BoolTest::lt: // x < y
1967 case BoolTest::le: // x <= y
1968 return (want_max ? yvalue : xvalue);
1969 case BoolTest::gt: // x > y
1970 case BoolTest::ge: // x >= y
1971 return (want_max ? xvalue : yvalue);
1972 default:
1973 break;
1974 }
1975 }
1976 }
1977
1978 // We failed to find a dominating test.
1979 // Let's pick a test that might GVN with prior tests.
1980 Node* best_bol = NULL;
1981 BoolTest::mask best_btest = BoolTest::illegal;
1982 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1983 Node* cmp = cmps[cmpn];
1984 if (cmp == NULL) continue;
1985 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1986 Node* bol = cmp->fast_out(j);
1987 if (!bol->is_Bool()) continue;
1988 BoolTest::mask btest = bol->as_Bool()->_test._test;
1989 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
1990 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1991 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1992 best_bol = bol->as_Bool();
1993 best_btest = btest;
1994 }
1995 }
1996 }
1997
1998 Node* answer_if_true = NULL;
1999 Node* answer_if_false = NULL;
2000 switch (best_btest) {
2001 default:
2002 if (cmpxy == NULL)
2003 cmpxy = ideal_cmpxy;
2004 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2005 // and fall through:
2006 case BoolTest::lt: // x < y
2007 case BoolTest::le: // x <= y
2008 answer_if_true = (want_max ? yvalue : xvalue);
2009 answer_if_false = (want_max ? xvalue : yvalue);
2010 break;
2011 case BoolTest::gt: // x > y
2012 case BoolTest::ge: // x >= y
2013 answer_if_true = (want_max ? xvalue : yvalue);
2014 answer_if_false = (want_max ? yvalue : xvalue);
2015 break;
2016 }
2017
2018 jint hi, lo;
2019 if (want_max) {
2020 // We can sharpen the minimum.
2021 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2022 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2023 } else {
2024 // We can sharpen the maximum.
2025 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2026 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2027 }
2028
2029 // Use a flow-free graph structure, to avoid creating excess control edges
2030 // which could hinder other optimizations.
2031 // Since Math.min/max is often used with arraycopy, we want
2032 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2033 Node* cmov = CMoveNode::make(NULL, best_bol,
2034 answer_if_false, answer_if_true,
2035 TypeInt::make(lo, hi, widen));
2036
2037 return _gvn.transform(cmov);
2038
2039 /*
2040 // This is not as desirable as it may seem, since Min and Max
2041 // nodes do not have a full set of optimizations.
2042 // And they would interfere, anyway, with 'if' optimizations
2043 // and with CMoveI canonical forms.
2044 switch (id) {
2045 case vmIntrinsics::_min:
2046 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2047 case vmIntrinsics::_max:
2048 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2049 default:
2050 ShouldNotReachHere();
2051 }
2052 */
2053 }
2054
2055 inline int
2056 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2057 const TypePtr* base_type = TypePtr::NULL_PTR;
2058 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2059 if (base_type == NULL) {
2060 // Unknown type.
2061 return Type::AnyPtr;
2062 } else if (base_type == TypePtr::NULL_PTR) {
2063 // Since this is a NULL+long form, we have to switch to a rawptr.
2064 base = _gvn.transform(new CastX2PNode(offset));
2065 offset = MakeConX(0);
2066 return Type::RawPtr;
2067 } else if (base_type->base() == Type::RawPtr) {
2068 return Type::RawPtr;
2069 } else if (base_type->isa_oopptr()) {
2070 // Base is never null => always a heap address.
2071 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2072 return Type::OopPtr;
2073 }
2074 // Offset is small => always a heap address.
2075 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2076 if (offset_type != NULL &&
2077 base_type->offset() == 0 && // (should always be?)
2078 offset_type->_lo >= 0 &&
2079 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2080 return Type::OopPtr;
2081 } else if (type == T_OBJECT) {
2082 // off heap access to an oop doesn't make any sense. Has to be on
2083 // heap.
2084 return Type::OopPtr;
2085 }
2086 // Otherwise, it might either be oop+off or NULL+addr.
2087 return Type::AnyPtr;
2088 } else {
2089 // No information:
2090 return Type::AnyPtr;
2091 }
2092 }
2093
2094 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2095 Node* uncasted_base = base;
2096 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2097 if (kind == Type::RawPtr) {
2098 return basic_plus_adr(top(), uncasted_base, offset);
2099 } else if (kind == Type::AnyPtr) {
2100 assert(base == uncasted_base, "unexpected base change");
2101 if (can_cast) {
2102 if (!_gvn.type(base)->speculative_maybe_null() &&
2103 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2104 // According to profiling, this access is always on
2105 // heap. Casting the base to not null and thus avoiding membars
2106 // around the access should allow better optimizations
2107 Node* null_ctl = top();
2108 base = null_check_oop(base, &null_ctl, true, true, true);
2109 assert(null_ctl->is_top(), "no null control here");
2110 return basic_plus_adr(base, offset);
2111 } else if (_gvn.type(base)->speculative_always_null() &&
2112 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2113 // According to profiling, this access is always off
2114 // heap.
2115 base = null_assert(base);
2116 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2117 offset = MakeConX(0);
2118 return basic_plus_adr(top(), raw_base, offset);
2119 }
2120 }
2121 // We don't know if it's an on heap or off heap access. Fall back
2122 // to raw memory access.
2123 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2124 return basic_plus_adr(top(), raw, offset);
2125 } else {
2126 assert(base == uncasted_base, "unexpected base change");
2127 // We know it's an on heap access so base can't be null
2128 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2129 base = must_be_not_null(base, true);
2130 }
2131 return basic_plus_adr(base, offset);
2132 }
2133 }
2134
2135 //--------------------------inline_number_methods-----------------------------
2136 // inline int Integer.numberOfLeadingZeros(int)
2137 // inline int Long.numberOfLeadingZeros(long)
2138 //
2139 // inline int Integer.numberOfTrailingZeros(int)
2140 // inline int Long.numberOfTrailingZeros(long)
2141 //
2142 // inline int Integer.bitCount(int)
2143 // inline int Long.bitCount(long)
2144 //
2145 // inline char Character.reverseBytes(char)
2146 // inline short Short.reverseBytes(short)
2147 // inline int Integer.reverseBytes(int)
2148 // inline long Long.reverseBytes(long)
2149 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2150 Node* arg = argument(0);
2151 Node* n = NULL;
2152 switch (id) {
2153 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2154 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2155 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2156 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2157 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2158 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2159 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2160 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2161 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2162 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2163 default: fatal_unexpected_iid(id); break;
2164 }
2165 set_result(_gvn.transform(n));
2166 return true;
2167 }
2168
2169 //----------------------------inline_unsafe_access----------------------------
2170
2171 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2172 // Attempt to infer a sharper value type from the offset and base type.
2173 ciKlass* sharpened_klass = NULL;
2174 bool null_free = false;
2175
2176 // See if it is an instance field, with an object type.
2177 if (alias_type->field() != NULL) {
2178 if (alias_type->field()->type()->is_klass()) {
2179 sharpened_klass = alias_type->field()->type()->as_klass();
2180 null_free = alias_type->field()->is_null_free();
2181 }
2182 }
2183
2184 // See if it is a narrow oop array.
2185 if (adr_type->isa_aryptr()) {
2186 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2187 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2188 null_free = adr_type->is_aryptr()->is_null_free();
2189 if (elem_type != NULL) {
2190 sharpened_klass = elem_type->klass();
2191 }
2192 }
2193 }
2194
2195 // The sharpened class might be unloaded if there is no class loader
2196 // contraint in place.
2197 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2198 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2199 if (null_free) {
2200 tjp = tjp->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2201 }
2202
2203 #ifndef PRODUCT
2204 if (C->print_intrinsics() || C->print_inlining()) {
2205 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2206 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2207 }
2208 #endif
2209 // Sharpen the value type.
2210 return tjp;
2211 }
2212 return NULL;
2213 }
2214
2215 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2216 switch (kind) {
2217 case Relaxed:
2218 return MO_UNORDERED;
2219 case Opaque:
2220 return MO_RELAXED;
2221 case Acquire:
2222 return MO_ACQUIRE;
2223 case Release:
2224 return MO_RELEASE;
2225 case Volatile:
2226 return MO_SEQ_CST;
2227 default:
2228 ShouldNotReachHere();
2229 return 0;
2230 }
2231 }
2232
2233 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2234 if (callee()->is_static()) return false; // caller must have the capability!
2235 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2236 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2237 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2238 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2239
2240 if (is_reference_type(type)) {
2241 decorators |= ON_UNKNOWN_OOP_REF;
2242 }
2243
2244 if (unaligned) {
2245 decorators |= C2_UNALIGNED;
2246 }
2247
2248 #ifndef PRODUCT
2249 {
2250 ResourceMark rm;
2251 // Check the signatures.
2252 ciSignature* sig = callee()->signature();
2253 #ifdef ASSERT
2254 if (!is_store) {
2255 // Object getReference(Object base, int/long offset), etc.
2256 BasicType rtype = sig->return_type()->basic_type();
2257 assert(rtype == type || (rtype == T_OBJECT && type == T_INLINE_TYPE), "getter must return the expected value");
2258 assert(sig->count() == 2 || (type == T_INLINE_TYPE && sig->count() == 3), "oop getter has 2 or 3 arguments");
2259 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2260 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2261 } else {
2262 // void putReference(Object base, int/long offset, Object x), etc.
2263 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2264 assert(sig->count() == 3 || (type == T_INLINE_TYPE && sig->count() == 4), "oop putter has 3 arguments");
2265 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2266 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2267 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2268 assert(vtype == type || (type == T_INLINE_TYPE && vtype == T_OBJECT), "putter must accept the expected value");
2269 }
2270 #endif // ASSERT
2271 }
2272 #endif //PRODUCT
2273
2274 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2275
2276 Node* receiver = argument(0); // type: oop
2277
2278 // Build address expression.
2279 Node* heap_base_oop = top();
2280
2281 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2282 Node* base = argument(1); // type: oop
2283 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2284 Node* offset = argument(2); // type: long
2285 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2286 // to be plain byte offsets, which are also the same as those accepted
2287 // by oopDesc::field_addr.
2288 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2289 "fieldOffset must be byte-scaled");
2290
2291 ciInlineKlass* inline_klass = NULL;
2292 if (type == T_INLINE_TYPE) {
2293 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2294 if (cls == NULL || cls->const_oop() == NULL) {
2295 return false;
2296 }
2297 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2298 if (!mirror_type->is_inlinetype()) {
2299 return false;
2300 }
2301 inline_klass = mirror_type->as_inline_klass();
2302 }
2303
2304 if (base->is_InlineTypeBase()) {
2305 InlineTypeBaseNode* vt = base->as_InlineTypeBase();
2306 if (is_store) {
2307 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->isa_inlinetype() || !_gvn.type(vt)->is_inlinetype()->larval()) {
2308 return false;
2309 }
2310 base = vt->get_oop();
2311 } else {
2312 if (offset->is_Con()) {
2313 long off = find_long_con(offset, 0);
2314 ciInlineKlass* vk = vt->type()->inline_klass();
2315 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2316 return false;
2317 }
2318
2319 ciField* field = vk->get_non_flattened_field_by_offset(off);
2320 if (field != NULL) {
2321 BasicType bt = field->layout_type();
2322 if (bt == T_ARRAY || bt == T_NARROWOOP || (bt == T_INLINE_TYPE && !field->is_flattened())) {
2323 bt = T_OBJECT;
2324 }
2325 if (bt == type && (bt != T_INLINE_TYPE || field->type() == inline_klass)) {
2326 set_result(vt->field_value_by_offset(off, false));
2327 return true;
2328 }
2329 }
2330 }
2331 if (vt->is_InlineType()) {
2332 // Re-execute the unsafe access if allocation triggers deoptimization.
2333 PreserveReexecuteState preexecs(this);
2334 jvms()->set_should_reexecute(true);
2335 vt = vt->buffer(this);
2336 }
2337 base = vt->get_oop();
2338 }
2339 }
2340
2341 // 32-bit machines ignore the high half!
2342 offset = ConvL2X(offset);
2343
2344 // Save state and restore on bailout
2345 uint old_sp = sp();
2346 SafePointNode* old_map = clone_map();
2347
2348 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2349
2350 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2351 if (type != T_OBJECT && (inline_klass == NULL || !inline_klass->has_object_fields())) {
2352 decorators |= IN_NATIVE; // off-heap primitive access
2353 } else {
2354 set_map(old_map);
2355 set_sp(old_sp);
2356 return false; // off-heap oop accesses are not supported
2357 }
2358 } else {
2359 heap_base_oop = base; // on-heap or mixed access
2360 }
2361
2362 // Can base be NULL? Otherwise, always on-heap access.
2363 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2364
2365 if (!can_access_non_heap) {
2366 decorators |= IN_HEAP;
2367 }
2368
2369 Node* val = is_store ? argument(4 + (type == T_INLINE_TYPE ? 1 : 0)) : NULL;
2370
2371 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2372 if (adr_type == TypePtr::NULL_PTR) {
2373 set_map(old_map);
2374 set_sp(old_sp);
2375 return false; // off-heap access with zero address
2376 }
2377
2378 // Try to categorize the address.
2379 Compile::AliasType* alias_type = C->alias_type(adr_type);
2380 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2381
2382 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2383 alias_type->adr_type() == TypeAryPtr::RANGE) {
2384 set_map(old_map);
2385 set_sp(old_sp);
2386 return false; // not supported
2387 }
2388
2389 bool mismatched = false;
2390 BasicType bt = T_ILLEGAL;
2391 ciField* field = NULL;
2392 if (adr_type->isa_instptr()) {
2393 const TypeInstPtr* instptr = adr_type->is_instptr();
2394 ciInstanceKlass* k = instptr->klass()->as_instance_klass();
2395 int off = instptr->offset();
2396 if (instptr->const_oop() != NULL &&
2397 instptr->klass() == ciEnv::current()->Class_klass() &&
2398 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) {
2399 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2400 field = k->get_field_by_offset(off, true);
2401 } else {
2402 field = k->get_non_flattened_field_by_offset(off);
2403 }
2404 if (field != NULL) {
2405 bt = field->layout_type();
2406 }
2407 assert(bt == alias_type->basic_type() || bt == T_INLINE_TYPE, "should match");
2408 if (field != NULL && bt == T_INLINE_TYPE && !field->is_flattened()) {
2409 bt = T_OBJECT;
2410 }
2411 } else {
2412 bt = alias_type->basic_type();
2413 }
2414
2415 if (bt != T_ILLEGAL) {
2416 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2417 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2418 // Alias type doesn't differentiate between byte[] and boolean[]).
2419 // Use address type to get the element type.
2420 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2421 }
2422 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2423 // accessing an array field with getReference is not a mismatch
2424 bt = T_OBJECT;
2425 }
2426 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2427 // Don't intrinsify mismatched object accesses
2428 set_map(old_map);
2429 set_sp(old_sp);
2430 return false;
2431 }
2432 mismatched = (bt != type);
2433 } else if (alias_type->adr_type()->isa_oopptr()) {
2434 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2435 }
2436
2437 if (type == T_INLINE_TYPE) {
2438 if (adr_type->isa_instptr()) {
2439 if (field == NULL || field->type() != inline_klass) {
2440 mismatched = true;
2441 }
2442 } else if (adr_type->isa_aryptr()) {
2443 const Type* elem = adr_type->is_aryptr()->elem();
2444 if (!elem->isa_inlinetype()) {
2445 mismatched = true;
2446 } else if (elem->inline_klass() != inline_klass) {
2447 mismatched = true;
2448 }
2449 } else {
2450 mismatched = true;
2451 }
2452 if (is_store) {
2453 const Type* val_t = _gvn.type(val);
2454 if (!val_t->isa_inlinetype() || val_t->inline_klass() != inline_klass) {
2455 set_map(old_map);
2456 set_sp(old_sp);
2457 return false;
2458 }
2459 }
2460 }
2461
2462 old_map->destruct(&_gvn);
2463 assert(!mismatched || type == T_INLINE_TYPE || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2464
2465 if (mismatched) {
2466 decorators |= C2_MISMATCHED;
2467 }
2468
2469 // First guess at the value type.
2470 const Type *value_type = Type::get_const_basic_type(type);
2471
2472 // Figure out the memory ordering.
2473 decorators |= mo_decorator_for_access_kind(kind);
2474
2475 if (!is_store) {
2476 if (type == T_OBJECT) {
2477 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2478 if (tjp != NULL) {
2479 value_type = tjp;
2480 }
2481 } else if (type == T_INLINE_TYPE) {
2482 value_type = NULL;
2483 }
2484 }
2485
2486 receiver = null_check(receiver);
2487 if (stopped()) {
2488 return true;
2489 }
2490 // Heap pointers get a null-check from the interpreter,
2491 // as a courtesy. However, this is not guaranteed by Unsafe,
2492 // and it is not possible to fully distinguish unintended nulls
2493 // from intended ones in this API.
2494
2495 if (!is_store) {
2496 Node* p = NULL;
2497 // Try to constant fold a load from a constant field
2498
2499 if (heap_base_oop != top() && field != NULL && field->is_constant() && !field->is_flattened() && !mismatched) {
2500 // final or stable field
2501 p = make_constant_from_field(field, heap_base_oop);
2502 }
2503
2504 if (p == NULL) { // Could not constant fold the load
2505 if (type == T_INLINE_TYPE) {
2506 if (adr_type->isa_instptr() && !mismatched) {
2507 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2508 int offset = adr_type->is_instptr()->offset();
2509 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, base, holder, offset, decorators);
2510 } else {
2511 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, adr, NULL, 0, decorators);
2512 }
2513 } else {
2514 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2515 const TypeOopPtr* ptr = value_type->make_oopptr();
2516 if (ptr != NULL && ptr->is_inlinetypeptr()) {
2517 // Load a non-flattened inline type from memory
2518 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2519 }
2520 }
2521 // Normalize the value returned by getBoolean in the following cases
2522 if (type == T_BOOLEAN &&
2523 (mismatched ||
2524 heap_base_oop == top() || // - heap_base_oop is NULL or
2525 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2526 // and the unsafe access is made to large offset
2527 // (i.e., larger than the maximum offset necessary for any
2528 // field access)
2529 ) {
2530 IdealKit ideal = IdealKit(this);
2531 #define __ ideal.
2532 IdealVariable normalized_result(ideal);
2533 __ declarations_done();
2534 __ set(normalized_result, p);
2535 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2536 __ set(normalized_result, ideal.ConI(1));
2537 ideal.end_if();
2538 final_sync(ideal);
2539 p = __ value(normalized_result);
2540 #undef __
2541 }
2542 }
2543 if (type == T_ADDRESS) {
2544 p = gvn().transform(new CastP2XNode(NULL, p));
2545 p = ConvX2UL(p);
2546 }
2547 // The load node has the control of the preceding MemBarCPUOrder. All
2548 // following nodes will have the control of the MemBarCPUOrder inserted at
2549 // the end of this method. So, pushing the load onto the stack at a later
2550 // point is fine.
2551 set_result(p);
2552 } else {
2553 if (bt == T_ADDRESS) {
2554 // Repackage the long as a pointer.
2555 val = ConvL2X(val);
2556 val = gvn().transform(new CastX2PNode(val));
2557 }
2558 if (type == T_INLINE_TYPE) {
2559 if (adr_type->isa_instptr() && !mismatched) {
2560 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2561 int offset = adr_type->is_instptr()->offset();
2562 val->as_InlineTypeBase()->store_flattened(this, base, base, holder, offset, decorators);
2563 } else {
2564 val->as_InlineTypeBase()->store_flattened(this, base, adr, NULL, 0, decorators);
2565 }
2566 } else {
2567 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2568 }
2569 }
2570
2571 if (argument(1)->is_InlineType() && is_store) {
2572 Node* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(base)->inline_klass());
2573 value = value->as_InlineType()->make_larval(this, false);
2574 replace_in_map(argument(1), value);
2575 }
2576
2577 return true;
2578 }
2579
2580 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2581 Node* receiver = argument(0);
2582 Node* value = argument(1);
2583 if (!value->is_InlineType()) {
2584 return false;
2585 }
2586
2587 receiver = null_check(receiver);
2588 if (stopped()) {
2589 return true;
2590 }
2591
2592 set_result(value->as_InlineType()->make_larval(this, true));
2593 return true;
2594 }
2595
2596 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2597 Node* receiver = argument(0);
2598 Node* buffer = argument(1);
2599 if (!buffer->is_InlineType()) {
2600 return false;
2601 }
2602 InlineTypeNode* vt = buffer->as_InlineType();
2603 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_inlinetype()->larval()) {
2604 return false;
2605 }
2606
2607 receiver = null_check(receiver);
2608 if (stopped()) {
2609 return true;
2610 }
2611
2612 set_result(vt->finish_larval(this));
2613 return true;
2614 }
2615
2616 //----------------------------inline_unsafe_load_store----------------------------
2617 // This method serves a couple of different customers (depending on LoadStoreKind):
2618 //
2619 // LS_cmp_swap:
2620 //
2621 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2622 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2623 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2624 //
2625 // LS_cmp_swap_weak:
2626 //
2627 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2628 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2629 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2630 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2631 //
2632 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2633 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2634 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2635 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2636 //
2637 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2638 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2639 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2640 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2641 //
2642 // LS_cmp_exchange:
2643 //
2644 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2645 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2646 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2647 //
2648 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2649 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2650 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2651 //
2652 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2653 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2654 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2655 //
2656 // LS_get_add:
2657 //
2658 // int getAndAddInt( Object o, long offset, int delta)
2659 // long getAndAddLong(Object o, long offset, long delta)
2660 //
2661 // LS_get_set:
2662 //
2663 // int getAndSet(Object o, long offset, int newValue)
2664 // long getAndSet(Object o, long offset, long newValue)
2665 // Object getAndSet(Object o, long offset, Object newValue)
2666 //
2667 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2668 // This basic scheme here is the same as inline_unsafe_access, but
2669 // differs in enough details that combining them would make the code
2670 // overly confusing. (This is a true fact! I originally combined
2671 // them, but even I was confused by it!) As much code/comments as
2672 // possible are retained from inline_unsafe_access though to make
2673 // the correspondences clearer. - dl
2674
2675 if (callee()->is_static()) return false; // caller must have the capability!
2676
2677 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2678 decorators |= mo_decorator_for_access_kind(access_kind);
2679
2680 #ifndef PRODUCT
2681 BasicType rtype;
2682 {
2683 ResourceMark rm;
2684 // Check the signatures.
2685 ciSignature* sig = callee()->signature();
2686 rtype = sig->return_type()->basic_type();
2687 switch(kind) {
2688 case LS_get_add:
2689 case LS_get_set: {
2690 // Check the signatures.
2691 #ifdef ASSERT
2692 assert(rtype == type, "get and set must return the expected type");
2693 assert(sig->count() == 3, "get and set has 3 arguments");
2694 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2695 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2696 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2697 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2698 #endif // ASSERT
2699 break;
2700 }
2701 case LS_cmp_swap:
2702 case LS_cmp_swap_weak: {
2703 // Check the signatures.
2704 #ifdef ASSERT
2705 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2706 assert(sig->count() == 4, "CAS has 4 arguments");
2707 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2708 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2709 #endif // ASSERT
2710 break;
2711 }
2712 case LS_cmp_exchange: {
2713 // Check the signatures.
2714 #ifdef ASSERT
2715 assert(rtype == type, "CAS must return the expected type");
2716 assert(sig->count() == 4, "CAS has 4 arguments");
2717 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2718 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2719 #endif // ASSERT
2720 break;
2721 }
2722 default:
2723 ShouldNotReachHere();
2724 }
2725 }
2726 #endif //PRODUCT
2727
2728 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2729
2730 // Get arguments:
2731 Node* receiver = NULL;
2732 Node* base = NULL;
2733 Node* offset = NULL;
2734 Node* oldval = NULL;
2735 Node* newval = NULL;
2736 switch(kind) {
2737 case LS_cmp_swap:
2738 case LS_cmp_swap_weak:
2739 case LS_cmp_exchange: {
2740 const bool two_slot_type = type2size[type] == 2;
2741 receiver = argument(0); // type: oop
2742 base = argument(1); // type: oop
2743 offset = argument(2); // type: long
2744 oldval = argument(4); // type: oop, int, or long
2745 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2746 break;
2747 }
2748 case LS_get_add:
2749 case LS_get_set: {
2750 receiver = argument(0); // type: oop
2751 base = argument(1); // type: oop
2752 offset = argument(2); // type: long
2753 oldval = NULL;
2754 newval = argument(4); // type: oop, int, or long
2755 break;
2756 }
2757 default:
2758 ShouldNotReachHere();
2759 }
2760
2761 // Build field offset expression.
2762 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2763 // to be plain byte offsets, which are also the same as those accepted
2764 // by oopDesc::field_addr.
2765 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2766 // 32-bit machines ignore the high half of long offsets
2767 offset = ConvL2X(offset);
2768 // Save state and restore on bailout
2769 uint old_sp = sp();
2770 SafePointNode* old_map = clone_map();
2771 Node* adr = make_unsafe_address(base, offset,type, false);
2772 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2773
2774 Compile::AliasType* alias_type = C->alias_type(adr_type);
2775 BasicType bt = alias_type->basic_type();
2776 if (bt != T_ILLEGAL &&
2777 (is_reference_type(bt) != (type == T_OBJECT))) {
2778 // Don't intrinsify mismatched object accesses.
2779 set_map(old_map);
2780 set_sp(old_sp);
2781 return false;
2782 }
2783
2784 old_map->destruct(&_gvn);
2785
2786 // For CAS, unlike inline_unsafe_access, there seems no point in
2787 // trying to refine types. Just use the coarse types here.
2788 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2789 const Type *value_type = Type::get_const_basic_type(type);
2790
2791 switch (kind) {
2792 case LS_get_set:
2793 case LS_cmp_exchange: {
2794 if (type == T_OBJECT) {
2795 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2796 if (tjp != NULL) {
2797 value_type = tjp;
2798 }
2799 }
2800 break;
2801 }
2802 case LS_cmp_swap:
2803 case LS_cmp_swap_weak:
2804 case LS_get_add:
2805 break;
2806 default:
2807 ShouldNotReachHere();
2808 }
2809
2810 // Null check receiver.
2811 receiver = null_check(receiver);
2812 if (stopped()) {
2813 return true;
2814 }
2815
2816 int alias_idx = C->get_alias_index(adr_type);
2817
2818 if (is_reference_type(type)) {
2819 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2820
2821 if (oldval != NULL && oldval->is_InlineType()) {
2822 // Re-execute the unsafe access if allocation triggers deoptimization.
2823 PreserveReexecuteState preexecs(this);
2824 jvms()->set_should_reexecute(true);
2825 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2826 }
2827 if (newval != NULL && newval->is_InlineType()) {
2828 // Re-execute the unsafe access if allocation triggers deoptimization.
2829 PreserveReexecuteState preexecs(this);
2830 jvms()->set_should_reexecute(true);
2831 newval = newval->as_InlineType()->buffer(this)->get_oop();
2832 }
2833
2834 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2835 // could be delayed during Parse (for example, in adjust_map_after_if()).
2836 // Execute transformation here to avoid barrier generation in such case.
2837 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2838 newval = _gvn.makecon(TypePtr::NULL_PTR);
2839
2840 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2841 // Refine the value to a null constant, when it is known to be null
2842 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2843 }
2844 }
2845
2846 Node* result = NULL;
2847 switch (kind) {
2848 case LS_cmp_exchange: {
2849 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2850 oldval, newval, value_type, type, decorators);
2851 break;
2852 }
2853 case LS_cmp_swap_weak:
2854 decorators |= C2_WEAK_CMPXCHG;
2855 case LS_cmp_swap: {
2856 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2857 oldval, newval, value_type, type, decorators);
2858 break;
2859 }
2860 case LS_get_set: {
2861 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2862 newval, value_type, type, decorators);
2863 break;
2864 }
2865 case LS_get_add: {
2866 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2867 newval, value_type, type, decorators);
2868 break;
2869 }
2870 default:
2871 ShouldNotReachHere();
2872 }
2873
2874 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2875 set_result(result);
2876 return true;
2877 }
2878
2879 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2880 // Regardless of form, don't allow previous ld/st to move down,
2881 // then issue acquire, release, or volatile mem_bar.
2882 insert_mem_bar(Op_MemBarCPUOrder);
2883 switch(id) {
2884 case vmIntrinsics::_loadFence:
2885 insert_mem_bar(Op_LoadFence);
2886 return true;
2887 case vmIntrinsics::_storeFence:
2888 insert_mem_bar(Op_StoreFence);
2889 return true;
2890 case vmIntrinsics::_fullFence:
2891 insert_mem_bar(Op_MemBarVolatile);
2892 return true;
2893 default:
2894 fatal_unexpected_iid(id);
2895 return false;
2896 }
2897 }
2898
2899 bool LibraryCallKit::inline_onspinwait() {
2900 insert_mem_bar(Op_OnSpinWait);
2901 return true;
2902 }
2903
2904 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2905 if (!kls->is_Con()) {
2906 return true;
2907 }
2908 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2909 if (klsptr == NULL) {
2910 return true;
2911 }
2912 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2913 // don't need a guard for a klass that is already initialized
2914 return !ik->is_initialized();
2915 }
2916
2917 //----------------------------inline_unsafe_writeback0-------------------------
2918 // public native void Unsafe.writeback0(long address)
2919 bool LibraryCallKit::inline_unsafe_writeback0() {
2920 if (!Matcher::has_match_rule(Op_CacheWB)) {
2921 return false;
2922 }
2923 #ifndef PRODUCT
2924 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2925 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2926 ciSignature* sig = callee()->signature();
2927 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2928 #endif
2929 null_check_receiver(); // null-check, then ignore
2930 Node *addr = argument(1);
2931 addr = new CastX2PNode(addr);
2932 addr = _gvn.transform(addr);
2933 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2934 flush = _gvn.transform(flush);
2935 set_memory(flush, TypeRawPtr::BOTTOM);
2936 return true;
2937 }
2938
2939 //----------------------------inline_unsafe_writeback0-------------------------
2940 // public native void Unsafe.writeback0(long address)
2941 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2942 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2943 return false;
2944 }
2945 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2946 return false;
2947 }
2948 #ifndef PRODUCT
2949 assert(Matcher::has_match_rule(Op_CacheWB),
2950 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2951 : "found match rule for CacheWBPostSync but not CacheWB"));
2952
2953 #endif
2954 null_check_receiver(); // null-check, then ignore
2955 Node *sync;
2956 if (is_pre) {
2957 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2958 } else {
2959 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2960 }
2961 sync = _gvn.transform(sync);
2962 set_memory(sync, TypeRawPtr::BOTTOM);
2963 return true;
2964 }
2965
2966 //----------------------------inline_unsafe_allocate---------------------------
2967 // public native Object Unsafe.allocateInstance(Class<?> cls);
2968 bool LibraryCallKit::inline_unsafe_allocate() {
2969 if (callee()->is_static()) return false; // caller must have the capability!
2970
2971 null_check_receiver(); // null-check, then ignore
2972 Node* cls = null_check(argument(1));
2973 if (stopped()) return true;
2974
2975 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2976 kls = null_check(kls);
2977 if (stopped()) return true; // argument was like int.class
2978
2979 Node* test = NULL;
2980 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2981 // Note: The argument might still be an illegal value like
2982 // Serializable.class or Object[].class. The runtime will handle it.
2983 // But we must make an explicit check for initialization.
2984 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2985 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2986 // can generate code to load it as unsigned byte.
2987 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2988 Node* bits = intcon(InstanceKlass::fully_initialized);
2989 test = _gvn.transform(new SubINode(inst, bits));
2990 // The 'test' is non-zero if we need to take a slow path.
2991 }
2992 Node* obj = NULL;
2993 ciKlass* klass = _gvn.type(kls)->is_klassptr()->klass();
2994 if (klass->is_inlinetype()) {
2995 obj = InlineTypeNode::make_default(_gvn, klass->as_inline_klass());
2996 } else {
2997 obj = new_instance(kls, test);
2998 }
2999 set_result(obj);
3000 return true;
3001 }
3002
3003 //------------------------inline_native_time_funcs--------------
3004 // inline code for System.currentTimeMillis() and System.nanoTime()
3005 // these have the same type and signature
3006 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3007 const TypeFunc* tf = OptoRuntime::void_long_Type();
3008 const TypePtr* no_memory_effects = NULL;
3009 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3010 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3011 #ifdef ASSERT
3012 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3013 assert(value_top == top(), "second value must be top");
3014 #endif
3015 set_result(value);
3016 return true;
3017 }
3018
3019 #ifdef JFR_HAVE_INTRINSICS
3020
3021 /**
3022 * if oop->klass != null
3023 * // normal class
3024 * epoch = _epoch_state ? 2 : 1
3025 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3026 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3027 * }
3028 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3029 * else
3030 * // primitive class
3031 * if oop->array_klass != null
3032 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3033 * else
3034 * id = LAST_TYPE_ID + 1 // void class path
3035 * if (!signaled)
3036 * signaled = true
3037 */
3038 bool LibraryCallKit::inline_native_classID() {
3039 Node* cls = argument(0);
3040
3041 IdealKit ideal(this);
3042 #define __ ideal.
3043 IdealVariable result(ideal); __ declarations_done();
3044 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3045 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3046 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3047
3048
3049 __ if_then(kls, BoolTest::ne, null()); {
3050 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3051 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3052
3053 Node* epoch_address = makecon(TypeRawPtr::make(Jfr::epoch_address()));
3054 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3055 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3056 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3057 mask = _gvn.transform(new OrLNode(mask, epoch));
3058 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3059
3060 float unlikely = PROB_UNLIKELY(0.999);
3061 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3062 sync_kit(ideal);
3063 make_runtime_call(RC_LEAF,
3064 OptoRuntime::get_class_id_intrinsic_Type(),
3065 CAST_FROM_FN_PTR(address, Jfr::get_class_id_intrinsic),
3066 "get_class_id_intrinsic",
3067 TypePtr::BOTTOM,
3068 kls);
3069 ideal.sync_kit(this);
3070 } __ end_if();
3071
3072 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3073 } __ else_(); {
3074 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
3075 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3076 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3077 __ if_then(array_kls, BoolTest::ne, null()); {
3078 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3079 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3080 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3081 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3082 } __ else_(); {
3083 // void class case
3084 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3085 } __ end_if();
3086
3087 Node* signaled_flag_address = makecon(TypeRawPtr::make(Jfr::signal_address()));
3088 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3089 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3090 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3091 } __ end_if();
3092 } __ end_if();
3093
3094 final_sync(ideal);
3095 set_result(ideal.value(result));
3096 #undef __
3097 return true;
3098 }
3099
3100 bool LibraryCallKit::inline_native_getEventWriter() {
3101 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3102
3103 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3104 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3105
3106 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3107
3108 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3109 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3110
3111 IfNode* iff_jobj_null =
3112 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3113
3114 enum { _normal_path = 1,
3115 _null_path = 2,
3116 PATH_LIMIT };
3117
3118 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3119 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3120
3121 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3122 result_rgn->init_req(_null_path, jobj_is_null);
3123 result_val->init_req(_null_path, null());
3124
3125 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3126 set_control(jobj_is_not_null);
3127 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3128 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3129 result_rgn->init_req(_normal_path, control());
3130 result_val->init_req(_normal_path, res);
3131
3132 set_result(result_rgn, result_val);
3133
3134 return true;
3135 }
3136
3137 #endif // JFR_HAVE_INTRINSICS
3138
3139 //------------------------inline_native_currentThread------------------
3140 bool LibraryCallKit::inline_native_currentThread() {
3141 Node* junk = NULL;
3142 set_result(generate_current_thread(junk));
3143 return true;
3144 }
3145
3146 //-----------------------load_klass_from_mirror_common-------------------------
3147 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3148 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3149 // and branch to the given path on the region.
3150 // If never_see_null, take an uncommon trap on null, so we can optimistically
3151 // compile for the non-null case.
3152 // If the region is NULL, force never_see_null = true.
3153 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3154 bool never_see_null,
3155 RegionNode* region,
3156 int null_path,
3157 int offset) {
3158 if (region == NULL) never_see_null = true;
3159 Node* p = basic_plus_adr(mirror, offset);
3160 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3161 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3162 Node* null_ctl = top();
3163 kls = null_check_oop(kls, &null_ctl, never_see_null);
3164 if (region != NULL) {
3165 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3166 region->init_req(null_path, null_ctl);
3167 } else {
3168 assert(null_ctl == top(), "no loose ends");
3169 }
3170 return kls;
3171 }
3172
3173 //--------------------(inline_native_Class_query helpers)---------------------
3174 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3175 // Fall through if (mods & mask) == bits, take the guard otherwise.
3176 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3177 // Branch around if the given klass has the given modifier bit set.
3178 // Like generate_guard, adds a new path onto the region.
3179 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3180 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3181 Node* mask = intcon(modifier_mask);
3182 Node* bits = intcon(modifier_bits);
3183 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3184 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3185 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3186 return generate_fair_guard(bol, region);
3187 }
3188
3189 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3190 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3191 }
3192 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3193 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3194 }
3195
3196 //-------------------------inline_native_Class_query-------------------
3197 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3198 const Type* return_type = TypeInt::BOOL;
3199 Node* prim_return_value = top(); // what happens if it's a primitive class?
3200 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3201 bool expect_prim = false; // most of these guys expect to work on refs
3202
3203 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3204
3205 Node* mirror = argument(0);
3206 Node* obj = top();
3207
3208 switch (id) {
3209 case vmIntrinsics::_isInstance:
3210 // nothing is an instance of a primitive type
3211 prim_return_value = intcon(0);
3212 obj = argument(1);
3213 break;
3214 case vmIntrinsics::_getModifiers:
3215 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3216 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3217 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3218 break;
3219 case vmIntrinsics::_isInterface:
3220 prim_return_value = intcon(0);
3221 break;
3222 case vmIntrinsics::_isArray:
3223 prim_return_value = intcon(0);
3224 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3225 break;
3226 case vmIntrinsics::_isPrimitive:
3227 prim_return_value = intcon(1);
3228 expect_prim = true; // obviously
3229 break;
3230 case vmIntrinsics::_isHidden:
3231 prim_return_value = intcon(0);
3232 break;
3233 case vmIntrinsics::_getSuperclass:
3234 prim_return_value = null();
3235 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3236 break;
3237 case vmIntrinsics::_getClassAccessFlags:
3238 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3239 return_type = TypeInt::INT; // not bool! 6297094
3240 break;
3241 default:
3242 fatal_unexpected_iid(id);
3243 break;
3244 }
3245
3246 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3247 if (mirror_con == NULL) return false; // cannot happen?
3248
3249 #ifndef PRODUCT
3250 if (C->print_intrinsics() || C->print_inlining()) {
3251 ciType* k = mirror_con->java_mirror_type();
3252 if (k) {
3253 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3254 k->print_name();
3255 tty->cr();
3256 }
3257 }
3258 #endif
3259
3260 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3261 RegionNode* region = new RegionNode(PATH_LIMIT);
3262 record_for_igvn(region);
3263 PhiNode* phi = new PhiNode(region, return_type);
3264
3265 // The mirror will never be null of Reflection.getClassAccessFlags, however
3266 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3267 // if it is. See bug 4774291.
3268
3269 // For Reflection.getClassAccessFlags(), the null check occurs in
3270 // the wrong place; see inline_unsafe_access(), above, for a similar
3271 // situation.
3272 mirror = null_check(mirror);
3273 // If mirror or obj is dead, only null-path is taken.
3274 if (stopped()) return true;
3275
3276 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3277
3278 // Now load the mirror's klass metaobject, and null-check it.
3279 // Side-effects region with the control path if the klass is null.
3280 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3281 // If kls is null, we have a primitive mirror.
3282 phi->init_req(_prim_path, prim_return_value);
3283 if (stopped()) { set_result(region, phi); return true; }
3284 bool safe_for_replace = (region->in(_prim_path) == top());
3285
3286 Node* p; // handy temp
3287 Node* null_ctl;
3288
3289 // Now that we have the non-null klass, we can perform the real query.
3290 // For constant classes, the query will constant-fold in LoadNode::Value.
3291 Node* query_value = top();
3292 switch (id) {
3293 case vmIntrinsics::_isInstance:
3294 // nothing is an instance of a primitive type
3295 query_value = gen_instanceof(obj, kls, safe_for_replace);
3296 break;
3297
3298 case vmIntrinsics::_getModifiers:
3299 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3300 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3301 break;
3302
3303 case vmIntrinsics::_isInterface:
3304 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3305 if (generate_interface_guard(kls, region) != NULL)
3306 // A guard was added. If the guard is taken, it was an interface.
3307 phi->add_req(intcon(1));
3308 // If we fall through, it's a plain class.
3309 query_value = intcon(0);
3310 break;
3311
3312 case vmIntrinsics::_isArray:
3313 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3314 if (generate_array_guard(kls, region) != NULL)
3315 // A guard was added. If the guard is taken, it was an array.
3316 phi->add_req(intcon(1));
3317 // If we fall through, it's a plain class.
3318 query_value = intcon(0);
3319 break;
3320
3321 case vmIntrinsics::_isPrimitive:
3322 query_value = intcon(0); // "normal" path produces false
3323 break;
3324
3325 case vmIntrinsics::_isHidden:
3326 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3327 if (generate_hidden_class_guard(kls, region) != NULL)
3328 // A guard was added. If the guard is taken, it was an hidden class.
3329 phi->add_req(intcon(1));
3330 // If we fall through, it's a plain class.
3331 query_value = intcon(0);
3332 break;
3333
3334
3335 case vmIntrinsics::_getSuperclass:
3336 // The rules here are somewhat unfortunate, but we can still do better
3337 // with random logic than with a JNI call.
3338 // Interfaces store null or Object as _super, but must report null.
3339 // Arrays store an intermediate super as _super, but must report Object.
3340 // Other types can report the actual _super.
3341 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3342 if (generate_interface_guard(kls, region) != NULL)
3343 // A guard was added. If the guard is taken, it was an interface.
3344 phi->add_req(null());
3345 if (generate_array_guard(kls, region) != NULL)
3346 // A guard was added. If the guard is taken, it was an array.
3347 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3348 // If we fall through, it's a plain class. Get its _super.
3349 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3350 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3351 null_ctl = top();
3352 kls = null_check_oop(kls, &null_ctl);
3353 if (null_ctl != top()) {
3354 // If the guard is taken, Object.superClass is null (both klass and mirror).
3355 region->add_req(null_ctl);
3356 phi ->add_req(null());
3357 }
3358 if (!stopped()) {
3359 query_value = load_mirror_from_klass(kls);
3360 }
3361 break;
3362
3363 case vmIntrinsics::_getClassAccessFlags:
3364 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3365 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3366 break;
3367
3368 default:
3369 fatal_unexpected_iid(id);
3370 break;
3371 }
3372
3373 // Fall-through is the normal case of a query to a real class.
3374 phi->init_req(1, query_value);
3375 region->init_req(1, control());
3376
3377 C->set_has_split_ifs(true); // Has chance for split-if optimization
3378 set_result(region, phi);
3379 return true;
3380 }
3381
3382 //-------------------------inline_primitive_Class_conversion-------------------
3383 // public Class<T> java.lang.Class.asPrimaryType();
3384 // public Class<T> java.lang.Class.asValueType()
3385 bool LibraryCallKit::inline_primitive_Class_conversion(vmIntrinsics::ID id) {
3386 Node* mirror = argument(0); // Receiver Class
3387 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3388 if (mirror_con == NULL) {
3389 return false;
3390 }
3391
3392 bool is_val_mirror = true;
3393 ciType* tm = mirror_con->java_mirror_type(&is_val_mirror);
3394 if (tm != NULL) {
3395 Node* result = mirror;
3396 if (id == vmIntrinsics::_asPrimaryType && is_val_mirror) {
3397 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->ref_mirror()));
3398 } else if (id == vmIntrinsics::_asValueType) {
3399 if (!tm->is_inlinetype()) {
3400 return false; // Throw UnsupportedOperationException
3401 } else if (!is_val_mirror) {
3402 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->val_mirror()));
3403 }
3404 }
3405 set_result(result);
3406 return true;
3407 }
3408 return false;
3409 }
3410
3411 //-------------------------inline_Class_cast-------------------
3412 bool LibraryCallKit::inline_Class_cast() {
3413 Node* mirror = argument(0); // Class
3414 Node* obj = argument(1);
3415 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3416 if (mirror_con == NULL) {
3417 return false; // dead path (mirror->is_top()).
3418 }
3419 if (obj == NULL || obj->is_top()) {
3420 return false; // dead path
3421 }
3422 ciKlass* obj_klass = NULL;
3423 const Type* obj_t = _gvn.type(obj);
3424 if (obj->is_InlineType()) {
3425 obj_klass = obj_t->inline_klass();
3426 } else if (obj_t->isa_oopptr()) {
3427 obj_klass = obj_t->is_oopptr()->klass();
3428 }
3429
3430 // First, see if Class.cast() can be folded statically.
3431 // java_mirror_type() returns non-null for compile-time Class constants.
3432 bool requires_null_check = false;
3433 ciType* tm = mirror_con->java_mirror_type(&requires_null_check);
3434 // Check for null if casting to QMyValue
3435 requires_null_check &= !obj->is_InlineType();
3436 if (tm != NULL && tm->is_klass() && obj_klass != NULL) {
3437 if (!obj_klass->is_loaded()) {
3438 // Don't use intrinsic when class is not loaded.
3439 return false;
3440 } else {
3441 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass);
3442 if (static_res == Compile::SSC_always_true) {
3443 // isInstance() is true - fold the code.
3444 if (requires_null_check) {
3445 obj = null_check(obj);
3446 }
3447 set_result(obj);
3448 return true;
3449 } else if (static_res == Compile::SSC_always_false) {
3450 // Don't use intrinsic, have to throw ClassCastException.
3451 // If the reference is null, the non-intrinsic bytecode will
3452 // be optimized appropriately.
3453 return false;
3454 }
3455 }
3456 }
3457
3458 // Bailout intrinsic and do normal inlining if exception path is frequent.
3459 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3460 return false;
3461 }
3462
3463 // Generate dynamic checks.
3464 // Class.cast() is java implementation of _checkcast bytecode.
3465 // Do checkcast (Parse::do_checkcast()) optimizations here.
3466
3467 if (requires_null_check) {
3468 obj = null_check(obj);
3469 }
3470 mirror = null_check(mirror);
3471 // If mirror is dead, only null-path is taken.
3472 if (stopped()) {
3473 return true;
3474 }
3475
3476 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3477 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
3478 RegionNode* region = new RegionNode(PATH_LIMIT);
3479 record_for_igvn(region);
3480
3481 // Now load the mirror's klass metaobject, and null-check it.
3482 // If kls is null, we have a primitive mirror and
3483 // nothing is an instance of a primitive type.
3484 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3485
3486 Node* res = top();
3487 if (!stopped()) {
3488 if (EnableValhalla && !obj->is_InlineType() && !requires_null_check) {
3489 // Check if we are casting to QMyValue
3490 Node* ctrl_val_mirror = generate_fair_guard(is_val_mirror(mirror), NULL);
3491 if (ctrl_val_mirror != NULL) {
3492 RegionNode* r = new RegionNode(3);
3493 record_for_igvn(r);
3494 r->init_req(1, control());
3495
3496 // Casting to QMyValue, check for null
3497 set_control(ctrl_val_mirror);
3498 { // PreserveJVMState because null check replaces obj in map
3499 PreserveJVMState pjvms(this);
3500 Node* null_ctr = top();
3501 null_check_oop(obj, &null_ctr);
3502 region->init_req(_npe_path, null_ctr);
3503 r->init_req(2, control());
3504 }
3505 set_control(_gvn.transform(r));
3506 }
3507 }
3508
3509 Node* bad_type_ctrl = top();
3510 // Do checkcast optimizations.
3511 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3512 region->init_req(_bad_type_path, bad_type_ctrl);
3513 }
3514 if (region->in(_prim_path) != top() ||
3515 region->in(_bad_type_path) != top() ||
3516 region->in(_npe_path) != top()) {
3517 // Let Interpreter throw ClassCastException.
3518 PreserveJVMState pjvms(this);
3519 set_control(_gvn.transform(region));
3520 uncommon_trap(Deoptimization::Reason_intrinsic,
3521 Deoptimization::Action_maybe_recompile);
3522 }
3523 if (!stopped()) {
3524 set_result(res);
3525 }
3526 return true;
3527 }
3528
3529
3530 //--------------------------inline_native_subtype_check------------------------
3531 // This intrinsic takes the JNI calls out of the heart of
3532 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3533 bool LibraryCallKit::inline_native_subtype_check() {
3534 // Pull both arguments off the stack.
3535 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3536 args[0] = argument(0);
3537 args[1] = argument(1);
3538 Node* klasses[2]; // corresponding Klasses: superk, subk
3539 klasses[0] = klasses[1] = top();
3540
3541 enum {
3542 // A full decision tree on {superc is prim, subc is prim}:
3543 _prim_0_path = 1, // {P,N} => false
3544 // {P,P} & superc!=subc => false
3545 _prim_same_path, // {P,P} & superc==subc => true
3546 _prim_1_path, // {N,P} => false
3547 _ref_subtype_path, // {N,N} & subtype check wins => true
3548 _both_ref_path, // {N,N} & subtype check loses => false
3549 PATH_LIMIT
3550 };
3551
3552 RegionNode* region = new RegionNode(PATH_LIMIT);
3553 RegionNode* prim_region = new RegionNode(2);
3554 Node* phi = new PhiNode(region, TypeInt::BOOL);
3555 record_for_igvn(region);
3556 record_for_igvn(prim_region);
3557
3558 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3559 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3560 int class_klass_offset = java_lang_Class::klass_offset();
3561
3562 // First null-check both mirrors and load each mirror's klass metaobject.
3563 int which_arg;
3564 for (which_arg = 0; which_arg <= 1; which_arg++) {
3565 Node* arg = args[which_arg];
3566 arg = null_check(arg);
3567 if (stopped()) break;
3568 args[which_arg] = arg;
3569
3570 Node* p = basic_plus_adr(arg, class_klass_offset);
3571 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3572 klasses[which_arg] = _gvn.transform(kls);
3573 }
3574
3575 // Having loaded both klasses, test each for null.
3576 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3577 for (which_arg = 0; which_arg <= 1; which_arg++) {
3578 Node* kls = klasses[which_arg];
3579 Node* null_ctl = top();
3580 kls = null_check_oop(kls, &null_ctl, never_see_null);
3581 if (which_arg == 0) {
3582 prim_region->init_req(1, null_ctl);
3583 } else {
3584 region->init_req(_prim_1_path, null_ctl);
3585 }
3586 if (stopped()) break;
3587 klasses[which_arg] = kls;
3588 }
3589
3590 if (!stopped()) {
3591 // now we have two reference types, in klasses[0..1]
3592 Node* subk = klasses[1]; // the argument to isAssignableFrom
3593 Node* superk = klasses[0]; // the receiver
3594 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3595 // If superc is an inline mirror, we also need to check if superc == subc because LMyValue
3596 // is not a subtype of QMyValue but due to subk == superk the subtype check will pass.
3597 generate_fair_guard(is_val_mirror(args[0]), prim_region);
3598 // now we have a successful reference subtype check
3599 region->set_req(_ref_subtype_path, control());
3600 }
3601
3602 // If both operands are primitive (both klasses null), then
3603 // we must return true when they are identical primitives.
3604 // It is convenient to test this after the first null klass check.
3605 // This path is also used if superc is a value mirror.
3606 set_control(_gvn.transform(prim_region));
3607 if (!stopped()) {
3608 // Since superc is primitive, make a guard for the superc==subc case.
3609 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3610 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3611 generate_fair_guard(bol_eq, region);
3612 if (region->req() == PATH_LIMIT+1) {
3613 // A guard was added. If the added guard is taken, superc==subc.
3614 region->swap_edges(PATH_LIMIT, _prim_same_path);
3615 region->del_req(PATH_LIMIT);
3616 }
3617 region->set_req(_prim_0_path, control()); // Not equal after all.
3618 }
3619
3620 // these are the only paths that produce 'true':
3621 phi->set_req(_prim_same_path, intcon(1));
3622 phi->set_req(_ref_subtype_path, intcon(1));
3623
3624 // pull together the cases:
3625 assert(region->req() == PATH_LIMIT, "sane region");
3626 for (uint i = 1; i < region->req(); i++) {
3627 Node* ctl = region->in(i);
3628 if (ctl == NULL || ctl == top()) {
3629 region->set_req(i, top());
3630 phi ->set_req(i, top());
3631 } else if (phi->in(i) == NULL) {
3632 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3633 }
3634 }
3635
3636 set_control(_gvn.transform(region));
3637 set_result(_gvn.transform(phi));
3638 return true;
3639 }
3640
3641 //---------------------generate_array_guard_common------------------------
3642 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
3643
3644 if (stopped()) {
3645 return NULL;
3646 }
3647
3648 // Like generate_guard, adds a new path onto the region.
3649 jint layout_con = 0;
3650 Node* layout_val = get_layout_helper(kls, layout_con);
3651 if (layout_val == NULL) {
3652 bool query = 0;
3653 switch(kind) {
3654 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
3655 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
3656 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
3657 case FlatArray: query = Klass::layout_helper_is_flatArray(layout_con); break;
3658 case NonFlatArray: query = !Klass::layout_helper_is_flatArray(layout_con); break;
3659 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
3660 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
3661 default:
3662 ShouldNotReachHere();
3663 }
3664 if (!query) {
3665 return NULL; // never a branch
3666 } else { // always a branch
3667 Node* always_branch = control();
3668 if (region != NULL)
3669 region->add_req(always_branch);
3670 set_control(top());
3671 return always_branch;
3672 }
3673 }
3674 unsigned int value = 0;
3675 BoolTest::mask btest = BoolTest::illegal;
3676 switch(kind) {
3677 case ObjectArray:
3678 case NonObjectArray: {
3679 value = Klass::_lh_array_tag_obj_value;
3680 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3681 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
3682 break;
3683 }
3684 case TypeArray: {
3685 value = Klass::_lh_array_tag_type_value;
3686 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3687 btest = BoolTest::eq;
3688 break;
3689 }
3690 case FlatArray:
3691 case NonFlatArray: {
3692 value = 0;
3693 layout_val = _gvn.transform(new AndINode(layout_val, intcon(Klass::_lh_array_tag_vt_value_bit_inplace)));
3694 btest = (kind == FlatArray) ? BoolTest::ne : BoolTest::eq;
3695 break;
3696 }
3697 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
3698 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
3699 default:
3700 ShouldNotReachHere();
3701 }
3702 // Now test the correct condition.
3703 jint nval = (jint)value;
3704 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3705 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3706 return generate_fair_guard(bol, region);
3707 }
3708
3709
3710 //-----------------------inline_native_newArray--------------------------
3711 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
3712 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3713 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3714 Node* mirror;
3715 Node* count_val;
3716 if (uninitialized) {
3717 mirror = argument(1);
3718 count_val = argument(2);
3719 } else {
3720 mirror = argument(0);
3721 count_val = argument(1);
3722 }
3723
3724 mirror = null_check(mirror);
3725 // If mirror or obj is dead, only null-path is taken.
3726 if (stopped()) return true;
3727
3728 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3729 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3730 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3731 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3732 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3733
3734 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3735 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3736 result_reg, _slow_path);
3737 Node* normal_ctl = control();
3738 Node* no_array_ctl = result_reg->in(_slow_path);
3739
3740 // Generate code for the slow case. We make a call to newArray().
3741 set_control(no_array_ctl);
3742 if (!stopped()) {
3743 // Either the input type is void.class, or else the
3744 // array klass has not yet been cached. Either the
3745 // ensuing call will throw an exception, or else it
3746 // will cache the array klass for next time.
3747 PreserveJVMState pjvms(this);
3748 CallJavaNode* slow_call = NULL;
3749 if (uninitialized) {
3750 // Generate optimized virtual call (holder class 'Unsafe' is final)
3751 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3752 } else {
3753 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3754 }
3755 Node* slow_result = set_results_for_java_call(slow_call);
3756 // this->control() comes from set_results_for_java_call
3757 result_reg->set_req(_slow_path, control());
3758 result_val->set_req(_slow_path, slow_result);
3759 result_io ->set_req(_slow_path, i_o());
3760 result_mem->set_req(_slow_path, reset_memory());
3761 }
3762
3763 set_control(normal_ctl);
3764 if (!stopped()) {
3765 // Normal case: The array type has been cached in the java.lang.Class.
3766 // The following call works fine even if the array type is polymorphic.
3767 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3768 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3769 result_reg->init_req(_normal_path, control());
3770 result_val->init_req(_normal_path, obj);
3771 result_io ->init_req(_normal_path, i_o());
3772 result_mem->init_req(_normal_path, reset_memory());
3773
3774 if (uninitialized) {
3775 // Mark the allocation so that zeroing is skipped
3776 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3777 alloc->maybe_set_complete(&_gvn);
3778 }
3779 }
3780
3781 // Return the combined state.
3782 set_i_o( _gvn.transform(result_io) );
3783 set_all_memory( _gvn.transform(result_mem));
3784
3785 C->set_has_split_ifs(true); // Has chance for split-if optimization
3786 set_result(result_reg, result_val);
3787 return true;
3788 }
3789
3790 //----------------------inline_native_getLength--------------------------
3791 // public static native int java.lang.reflect.Array.getLength(Object array);
3792 bool LibraryCallKit::inline_native_getLength() {
3793 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3794
3795 Node* array = null_check(argument(0));
3796 // If array is dead, only null-path is taken.
3797 if (stopped()) return true;
3798
3799 // Deoptimize if it is a non-array.
3800 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3801
3802 if (non_array != NULL) {
3803 PreserveJVMState pjvms(this);
3804 set_control(non_array);
3805 uncommon_trap(Deoptimization::Reason_intrinsic,
3806 Deoptimization::Action_maybe_recompile);
3807 }
3808
3809 // If control is dead, only non-array-path is taken.
3810 if (stopped()) return true;
3811
3812 // The works fine even if the array type is polymorphic.
3813 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3814 Node* result = load_array_length(array);
3815
3816 C->set_has_split_ifs(true); // Has chance for split-if optimization
3817 set_result(result);
3818 return true;
3819 }
3820
3821 //------------------------inline_array_copyOf----------------------------
3822 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3823 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3824 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3825 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3826
3827 // Get the arguments.
3828 Node* original = argument(0);
3829 Node* start = is_copyOfRange? argument(1): intcon(0);
3830 Node* end = is_copyOfRange? argument(2): argument(1);
3831 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3832
3833 Node* newcopy = NULL;
3834
3835 // Set the original stack and the reexecute bit for the interpreter to reexecute
3836 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3837 { PreserveReexecuteState preexecs(this);
3838 jvms()->set_should_reexecute(true);
3839
3840 array_type_mirror = null_check(array_type_mirror);
3841 original = null_check(original);
3842
3843 // Check if a null path was taken unconditionally.
3844 if (stopped()) return true;
3845
3846 Node* orig_length = load_array_length(original);
3847
3848 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3849 klass_node = null_check(klass_node);
3850
3851 RegionNode* bailout = new RegionNode(1);
3852 record_for_igvn(bailout);
3853
3854 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3855 // Bail out if that is so.
3856 // Inline type array may have object field that would require a
3857 // write barrier. Conservatively, go to slow path.
3858 // TODO 8251971: Optimize for the case when flat src/dst are later found
3859 // to not contain oops (i.e., move this check to the macro expansion phase).
3860 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
3861 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
3862 ciKlass* klass = _gvn.type(klass_node)->is_klassptr()->klass();
3863 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
3864 // Can src array be flat and contain oops?
3865 (orig_t == NULL || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
3866 // Can dest array be flat and contain oops?
3867 klass->can_be_inline_array_klass() && (!klass->is_flat_array_klass() || klass->as_flat_array_klass()->element_klass()->as_inline_klass()->contains_oops());
3868 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
3869 if (not_objArray != NULL) {
3870 // Improve the klass node's type from the new optimistic assumption:
3871 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3872 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
3873 Node* cast = new CastPPNode(klass_node, akls);
3874 cast->init_req(0, control());
3875 klass_node = _gvn.transform(cast);
3876 }
3877
3878 Node* original_kls = load_object_klass(original);
3879 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3880 // loads/stores but it is legal only if we're sure the
3881 // Arrays.copyOf would succeed. So we need all input arguments
3882 // to the copyOf to be validated, including that the copy to the
3883 // new array won't trigger an ArrayStoreException. That subtype
3884 // check can be optimized if we know something on the type of
3885 // the input array from type speculation.
3886 if (_gvn.type(klass_node)->singleton() && !stopped()) {
3887 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass();
3888 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3889
3890 int test = C->static_subtype_check(superk, subk);
3891 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3892 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3893 if (t_original->speculative_type() != NULL) {
3894 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3895 original_kls = load_object_klass(original);
3896 }
3897 }
3898 }
3899
3900 // Bail out if either start or end is negative.
3901 generate_negative_guard(start, bailout, &start);
3902 generate_negative_guard(end, bailout, &end);
3903
3904 Node* length = end;
3905 if (_gvn.type(start) != TypeInt::ZERO) {
3906 length = _gvn.transform(new SubINode(end, start));
3907 }
3908
3909 // Bail out if length is negative.
3910 // Without this the new_array would throw
3911 // NegativeArraySizeException but IllegalArgumentException is what
3912 // should be thrown
3913 generate_negative_guard(length, bailout, &length);
3914
3915 // Handle inline type arrays
3916 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
3917 if (!stopped()) {
3918 orig_t = _gvn.type(original)->isa_aryptr();
3919 if (orig_t != NULL && orig_t->is_flat()) {
3920 // Src is flat, check that dest is flat as well
3921 if (exclude_flat) {
3922 // Dest can't be flat, bail out
3923 bailout->add_req(control());
3924 set_control(top());
3925 } else {
3926 generate_non_flatArray_guard(klass_node, bailout);
3927 }
3928 } else if (UseFlatArray && (orig_t == NULL || !orig_t->is_not_flat()) &&
3929 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
3930 ((!klass->is_flat_array_klass() && klass->can_be_inline_array_klass()) || !can_validate)) {
3931 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
3932 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
3933 generate_flatArray_guard(original_kls, bailout);
3934 if (orig_t != NULL) {
3935 orig_t = orig_t->cast_to_not_flat();
3936 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
3937 }
3938 }
3939 if (!can_validate) {
3940 // No validation. The subtype check emitted at macro expansion time will not go to the slow
3941 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
3942 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
3943 generate_fair_guard(null_free_array_test(klass_node), bailout);
3944 }
3945 }
3946
3947 if (bailout->req() > 1) {
3948 PreserveJVMState pjvms(this);
3949 set_control(_gvn.transform(bailout));
3950 uncommon_trap(Deoptimization::Reason_intrinsic,
3951 Deoptimization::Action_maybe_recompile);
3952 }
3953
3954 if (!stopped()) {
3955 // How many elements will we copy from the original?
3956 // The answer is MinI(orig_length - start, length).
3957 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3958 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3959
3960 // Generate a direct call to the right arraycopy function(s).
3961 // We know the copy is disjoint but we might not know if the
3962 // oop stores need checking.
3963 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3964 // This will fail a store-check if x contains any non-nulls.
3965
3966 bool validated = false;
3967 // Reason_class_check rather than Reason_intrinsic because we
3968 // want to intrinsify even if this traps.
3969 if (can_validate) {
3970 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3971
3972 if (not_subtype_ctrl != top()) {
3973 PreserveJVMState pjvms(this);
3974 set_control(not_subtype_ctrl);
3975 uncommon_trap(Deoptimization::Reason_class_check,
3976 Deoptimization::Action_make_not_entrant);
3977 assert(stopped(), "Should be stopped");
3978 }
3979 validated = true;
3980 }
3981
3982 if (!stopped()) {
3983 newcopy = new_array(klass_node, length, 0); // no arguments to push
3984
3985 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3986 original_kls, klass_node);
3987 if (!is_copyOfRange) {
3988 ac->set_copyof(validated);
3989 } else {
3990 ac->set_copyofrange(validated);
3991 }
3992 Node* n = _gvn.transform(ac);
3993 if (n == ac) {
3994 ac->connect_outputs(this);
3995 } else {
3996 assert(validated, "shouldn't transform if all arguments not validated");
3997 set_all_memory(n);
3998 }
3999 }
4000 }
4001 } // original reexecute is set back here
4002
4003 C->set_has_split_ifs(true); // Has chance for split-if optimization
4004 if (!stopped()) {
4005 set_result(newcopy);
4006 }
4007 return true;
4008 }
4009
4010
4011 //----------------------generate_virtual_guard---------------------------
4012 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4013 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4014 RegionNode* slow_region) {
4015 ciMethod* method = callee();
4016 int vtable_index = method->vtable_index();
4017 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4018 "bad index %d", vtable_index);
4019 // Get the Method* out of the appropriate vtable entry.
4020 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4021 vtable_index*vtableEntry::size_in_bytes() +
4022 vtableEntry::method_offset_in_bytes();
4023 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4024 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4025
4026 // Compare the target method with the expected method (e.g., Object.hashCode).
4027 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4028
4029 Node* native_call = makecon(native_call_addr);
4030 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4031 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4032
4033 return generate_slow_guard(test_native, slow_region);
4034 }
4035
4036 //-----------------------generate_method_call----------------------------
4037 // Use generate_method_call to make a slow-call to the real
4038 // method if the fast path fails. An alternative would be to
4039 // use a stub like OptoRuntime::slow_arraycopy_Java.
4040 // This only works for expanding the current library call,
4041 // not another intrinsic. (E.g., don't use this for making an
4042 // arraycopy call inside of the copyOf intrinsic.)
4043 CallJavaNode*
4044 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4045 // When compiling the intrinsic method itself, do not use this technique.
4046 guarantee(callee() != C->method(), "cannot make slow-call to self");
4047
4048 ciMethod* method = callee();
4049 // ensure the JVMS we have will be correct for this call
4050 guarantee(method_id == method->intrinsic_id(), "must match");
4051
4052 const TypeFunc* tf = TypeFunc::make(method);
4053 CallJavaNode* slow_call;
4054 if (is_static) {
4055 assert(!is_virtual, "");
4056 slow_call = new CallStaticJavaNode(C, tf,
4057 SharedRuntime::get_resolve_static_call_stub(), method);
4058 } else if (is_virtual) {
4059 null_check_receiver();
4060 int vtable_index = Method::invalid_vtable_index;
4061 if (UseInlineCaches) {
4062 // Suppress the vtable call
4063 } else {
4064 // hashCode and clone are not a miranda methods,
4065 // so the vtable index is fixed.
4066 // No need to use the linkResolver to get it.
4067 vtable_index = method->vtable_index();
4068 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4069 "bad index %d", vtable_index);
4070 }
4071 slow_call = new CallDynamicJavaNode(tf,
4072 SharedRuntime::get_resolve_virtual_call_stub(),
4073 method, vtable_index);
4074 } else { // neither virtual nor static: opt_virtual
4075 null_check_receiver();
4076 slow_call = new CallStaticJavaNode(C, tf,
4077 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4078 slow_call->set_optimized_virtual(true);
4079 }
4080 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4081 // To be able to issue a direct call (optimized virtual or virtual)
4082 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4083 // about the method being invoked should be attached to the call site to
4084 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4085 slow_call->set_override_symbolic_info(true);
4086 }
4087 set_arguments_for_java_call(slow_call);
4088 set_edges_for_java_call(slow_call);
4089 return slow_call;
4090 }
4091
4092
4093 /**
4094 * Build special case code for calls to hashCode on an object. This call may
4095 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4096 * slightly different code.
4097 */
4098 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4099 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4100 assert(!(is_virtual && is_static), "either virtual, special, or static");
4101
4102 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4103
4104 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4105 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4106 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4107 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4108 Node* obj = argument(0);
4109
4110 if (obj->is_InlineType() || gvn().type(obj)->is_inlinetypeptr()) {
4111 return false;
4112 }
4113
4114 if (!is_static) {
4115 // Check for hashing null object
4116 obj = null_check_receiver();
4117 if (stopped()) return true; // unconditionally null
4118 result_reg->init_req(_null_path, top());
4119 result_val->init_req(_null_path, top());
4120 } else {
4121 // Do a null check, and return zero if null.
4122 // System.identityHashCode(null) == 0
4123 Node* null_ctl = top();
4124 obj = null_check_oop(obj, &null_ctl);
4125 result_reg->init_req(_null_path, null_ctl);
4126 result_val->init_req(_null_path, _gvn.intcon(0));
4127 }
4128
4129 // Unconditionally null? Then return right away.
4130 if (stopped()) {
4131 set_control( result_reg->in(_null_path));
4132 if (!stopped())
4133 set_result(result_val->in(_null_path));
4134 return true;
4135 }
4136
4137 // We only go to the fast case code if we pass a number of guards. The
4138 // paths which do not pass are accumulated in the slow_region.
4139 RegionNode* slow_region = new RegionNode(1);
4140 record_for_igvn(slow_region);
4141
4142 // If this is a virtual call, we generate a funny guard. We pull out
4143 // the vtable entry corresponding to hashCode() from the target object.
4144 // If the target method which we are calling happens to be the native
4145 // Object hashCode() method, we pass the guard. We do not need this
4146 // guard for non-virtual calls -- the caller is known to be the native
4147 // Object hashCode().
4148 if (is_virtual) {
4149 // After null check, get the object's klass.
4150 Node* obj_klass = load_object_klass(obj);
4151 generate_virtual_guard(obj_klass, slow_region);
4152 }
4153
4154 // Get the header out of the object, use LoadMarkNode when available
4155 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4156 // The control of the load must be NULL. Otherwise, the load can move before
4157 // the null check after castPP removal.
4158 Node* no_ctrl = NULL;
4159 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4160
4161 // Test the header to see if it is unlocked.
4162 // This also serves as guard against inline types
4163 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4164 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4165 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4166 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4167 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4168
4169 generate_slow_guard(test_unlocked, slow_region);
4170
4171 // Get the hash value and check to see that it has been properly assigned.
4172 // We depend on hash_mask being at most 32 bits and avoid the use of
4173 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4174 // vm: see markWord.hpp.
4175 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4176 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4177 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4178 // This hack lets the hash bits live anywhere in the mark object now, as long
4179 // as the shift drops the relevant bits into the low 32 bits. Note that
4180 // Java spec says that HashCode is an int so there's no point in capturing
4181 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4182 hshifted_header = ConvX2I(hshifted_header);
4183 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4184
4185 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4186 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4187 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4188
4189 generate_slow_guard(test_assigned, slow_region);
4190
4191 Node* init_mem = reset_memory();
4192 // fill in the rest of the null path:
4193 result_io ->init_req(_null_path, i_o());
4194 result_mem->init_req(_null_path, init_mem);
4195
4196 result_val->init_req(_fast_path, hash_val);
4197 result_reg->init_req(_fast_path, control());
4198 result_io ->init_req(_fast_path, i_o());
4199 result_mem->init_req(_fast_path, init_mem);
4200
4201 // Generate code for the slow case. We make a call to hashCode().
4202 set_control(_gvn.transform(slow_region));
4203 if (!stopped()) {
4204 // No need for PreserveJVMState, because we're using up the present state.
4205 set_all_memory(init_mem);
4206 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4207 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4208 Node* slow_result = set_results_for_java_call(slow_call);
4209 // this->control() comes from set_results_for_java_call
4210 result_reg->init_req(_slow_path, control());
4211 result_val->init_req(_slow_path, slow_result);
4212 result_io ->set_req(_slow_path, i_o());
4213 result_mem ->set_req(_slow_path, reset_memory());
4214 }
4215
4216 // Return the combined state.
4217 set_i_o( _gvn.transform(result_io) );
4218 set_all_memory( _gvn.transform(result_mem));
4219
4220 set_result(result_reg, result_val);
4221 return true;
4222 }
4223
4224 //---------------------------inline_native_getClass----------------------------
4225 // public final native Class<?> java.lang.Object.getClass();
4226 //
4227 // Build special case code for calls to getClass on an object.
4228 bool LibraryCallKit::inline_native_getClass() {
4229 Node* obj = argument(0);
4230 if (obj->is_InlineTypeBase()) {
4231 const Type* t = _gvn.type(obj);
4232 if (t->maybe_null()) {
4233 null_check(obj);
4234 }
4235 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4236 return true;
4237 }
4238 obj = null_check_receiver();
4239 if (stopped()) return true;
4240 set_result(load_mirror_from_klass(load_object_klass(obj)));
4241 return true;
4242 }
4243
4244 //-----------------inline_native_Reflection_getCallerClass---------------------
4245 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4246 //
4247 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4248 //
4249 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4250 // in that it must skip particular security frames and checks for
4251 // caller sensitive methods.
4252 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4253 #ifndef PRODUCT
4254 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4255 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4256 }
4257 #endif
4258
4259 if (!jvms()->has_method()) {
4260 #ifndef PRODUCT
4261 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4262 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4263 }
4264 #endif
4265 return false;
4266 }
4267
4268 // Walk back up the JVM state to find the caller at the required
4269 // depth.
4270 JVMState* caller_jvms = jvms();
4271
4272 // Cf. JVM_GetCallerClass
4273 // NOTE: Start the loop at depth 1 because the current JVM state does
4274 // not include the Reflection.getCallerClass() frame.
4275 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4276 ciMethod* m = caller_jvms->method();
4277 switch (n) {
4278 case 0:
4279 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4280 break;
4281 case 1:
4282 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4283 if (!m->caller_sensitive()) {
4284 #ifndef PRODUCT
4285 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4286 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4287 }
4288 #endif
4289 return false; // bail-out; let JVM_GetCallerClass do the work
4290 }
4291 break;
4292 default:
4293 if (!m->is_ignored_by_security_stack_walk()) {
4294 // We have reached the desired frame; return the holder class.
4295 // Acquire method holder as java.lang.Class and push as constant.
4296 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4297 ciInstance* caller_mirror = caller_klass->java_mirror();
4298 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4299
4300 #ifndef PRODUCT
4301 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4302 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());
4303 tty->print_cr(" JVM state at this point:");
4304 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4305 ciMethod* m = jvms()->of_depth(i)->method();
4306 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4307 }
4308 }
4309 #endif
4310 return true;
4311 }
4312 break;
4313 }
4314 }
4315
4316 #ifndef PRODUCT
4317 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4318 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4319 tty->print_cr(" JVM state at this point:");
4320 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4321 ciMethod* m = jvms()->of_depth(i)->method();
4322 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4323 }
4324 }
4325 #endif
4326
4327 return false; // bail-out; let JVM_GetCallerClass do the work
4328 }
4329
4330 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4331 Node* arg = argument(0);
4332 Node* result = NULL;
4333
4334 switch (id) {
4335 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4336 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4337 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4338 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4339
4340 case vmIntrinsics::_doubleToLongBits: {
4341 // two paths (plus control) merge in a wood
4342 RegionNode *r = new RegionNode(3);
4343 Node *phi = new PhiNode(r, TypeLong::LONG);
4344
4345 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4346 // Build the boolean node
4347 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4348
4349 // Branch either way.
4350 // NaN case is less traveled, which makes all the difference.
4351 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4352 Node *opt_isnan = _gvn.transform(ifisnan);
4353 assert( opt_isnan->is_If(), "Expect an IfNode");
4354 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4355 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4356
4357 set_control(iftrue);
4358
4359 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4360 Node *slow_result = longcon(nan_bits); // return NaN
4361 phi->init_req(1, _gvn.transform( slow_result ));
4362 r->init_req(1, iftrue);
4363
4364 // Else fall through
4365 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4366 set_control(iffalse);
4367
4368 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4369 r->init_req(2, iffalse);
4370
4371 // Post merge
4372 set_control(_gvn.transform(r));
4373 record_for_igvn(r);
4374
4375 C->set_has_split_ifs(true); // Has chance for split-if optimization
4376 result = phi;
4377 assert(result->bottom_type()->isa_long(), "must be");
4378 break;
4379 }
4380
4381 case vmIntrinsics::_floatToIntBits: {
4382 // two paths (plus control) merge in a wood
4383 RegionNode *r = new RegionNode(3);
4384 Node *phi = new PhiNode(r, TypeInt::INT);
4385
4386 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4387 // Build the boolean node
4388 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4389
4390 // Branch either way.
4391 // NaN case is less traveled, which makes all the difference.
4392 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4393 Node *opt_isnan = _gvn.transform(ifisnan);
4394 assert( opt_isnan->is_If(), "Expect an IfNode");
4395 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4396 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4397
4398 set_control(iftrue);
4399
4400 static const jint nan_bits = 0x7fc00000;
4401 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4402 phi->init_req(1, _gvn.transform( slow_result ));
4403 r->init_req(1, iftrue);
4404
4405 // Else fall through
4406 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4407 set_control(iffalse);
4408
4409 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4410 r->init_req(2, iffalse);
4411
4412 // Post merge
4413 set_control(_gvn.transform(r));
4414 record_for_igvn(r);
4415
4416 C->set_has_split_ifs(true); // Has chance for split-if optimization
4417 result = phi;
4418 assert(result->bottom_type()->isa_int(), "must be");
4419 break;
4420 }
4421
4422 default:
4423 fatal_unexpected_iid(id);
4424 break;
4425 }
4426 set_result(_gvn.transform(result));
4427 return true;
4428 }
4429
4430 //----------------------inline_unsafe_copyMemory-------------------------
4431 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4432
4433 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4434 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4435 const Type* base_t = gvn.type(base);
4436
4437 bool in_native = (base_t == TypePtr::NULL_PTR);
4438 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4439 bool is_mixed = !in_heap && !in_native;
4440
4441 if (is_mixed) {
4442 return true; // mixed accesses can touch both on-heap and off-heap memory
4443 }
4444 if (in_heap) {
4445 bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4446 if (!is_prim_array) {
4447 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4448 // there's not enough type information available to determine proper memory slice for it.
4449 return true;
4450 }
4451 }
4452 return false;
4453 }
4454
4455 bool LibraryCallKit::inline_unsafe_copyMemory() {
4456 if (callee()->is_static()) return false; // caller must have the capability!
4457 null_check_receiver(); // null-check receiver
4458 if (stopped()) return true;
4459
4460 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4461
4462 Node* src_base = argument(1); // type: oop
4463 Node* src_off = ConvL2X(argument(2)); // type: long
4464 Node* dst_base = argument(4); // type: oop
4465 Node* dst_off = ConvL2X(argument(5)); // type: long
4466 Node* size = ConvL2X(argument(7)); // type: long
4467
4468 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4469 "fieldOffset must be byte-scaled");
4470
4471 Node* src_addr = make_unsafe_address(src_base, src_off);
4472 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4473
4474 Node* thread = _gvn.transform(new ThreadLocalNode());
4475 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4476 BasicType doing_unsafe_access_bt = T_BYTE;
4477 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4478
4479 // update volatile field
4480 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4481
4482 int flags = RC_LEAF | RC_NO_FP;
4483
4484 const TypePtr* dst_type = TypePtr::BOTTOM;
4485
4486 // Adjust memory effects of the runtime call based on input values.
4487 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4488 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4489 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4490
4491 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4492 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4493 flags |= RC_NARROW_MEM; // narrow in memory
4494 }
4495 }
4496
4497 // Call it. Note that the length argument is not scaled.
4498 make_runtime_call(flags,
4499 OptoRuntime::fast_arraycopy_Type(),
4500 StubRoutines::unsafe_arraycopy(),
4501 "unsafe_arraycopy",
4502 dst_type,
4503 src_addr, dst_addr, size XTOP);
4504
4505 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4506
4507 return true;
4508 }
4509
4510 #undef XTOP
4511
4512 //------------------------clone_coping-----------------------------------
4513 // Helper function for inline_native_clone.
4514 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4515 assert(obj_size != NULL, "");
4516 Node* raw_obj = alloc_obj->in(1);
4517 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4518
4519 AllocateNode* alloc = NULL;
4520 if (ReduceBulkZeroing) {
4521 // We will be completely responsible for initializing this object -
4522 // mark Initialize node as complete.
4523 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4524 // The object was just allocated - there should be no any stores!
4525 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4526 // Mark as complete_with_arraycopy so that on AllocateNode
4527 // expansion, we know this AllocateNode is initialized by an array
4528 // copy and a StoreStore barrier exists after the array copy.
4529 alloc->initialization()->set_complete_with_arraycopy();
4530 }
4531
4532 Node* size = _gvn.transform(obj_size);
4533 access_clone(obj, alloc_obj, size, is_array);
4534
4535 // Do not let reads from the cloned object float above the arraycopy.
4536 if (alloc != NULL) {
4537 // Do not let stores that initialize this object be reordered with
4538 // a subsequent store that would make this object accessible by
4539 // other threads.
4540 // Record what AllocateNode this StoreStore protects so that
4541 // escape analysis can go from the MemBarStoreStoreNode to the
4542 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4543 // based on the escape status of the AllocateNode.
4544 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4545 } else {
4546 insert_mem_bar(Op_MemBarCPUOrder);
4547 }
4548 }
4549
4550 //------------------------inline_native_clone----------------------------
4551 // protected native Object java.lang.Object.clone();
4552 //
4553 // Here are the simple edge cases:
4554 // null receiver => normal trap
4555 // virtual and clone was overridden => slow path to out-of-line clone
4556 // not cloneable or finalizer => slow path to out-of-line Object.clone
4557 //
4558 // The general case has two steps, allocation and copying.
4559 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4560 //
4561 // Copying also has two cases, oop arrays and everything else.
4562 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4563 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4564 //
4565 // These steps fold up nicely if and when the cloned object's klass
4566 // can be sharply typed as an object array, a type array, or an instance.
4567 //
4568 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4569 PhiNode* result_val;
4570
4571 // Set the reexecute bit for the interpreter to reexecute
4572 // the bytecode that invokes Object.clone if deoptimization happens.
4573 { PreserveReexecuteState preexecs(this);
4574 jvms()->set_should_reexecute(true);
4575
4576 Node* obj = argument(0);
4577 if (obj->is_InlineType()) {
4578 return false;
4579 }
4580
4581 obj = null_check_receiver();
4582 if (stopped()) return true;
4583
4584 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4585
4586 // If we are going to clone an instance, we need its exact type to
4587 // know the number and types of fields to convert the clone to
4588 // loads/stores. Maybe a speculative type can help us.
4589 if (!obj_type->klass_is_exact() &&
4590 obj_type->speculative_type() != NULL &&
4591 obj_type->speculative_type()->is_instance_klass() &&
4592 !obj_type->speculative_type()->is_inlinetype()) {
4593 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4594 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4595 !spec_ik->has_injected_fields()) {
4596 ciKlass* k = obj_type->klass();
4597 if (!k->is_instance_klass() ||
4598 k->as_instance_klass()->is_interface() ||
4599 k->as_instance_klass()->has_subklass()) {
4600 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4601 }
4602 }
4603 }
4604
4605 // Conservatively insert a memory barrier on all memory slices.
4606 // Do not let writes into the original float below the clone.
4607 insert_mem_bar(Op_MemBarCPUOrder);
4608
4609 // paths into result_reg:
4610 enum {
4611 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4612 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4613 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4614 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4615 PATH_LIMIT
4616 };
4617 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4618 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4619 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4620 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4621 record_for_igvn(result_reg);
4622
4623 Node* obj_klass = load_object_klass(obj);
4624 // We only go to the fast case code if we pass a number of guards.
4625 // The paths which do not pass are accumulated in the slow_region.
4626 RegionNode* slow_region = new RegionNode(1);
4627 record_for_igvn(slow_region);
4628
4629 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4630 if (array_ctl != NULL) {
4631 // It's an array.
4632 PreserveJVMState pjvms(this);
4633 set_control(array_ctl);
4634
4635 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4636 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
4637 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
4638 obj_type->klass()->can_be_inline_array_klass() &&
4639 (ary_ptr == NULL || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
4640 // Flattened inline type array may have object field that would require a
4641 // write barrier. Conservatively, go to slow path.
4642 generate_flatArray_guard(obj_klass, slow_region);
4643 }
4644
4645 if (!stopped()) {
4646 Node* obj_length = load_array_length(obj);
4647 Node* obj_size = NULL;
4648 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4649
4650 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4651 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4652 // If it is an oop array, it requires very special treatment,
4653 // because gc barriers are required when accessing the array.
4654 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4655 if (is_obja != NULL) {
4656 PreserveJVMState pjvms2(this);
4657 set_control(is_obja);
4658 // Generate a direct call to the right arraycopy function(s).
4659 // Clones are always tightly coupled.
4660 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4661 ac->set_clone_oop_array();
4662 Node* n = _gvn.transform(ac);
4663 assert(n == ac, "cannot disappear");
4664 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4665
4666 result_reg->init_req(_objArray_path, control());
4667 result_val->init_req(_objArray_path, alloc_obj);
4668 result_i_o ->set_req(_objArray_path, i_o());
4669 result_mem ->set_req(_objArray_path, reset_memory());
4670 }
4671 }
4672 // Otherwise, there are no barriers to worry about.
4673 // (We can dispense with card marks if we know the allocation
4674 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4675 // causes the non-eden paths to take compensating steps to
4676 // simulate a fresh allocation, so that no further
4677 // card marks are required in compiled code to initialize
4678 // the object.)
4679
4680 if (!stopped()) {
4681 copy_to_clone(obj, alloc_obj, obj_size, true);
4682
4683 // Present the results of the copy.
4684 result_reg->init_req(_array_path, control());
4685 result_val->init_req(_array_path, alloc_obj);
4686 result_i_o ->set_req(_array_path, i_o());
4687 result_mem ->set_req(_array_path, reset_memory());
4688 }
4689 }
4690 }
4691
4692 if (!stopped()) {
4693 // It's an instance (we did array above). Make the slow-path tests.
4694 // If this is a virtual call, we generate a funny guard. We grab
4695 // the vtable entry corresponding to clone() from the target object.
4696 // If the target method which we are calling happens to be the
4697 // Object clone() method, we pass the guard. We do not need this
4698 // guard for non-virtual calls; the caller is known to be the native
4699 // Object clone().
4700 if (is_virtual) {
4701 generate_virtual_guard(obj_klass, slow_region);
4702 }
4703
4704 // The object must be easily cloneable and must not have a finalizer.
4705 // Both of these conditions may be checked in a single test.
4706 // We could optimize the test further, but we don't care.
4707 generate_access_flags_guard(obj_klass,
4708 // Test both conditions:
4709 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4710 // Must be cloneable but not finalizer:
4711 JVM_ACC_IS_CLONEABLE_FAST,
4712 slow_region);
4713 }
4714
4715 if (!stopped()) {
4716 // It's an instance, and it passed the slow-path tests.
4717 PreserveJVMState pjvms(this);
4718 Node* obj_size = NULL;
4719 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4720 // is reexecuted if deoptimization occurs and there could be problems when merging
4721 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4722 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4723
4724 copy_to_clone(obj, alloc_obj, obj_size, false);
4725
4726 // Present the results of the slow call.
4727 result_reg->init_req(_instance_path, control());
4728 result_val->init_req(_instance_path, alloc_obj);
4729 result_i_o ->set_req(_instance_path, i_o());
4730 result_mem ->set_req(_instance_path, reset_memory());
4731 }
4732
4733 // Generate code for the slow case. We make a call to clone().
4734 set_control(_gvn.transform(slow_region));
4735 if (!stopped()) {
4736 PreserveJVMState pjvms(this);
4737 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4738 // We need to deoptimize on exception (see comment above)
4739 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4740 // this->control() comes from set_results_for_java_call
4741 result_reg->init_req(_slow_path, control());
4742 result_val->init_req(_slow_path, slow_result);
4743 result_i_o ->set_req(_slow_path, i_o());
4744 result_mem ->set_req(_slow_path, reset_memory());
4745 }
4746
4747 // Return the combined state.
4748 set_control( _gvn.transform(result_reg));
4749 set_i_o( _gvn.transform(result_i_o));
4750 set_all_memory( _gvn.transform(result_mem));
4751 } // original reexecute is set back here
4752
4753 set_result(_gvn.transform(result_val));
4754 return true;
4755 }
4756
4757 // If we have a tightly coupled allocation, the arraycopy may take care
4758 // of the array initialization. If one of the guards we insert between
4759 // the allocation and the arraycopy causes a deoptimization, an
4760 // unitialized array will escape the compiled method. To prevent that
4761 // we set the JVM state for uncommon traps between the allocation and
4762 // the arraycopy to the state before the allocation so, in case of
4763 // deoptimization, we'll reexecute the allocation and the
4764 // initialization.
4765 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4766 if (alloc != NULL) {
4767 ciMethod* trap_method = alloc->jvms()->method();
4768 int trap_bci = alloc->jvms()->bci();
4769
4770 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4771 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4772 // Make sure there's no store between the allocation and the
4773 // arraycopy otherwise visible side effects could be rexecuted
4774 // in case of deoptimization and cause incorrect execution.
4775 bool no_interfering_store = true;
4776 Node* mem = alloc->in(TypeFunc::Memory);
4777 if (mem->is_MergeMem()) {
4778 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4779 Node* n = mms.memory();
4780 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4781 assert(n->is_Store(), "what else?");
4782 no_interfering_store = false;
4783 break;
4784 }
4785 }
4786 } else {
4787 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4788 Node* n = mms.memory();
4789 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4790 assert(n->is_Store(), "what else?");
4791 no_interfering_store = false;
4792 break;
4793 }
4794 }
4795 }
4796
4797 if (no_interfering_store) {
4798 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4799 uint size = alloc->req();
4800 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4801 old_jvms->set_map(sfpt);
4802 for (uint i = 0; i < size; i++) {
4803 sfpt->init_req(i, alloc->in(i));
4804 }
4805 // re-push array length for deoptimization
4806 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4807 old_jvms->set_sp(old_jvms->sp()+1);
4808 old_jvms->set_monoff(old_jvms->monoff()+1);
4809 old_jvms->set_scloff(old_jvms->scloff()+1);
4810 old_jvms->set_endoff(old_jvms->endoff()+1);
4811 old_jvms->set_should_reexecute(true);
4812
4813 sfpt->set_i_o(map()->i_o());
4814 sfpt->set_memory(map()->memory());
4815 sfpt->set_control(map()->control());
4816
4817 JVMState* saved_jvms = jvms();
4818 saved_reexecute_sp = _reexecute_sp;
4819
4820 set_jvms(sfpt->jvms());
4821 _reexecute_sp = jvms()->sp();
4822
4823 return saved_jvms;
4824 }
4825 }
4826 }
4827 return NULL;
4828 }
4829
4830 // In case of a deoptimization, we restart execution at the
4831 // allocation, allocating a new array. We would leave an uninitialized
4832 // array in the heap that GCs wouldn't expect. Move the allocation
4833 // after the traps so we don't allocate the array if we
4834 // deoptimize. This is possible because tightly_coupled_allocation()
4835 // guarantees there's no observer of the allocated array at this point
4836 // and the control flow is simple enough.
4837 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4838 int saved_reexecute_sp, uint new_idx) {
4839 if (saved_jvms != NULL && !stopped()) {
4840 assert(alloc != NULL, "only with a tightly coupled allocation");
4841 // restore JVM state to the state at the arraycopy
4842 saved_jvms->map()->set_control(map()->control());
4843 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4844 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4845 // If we've improved the types of some nodes (null check) while
4846 // emitting the guards, propagate them to the current state
4847 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4848 set_jvms(saved_jvms);
4849 _reexecute_sp = saved_reexecute_sp;
4850
4851 // Remove the allocation from above the guards
4852 CallProjections* callprojs = alloc->extract_projections(true);
4853 InitializeNode* init = alloc->initialization();
4854 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4855 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4856 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4857
4858 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4859 // the allocation (i.e. is only valid if the allocation succeeds):
4860 // 1) replace CastIINode with AllocateArrayNode's length here
4861 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4862 //
4863 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4864 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4865 Node* init_control = init->proj_out(TypeFunc::Control);
4866 Node* alloc_length = alloc->Ideal_length();
4867 #ifdef ASSERT
4868 Node* prev_cast = NULL;
4869 #endif
4870 for (uint i = 0; i < init_control->outcnt(); i++) {
4871 Node* init_out = init_control->raw_out(i);
4872 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4873 #ifdef ASSERT
4874 if (prev_cast == NULL) {
4875 prev_cast = init_out;
4876 } else {
4877 if (prev_cast->cmp(*init_out) == false) {
4878 prev_cast->dump();
4879 init_out->dump();
4880 assert(false, "not equal CastIINode");
4881 }
4882 }
4883 #endif
4884 C->gvn_replace_by(init_out, alloc_length);
4885 }
4886 }
4887 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4888
4889 // move the allocation here (after the guards)
4890 _gvn.hash_delete(alloc);
4891 alloc->set_req(TypeFunc::Control, control());
4892 alloc->set_req(TypeFunc::I_O, i_o());
4893 Node *mem = reset_memory();
4894 set_all_memory(mem);
4895 alloc->set_req(TypeFunc::Memory, mem);
4896 set_control(init->proj_out_or_null(TypeFunc::Control));
4897 set_i_o(callprojs->fallthrough_ioproj);
4898
4899 // Update memory as done in GraphKit::set_output_for_allocation()
4900 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4901 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4902 if (ary_type->isa_aryptr() && length_type != NULL) {
4903 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4904 }
4905 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4906 int elemidx = C->get_alias_index(telemref);
4907 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4908 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4909
4910 Node* allocx = _gvn.transform(alloc);
4911 assert(allocx == alloc, "where has the allocation gone?");
4912 assert(dest->is_CheckCastPP(), "not an allocation result?");
4913
4914 _gvn.hash_delete(dest);
4915 dest->set_req(0, control());
4916 Node* destx = _gvn.transform(dest);
4917 assert(destx == dest, "where has the allocation result gone?");
4918
4919 array_ideal_length(alloc, ary_type, true);
4920 }
4921 }
4922
4923
4924 //------------------------------inline_arraycopy-----------------------
4925 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4926 // Object dest, int destPos,
4927 // int length);
4928 bool LibraryCallKit::inline_arraycopy() {
4929 // Get the arguments.
4930 Node* src = argument(0); // type: oop
4931 Node* src_offset = argument(1); // type: int
4932 Node* dest = argument(2); // type: oop
4933 Node* dest_offset = argument(3); // type: int
4934 Node* length = argument(4); // type: int
4935
4936 uint new_idx = C->unique();
4937
4938 // Check for allocation before we add nodes that would confuse
4939 // tightly_coupled_allocation()
4940 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
4941
4942 int saved_reexecute_sp = -1;
4943 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4944 // See arraycopy_restore_alloc_state() comment
4945 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4946 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4947 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4948 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4949
4950 // The following tests must be performed
4951 // (1) src and dest are arrays.
4952 // (2) src and dest arrays must have elements of the same BasicType
4953 // (3) src and dest must not be null.
4954 // (4) src_offset must not be negative.
4955 // (5) dest_offset must not be negative.
4956 // (6) length must not be negative.
4957 // (7) src_offset + length must not exceed length of src.
4958 // (8) dest_offset + length must not exceed length of dest.
4959 // (9) each element of an oop array must be assignable
4960
4961 // (3) src and dest must not be null.
4962 // always do this here because we need the JVM state for uncommon traps
4963 Node* null_ctl = top();
4964 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4965 assert(null_ctl->is_top(), "no null control here");
4966 dest = null_check(dest, T_ARRAY);
4967
4968 if (!can_emit_guards) {
4969 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4970 // guards but the arraycopy node could still take advantage of a
4971 // tightly allocated allocation. tightly_coupled_allocation() is
4972 // called again to make sure it takes the null check above into
4973 // account: the null check is mandatory and if it caused an
4974 // uncommon trap to be emitted then the allocation can't be
4975 // considered tightly coupled in this context.
4976 alloc = tightly_coupled_allocation(dest);
4977 }
4978
4979 bool validated = false;
4980
4981 const Type* src_type = _gvn.type(src);
4982 const Type* dest_type = _gvn.type(dest);
4983 const TypeAryPtr* top_src = src_type->isa_aryptr();
4984 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4985
4986 // Do we have the type of src?
4987 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4988 // Do we have the type of dest?
4989 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4990 // Is the type for src from speculation?
4991 bool src_spec = false;
4992 // Is the type for dest from speculation?
4993 bool dest_spec = false;
4994
4995 if ((!has_src || !has_dest) && can_emit_guards) {
4996 // We don't have sufficient type information, let's see if
4997 // speculative types can help. We need to have types for both src
4998 // and dest so that it pays off.
4999
5000 // Do we already have or could we have type information for src
5001 bool could_have_src = has_src;
5002 // Do we already have or could we have type information for dest
5003 bool could_have_dest = has_dest;
5004
5005 ciKlass* src_k = NULL;
5006 if (!has_src) {
5007 src_k = src_type->speculative_type_not_null();
5008 if (src_k != NULL && src_k->is_array_klass()) {
5009 could_have_src = true;
5010 }
5011 }
5012
5013 ciKlass* dest_k = NULL;
5014 if (!has_dest) {
5015 dest_k = dest_type->speculative_type_not_null();
5016 if (dest_k != NULL && dest_k->is_array_klass()) {
5017 could_have_dest = true;
5018 }
5019 }
5020
5021 if (could_have_src && could_have_dest) {
5022 // This is going to pay off so emit the required guards
5023 if (!has_src) {
5024 src = maybe_cast_profiled_obj(src, src_k, true);
5025 src_type = _gvn.type(src);
5026 top_src = src_type->isa_aryptr();
5027 has_src = (top_src != NULL && top_src->klass() != NULL);
5028 src_spec = true;
5029 }
5030 if (!has_dest) {
5031 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5032 dest_type = _gvn.type(dest);
5033 top_dest = dest_type->isa_aryptr();
5034 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5035 dest_spec = true;
5036 }
5037 }
5038 }
5039
5040 if (has_src && has_dest && can_emit_guards) {
5041 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5042 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5043 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
5044 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
5045
5046 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5047 // If both arrays are object arrays then having the exact types
5048 // for both will remove the need for a subtype check at runtime
5049 // before the call and may make it possible to pick a faster copy
5050 // routine (without a subtype check on every element)
5051 // Do we have the exact type of src?
5052 bool could_have_src = src_spec;
5053 // Do we have the exact type of dest?
5054 bool could_have_dest = dest_spec;
5055 ciKlass* src_k = top_src->klass();
5056 ciKlass* dest_k = top_dest->klass();
5057 if (!src_spec) {
5058 src_k = src_type->speculative_type_not_null();
5059 if (src_k != NULL && src_k->is_array_klass()) {
5060 could_have_src = true;
5061 }
5062 }
5063 if (!dest_spec) {
5064 dest_k = dest_type->speculative_type_not_null();
5065 if (dest_k != NULL && dest_k->is_array_klass()) {
5066 could_have_dest = true;
5067 }
5068 }
5069 if (could_have_src && could_have_dest) {
5070 // If we can have both exact types, emit the missing guards
5071 if (could_have_src && !src_spec) {
5072 src = maybe_cast_profiled_obj(src, src_k, true);
5073 src_type = _gvn.type(src);
5074 top_src = src_type->isa_aryptr();
5075 }
5076 if (could_have_dest && !dest_spec) {
5077 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5078 dest_type = _gvn.type(dest);
5079 top_dest = dest_type->isa_aryptr();
5080 }
5081 }
5082 }
5083 }
5084
5085 ciMethod* trap_method = method();
5086 int trap_bci = bci();
5087 if (saved_jvms != NULL) {
5088 trap_method = alloc->jvms()->method();
5089 trap_bci = alloc->jvms()->bci();
5090 }
5091
5092 bool negative_length_guard_generated = false;
5093
5094 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5095 can_emit_guards && !src->is_top() && !dest->is_top()) {
5096 // validate arguments: enables transformation the ArrayCopyNode
5097 validated = true;
5098
5099 RegionNode* slow_region = new RegionNode(1);
5100 record_for_igvn(slow_region);
5101
5102 // (1) src and dest are arrays.
5103 generate_non_array_guard(load_object_klass(src), slow_region);
5104 generate_non_array_guard(load_object_klass(dest), slow_region);
5105
5106 // (2) src and dest arrays must have elements of the same BasicType
5107 // done at macro expansion or at Ideal transformation time
5108
5109 // (4) src_offset must not be negative.
5110 generate_negative_guard(src_offset, slow_region);
5111
5112 // (5) dest_offset must not be negative.
5113 generate_negative_guard(dest_offset, slow_region);
5114
5115 // (7) src_offset + length must not exceed length of src.
5116 generate_limit_guard(src_offset, length,
5117 load_array_length(src),
5118 slow_region);
5119
5120 // (8) dest_offset + length must not exceed length of dest.
5121 generate_limit_guard(dest_offset, length,
5122 load_array_length(dest),
5123 slow_region);
5124
5125 // (6) length must not be negative.
5126 // This is also checked in generate_arraycopy() during macro expansion, but
5127 // we also have to check it here for the case where the ArrayCopyNode will
5128 // be eliminated by Escape Analysis.
5129 if (EliminateAllocations) {
5130 generate_negative_guard(length, slow_region);
5131 negative_length_guard_generated = true;
5132 }
5133
5134 // (9) each element of an oop array must be assignable
5135 Node* dest_klass = load_object_klass(dest);
5136 if (src != dest) {
5137 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5138 slow_region->add_req(not_subtype_ctrl);
5139 }
5140
5141 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5142 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
5143 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5144 src_type = _gvn.type(src);
5145 top_src = src_type->isa_aryptr();
5146
5147 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
5148 if (!stopped() && UseFlatArray) {
5149 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
5150 assert(top_dest == NULL || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
5151 if (top_src != NULL && top_src->is_flat()) {
5152 // Src is flat, check that dest is flat as well
5153 if (top_dest != NULL && !top_dest->is_flat()) {
5154 generate_non_flatArray_guard(dest_klass, slow_region);
5155 // Since dest is flat and src <: dest, dest must have the same type as src.
5156 top_dest = TypeOopPtr::make_from_klass(top_src->klass())->isa_aryptr();
5157 assert(top_dest->is_flat(), "dest must be flat");
5158 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
5159 }
5160 } else if (top_src == NULL || !top_src->is_not_flat()) {
5161 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5162 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5163 assert(top_dest == NULL || !top_dest->is_flat(), "dest array must not be flat");
5164 generate_flatArray_guard(load_object_klass(src), slow_region);
5165 if (top_src != NULL) {
5166 top_src = top_src->cast_to_not_flat();
5167 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
5168 }
5169 }
5170 }
5171
5172 {
5173 PreserveJVMState pjvms(this);
5174 set_control(_gvn.transform(slow_region));
5175 uncommon_trap(Deoptimization::Reason_intrinsic,
5176 Deoptimization::Action_make_not_entrant);
5177 assert(stopped(), "Should be stopped");
5178 }
5179 }
5180
5181 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5182
5183 if (stopped()) {
5184 return true;
5185 }
5186
5187 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5188 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5189 // so the compiler has a chance to eliminate them: during macro expansion,
5190 // we have to set their control (CastPP nodes are eliminated).
5191 load_object_klass(src), load_object_klass(dest),
5192 load_array_length(src), load_array_length(dest));
5193
5194 ac->set_arraycopy(validated);
5195
5196 Node* n = _gvn.transform(ac);
5197 if (n == ac) {
5198 ac->connect_outputs(this);
5199 } else {
5200 assert(validated, "shouldn't transform if all arguments not validated");
5201 set_all_memory(n);
5202 }
5203 clear_upper_avx();
5204
5205
5206 return true;
5207 }
5208
5209
5210 // Helper function which determines if an arraycopy immediately follows
5211 // an allocation, with no intervening tests or other escapes for the object.
5212 AllocateArrayNode*
5213 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5214 if (stopped()) return NULL; // no fast path
5215 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5216
5217 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5218 if (alloc == NULL) return NULL;
5219
5220 Node* rawmem = memory(Compile::AliasIdxRaw);
5221 // Is the allocation's memory state untouched?
5222 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5223 // Bail out if there have been raw-memory effects since the allocation.
5224 // (Example: There might have been a call or safepoint.)
5225 return NULL;
5226 }
5227 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5228 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5229 return NULL;
5230 }
5231
5232 // There must be no unexpected observers of this allocation.
5233 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5234 Node* obs = ptr->fast_out(i);
5235 if (obs != this->map()) {
5236 return NULL;
5237 }
5238 }
5239
5240 // This arraycopy must unconditionally follow the allocation of the ptr.
5241 Node* alloc_ctl = ptr->in(0);
5242 Node* ctl = control();
5243 while (ctl != alloc_ctl) {
5244 // There may be guards which feed into the slow_region.
5245 // Any other control flow means that we might not get a chance
5246 // to finish initializing the allocated object.
5247 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5248 IfNode* iff = ctl->in(0)->as_If();
5249 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5250 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5251 // One more try: Various low-level checks bottom out in
5252 // uncommon traps. If the debug-info of the trap omits
5253 // any reference to the allocation, as we've already
5254 // observed, then there can be no objection to the trap.
5255 bool found_trap = false;
5256 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5257 Node* obs = not_ctl->fast_out(j);
5258 if (obs->in(0) == not_ctl && obs->is_Call() &&
5259 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5260 found_trap = true; break;
5261 }
5262 }
5263 if (found_trap) {
5264 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5265 continue;
5266 }
5267 }
5268 return NULL;
5269 }
5270
5271 // If we get this far, we have an allocation which immediately
5272 // precedes the arraycopy, and we can take over zeroing the new object.
5273 // The arraycopy will finish the initialization, and provide
5274 // a new control state to which we will anchor the destination pointer.
5275
5276 return alloc;
5277 }
5278
5279 //-------------inline_encodeISOArray-----------------------------------
5280 // encode char[] to byte[] in ISO_8859_1 or ASCII
5281 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5282 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5283 // no receiver since it is static method
5284 Node *src = argument(0);
5285 Node *src_offset = argument(1);
5286 Node *dst = argument(2);
5287 Node *dst_offset = argument(3);
5288 Node *length = argument(4);
5289
5290 src = must_be_not_null(src, true);
5291 dst = must_be_not_null(dst, true);
5292
5293 const Type* src_type = src->Value(&_gvn);
5294 const Type* dst_type = dst->Value(&_gvn);
5295 const TypeAryPtr* top_src = src_type->isa_aryptr();
5296 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5297 if (top_src == NULL || top_src->klass() == NULL ||
5298 top_dest == NULL || top_dest->klass() == NULL) {
5299 // failed array check
5300 return false;
5301 }
5302
5303 // Figure out the size and type of the elements we will be copying.
5304 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5305 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5306 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5307 return false;
5308 }
5309
5310 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5311 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5312 // 'src_start' points to src array + scaled offset
5313 // 'dst_start' points to dst array + scaled offset
5314
5315 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5316 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5317 enc = _gvn.transform(enc);
5318 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5319 set_memory(res_mem, mtype);
5320 set_result(enc);
5321 clear_upper_avx();
5322
5323 return true;
5324 }
5325
5326 //-------------inline_multiplyToLen-----------------------------------
5327 bool LibraryCallKit::inline_multiplyToLen() {
5328 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5329
5330 address stubAddr = StubRoutines::multiplyToLen();
5331 if (stubAddr == NULL) {
5332 return false; // Intrinsic's stub is not implemented on this platform
5333 }
5334 const char* stubName = "multiplyToLen";
5335
5336 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5337
5338 // no receiver because it is a static method
5339 Node* x = argument(0);
5340 Node* xlen = argument(1);
5341 Node* y = argument(2);
5342 Node* ylen = argument(3);
5343 Node* z = argument(4);
5344
5345 x = must_be_not_null(x, true);
5346 y = must_be_not_null(y, true);
5347
5348 const Type* x_type = x->Value(&_gvn);
5349 const Type* y_type = y->Value(&_gvn);
5350 const TypeAryPtr* top_x = x_type->isa_aryptr();
5351 const TypeAryPtr* top_y = y_type->isa_aryptr();
5352 if (top_x == NULL || top_x->klass() == NULL ||
5353 top_y == NULL || top_y->klass() == NULL) {
5354 // failed array check
5355 return false;
5356 }
5357
5358 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5359 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5360 if (x_elem != T_INT || y_elem != T_INT) {
5361 return false;
5362 }
5363
5364 // Set the original stack and the reexecute bit for the interpreter to reexecute
5365 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5366 // on the return from z array allocation in runtime.
5367 { PreserveReexecuteState preexecs(this);
5368 jvms()->set_should_reexecute(true);
5369
5370 Node* x_start = array_element_address(x, intcon(0), x_elem);
5371 Node* y_start = array_element_address(y, intcon(0), y_elem);
5372 // 'x_start' points to x array + scaled xlen
5373 // 'y_start' points to y array + scaled ylen
5374
5375 // Allocate the result array
5376 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5377 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5378 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5379
5380 IdealKit ideal(this);
5381
5382 #define __ ideal.
5383 Node* one = __ ConI(1);
5384 Node* zero = __ ConI(0);
5385 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5386 __ set(need_alloc, zero);
5387 __ set(z_alloc, z);
5388 __ if_then(z, BoolTest::eq, null()); {
5389 __ increment (need_alloc, one);
5390 } __ else_(); {
5391 // Update graphKit memory and control from IdealKit.
5392 sync_kit(ideal);
5393 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5394 cast->init_req(0, control());
5395 _gvn.set_type(cast, cast->bottom_type());
5396 C->record_for_igvn(cast);
5397
5398 Node* zlen_arg = load_array_length(cast);
5399 // Update IdealKit memory and control from graphKit.
5400 __ sync_kit(this);
5401 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5402 __ increment (need_alloc, one);
5403 } __ end_if();
5404 } __ end_if();
5405
5406 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5407 // Update graphKit memory and control from IdealKit.
5408 sync_kit(ideal);
5409 Node * narr = new_array(klass_node, zlen, 1);
5410 // Update IdealKit memory and control from graphKit.
5411 __ sync_kit(this);
5412 __ set(z_alloc, narr);
5413 } __ end_if();
5414
5415 sync_kit(ideal);
5416 z = __ value(z_alloc);
5417 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5418 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5419 // Final sync IdealKit and GraphKit.
5420 final_sync(ideal);
5421 #undef __
5422
5423 Node* z_start = array_element_address(z, intcon(0), T_INT);
5424
5425 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5426 OptoRuntime::multiplyToLen_Type(),
5427 stubAddr, stubName, TypePtr::BOTTOM,
5428 x_start, xlen, y_start, ylen, z_start, zlen);
5429 } // original reexecute is set back here
5430
5431 C->set_has_split_ifs(true); // Has chance for split-if optimization
5432 set_result(z);
5433 return true;
5434 }
5435
5436 //-------------inline_squareToLen------------------------------------
5437 bool LibraryCallKit::inline_squareToLen() {
5438 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5439
5440 address stubAddr = StubRoutines::squareToLen();
5441 if (stubAddr == NULL) {
5442 return false; // Intrinsic's stub is not implemented on this platform
5443 }
5444 const char* stubName = "squareToLen";
5445
5446 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5447
5448 Node* x = argument(0);
5449 Node* len = argument(1);
5450 Node* z = argument(2);
5451 Node* zlen = argument(3);
5452
5453 x = must_be_not_null(x, true);
5454 z = must_be_not_null(z, true);
5455
5456 const Type* x_type = x->Value(&_gvn);
5457 const Type* z_type = z->Value(&_gvn);
5458 const TypeAryPtr* top_x = x_type->isa_aryptr();
5459 const TypeAryPtr* top_z = z_type->isa_aryptr();
5460 if (top_x == NULL || top_x->klass() == NULL ||
5461 top_z == NULL || top_z->klass() == NULL) {
5462 // failed array check
5463 return false;
5464 }
5465
5466 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5467 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5468 if (x_elem != T_INT || z_elem != T_INT) {
5469 return false;
5470 }
5471
5472
5473 Node* x_start = array_element_address(x, intcon(0), x_elem);
5474 Node* z_start = array_element_address(z, intcon(0), z_elem);
5475
5476 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5477 OptoRuntime::squareToLen_Type(),
5478 stubAddr, stubName, TypePtr::BOTTOM,
5479 x_start, len, z_start, zlen);
5480
5481 set_result(z);
5482 return true;
5483 }
5484
5485 //-------------inline_mulAdd------------------------------------------
5486 bool LibraryCallKit::inline_mulAdd() {
5487 assert(UseMulAddIntrinsic, "not implemented on this platform");
5488
5489 address stubAddr = StubRoutines::mulAdd();
5490 if (stubAddr == NULL) {
5491 return false; // Intrinsic's stub is not implemented on this platform
5492 }
5493 const char* stubName = "mulAdd";
5494
5495 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5496
5497 Node* out = argument(0);
5498 Node* in = argument(1);
5499 Node* offset = argument(2);
5500 Node* len = argument(3);
5501 Node* k = argument(4);
5502
5503 out = must_be_not_null(out, true);
5504
5505 const Type* out_type = out->Value(&_gvn);
5506 const Type* in_type = in->Value(&_gvn);
5507 const TypeAryPtr* top_out = out_type->isa_aryptr();
5508 const TypeAryPtr* top_in = in_type->isa_aryptr();
5509 if (top_out == NULL || top_out->klass() == NULL ||
5510 top_in == NULL || top_in->klass() == NULL) {
5511 // failed array check
5512 return false;
5513 }
5514
5515 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5516 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5517 if (out_elem != T_INT || in_elem != T_INT) {
5518 return false;
5519 }
5520
5521 Node* outlen = load_array_length(out);
5522 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5523 Node* out_start = array_element_address(out, intcon(0), out_elem);
5524 Node* in_start = array_element_address(in, intcon(0), in_elem);
5525
5526 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5527 OptoRuntime::mulAdd_Type(),
5528 stubAddr, stubName, TypePtr::BOTTOM,
5529 out_start,in_start, new_offset, len, k);
5530 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5531 set_result(result);
5532 return true;
5533 }
5534
5535 //-------------inline_montgomeryMultiply-----------------------------------
5536 bool LibraryCallKit::inline_montgomeryMultiply() {
5537 address stubAddr = StubRoutines::montgomeryMultiply();
5538 if (stubAddr == NULL) {
5539 return false; // Intrinsic's stub is not implemented on this platform
5540 }
5541
5542 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5543 const char* stubName = "montgomery_multiply";
5544
5545 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5546
5547 Node* a = argument(0);
5548 Node* b = argument(1);
5549 Node* n = argument(2);
5550 Node* len = argument(3);
5551 Node* inv = argument(4);
5552 Node* m = argument(6);
5553
5554 const Type* a_type = a->Value(&_gvn);
5555 const TypeAryPtr* top_a = a_type->isa_aryptr();
5556 const Type* b_type = b->Value(&_gvn);
5557 const TypeAryPtr* top_b = b_type->isa_aryptr();
5558 const Type* n_type = a->Value(&_gvn);
5559 const TypeAryPtr* top_n = n_type->isa_aryptr();
5560 const Type* m_type = a->Value(&_gvn);
5561 const TypeAryPtr* top_m = m_type->isa_aryptr();
5562 if (top_a == NULL || top_a->klass() == NULL ||
5563 top_b == NULL || top_b->klass() == NULL ||
5564 top_n == NULL || top_n->klass() == NULL ||
5565 top_m == NULL || top_m->klass() == NULL) {
5566 // failed array check
5567 return false;
5568 }
5569
5570 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5571 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5572 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5573 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5574 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5575 return false;
5576 }
5577
5578 // Make the call
5579 {
5580 Node* a_start = array_element_address(a, intcon(0), a_elem);
5581 Node* b_start = array_element_address(b, intcon(0), b_elem);
5582 Node* n_start = array_element_address(n, intcon(0), n_elem);
5583 Node* m_start = array_element_address(m, intcon(0), m_elem);
5584
5585 Node* call = make_runtime_call(RC_LEAF,
5586 OptoRuntime::montgomeryMultiply_Type(),
5587 stubAddr, stubName, TypePtr::BOTTOM,
5588 a_start, b_start, n_start, len, inv, top(),
5589 m_start);
5590 set_result(m);
5591 }
5592
5593 return true;
5594 }
5595
5596 bool LibraryCallKit::inline_montgomerySquare() {
5597 address stubAddr = StubRoutines::montgomerySquare();
5598 if (stubAddr == NULL) {
5599 return false; // Intrinsic's stub is not implemented on this platform
5600 }
5601
5602 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5603 const char* stubName = "montgomery_square";
5604
5605 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5606
5607 Node* a = argument(0);
5608 Node* n = argument(1);
5609 Node* len = argument(2);
5610 Node* inv = argument(3);
5611 Node* m = argument(5);
5612
5613 const Type* a_type = a->Value(&_gvn);
5614 const TypeAryPtr* top_a = a_type->isa_aryptr();
5615 const Type* n_type = a->Value(&_gvn);
5616 const TypeAryPtr* top_n = n_type->isa_aryptr();
5617 const Type* m_type = a->Value(&_gvn);
5618 const TypeAryPtr* top_m = m_type->isa_aryptr();
5619 if (top_a == NULL || top_a->klass() == NULL ||
5620 top_n == NULL || top_n->klass() == NULL ||
5621 top_m == NULL || top_m->klass() == NULL) {
5622 // failed array check
5623 return false;
5624 }
5625
5626 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5627 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5628 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5629 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5630 return false;
5631 }
5632
5633 // Make the call
5634 {
5635 Node* a_start = array_element_address(a, intcon(0), a_elem);
5636 Node* n_start = array_element_address(n, intcon(0), n_elem);
5637 Node* m_start = array_element_address(m, intcon(0), m_elem);
5638
5639 Node* call = make_runtime_call(RC_LEAF,
5640 OptoRuntime::montgomerySquare_Type(),
5641 stubAddr, stubName, TypePtr::BOTTOM,
5642 a_start, n_start, len, inv, top(),
5643 m_start);
5644 set_result(m);
5645 }
5646
5647 return true;
5648 }
5649
5650 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5651 address stubAddr = NULL;
5652 const char* stubName = NULL;
5653
5654 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5655 if (stubAddr == NULL) {
5656 return false; // Intrinsic's stub is not implemented on this platform
5657 }
5658
5659 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5660
5661 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5662
5663 Node* newArr = argument(0);
5664 Node* oldArr = argument(1);
5665 Node* newIdx = argument(2);
5666 Node* shiftCount = argument(3);
5667 Node* numIter = argument(4);
5668
5669 const Type* newArr_type = newArr->Value(&_gvn);
5670 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5671 const Type* oldArr_type = oldArr->Value(&_gvn);
5672 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5673 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5674 || top_oldArr->klass() == NULL) {
5675 return false;
5676 }
5677
5678 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5679 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5680 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5681 return false;
5682 }
5683
5684 // Make the call
5685 {
5686 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5687 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5688
5689 Node* call = make_runtime_call(RC_LEAF,
5690 OptoRuntime::bigIntegerShift_Type(),
5691 stubAddr,
5692 stubName,
5693 TypePtr::BOTTOM,
5694 newArr_start,
5695 oldArr_start,
5696 newIdx,
5697 shiftCount,
5698 numIter);
5699 }
5700
5701 return true;
5702 }
5703
5704 //-------------inline_vectorizedMismatch------------------------------
5705 bool LibraryCallKit::inline_vectorizedMismatch() {
5706 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5707
5708 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5709 Node* obja = argument(0); // Object
5710 Node* aoffset = argument(1); // long
5711 Node* objb = argument(3); // Object
5712 Node* boffset = argument(4); // long
5713 Node* length = argument(6); // int
5714 Node* scale = argument(7); // int
5715
5716 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5717 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5718 if (obja_t == NULL || obja_t->klass() == NULL ||
5719 objb_t == NULL || objb_t->klass() == NULL ||
5720 scale == top()) {
5721 return false; // failed input validation
5722 }
5723
5724 Node* obja_adr = make_unsafe_address(obja, aoffset);
5725 Node* objb_adr = make_unsafe_address(objb, boffset);
5726
5727 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5728 //
5729 // inline_limit = ArrayOperationPartialInlineSize / element_size;
5730 // if (length <= inline_limit) {
5731 // inline_path:
5732 // vmask = VectorMaskGen length
5733 // vload1 = LoadVectorMasked obja, vmask
5734 // vload2 = LoadVectorMasked objb, vmask
5735 // result1 = VectorCmpMasked vload1, vload2, vmask
5736 // } else {
5737 // call_stub_path:
5738 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5739 // }
5740 // exit_block:
5741 // return Phi(result1, result2);
5742 //
5743 enum { inline_path = 1, // input is small enough to process it all at once
5744 stub_path = 2, // input is too large; call into the VM
5745 PATH_LIMIT = 3
5746 };
5747
5748 Node* exit_block = new RegionNode(PATH_LIMIT);
5749 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5750 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5751
5752 Node* call_stub_path = control();
5753
5754 BasicType elem_bt = T_ILLEGAL;
5755
5756 const TypeInt* scale_t = _gvn.type(scale)->is_int();
5757 if (scale_t->is_con()) {
5758 switch (scale_t->get_con()) {
5759 case 0: elem_bt = T_BYTE; break;
5760 case 1: elem_bt = T_SHORT; break;
5761 case 2: elem_bt = T_INT; break;
5762 case 3: elem_bt = T_LONG; break;
5763
5764 default: elem_bt = T_ILLEGAL; break; // not supported
5765 }
5766 }
5767
5768 int inline_limit = 0;
5769 bool do_partial_inline = false;
5770
5771 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5772 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5773 do_partial_inline = inline_limit >= 16;
5774 }
5775
5776 if (do_partial_inline) {
5777 assert(elem_bt != T_ILLEGAL, "sanity");
5778
5779 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
5780 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5781 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
5782
5783 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5784 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5785 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5786
5787 call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5788
5789 if (!stopped()) {
5790 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5791
5792 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5793 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5794 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5795 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5796
5797 Node* vmask = _gvn.transform(new VectorMaskGenNode(ConvI2X(casted_length), TypeVect::VECTMASK, elem_bt));
5798 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5799 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5800 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5801
5802 exit_block->init_req(inline_path, control());
5803 memory_phi->init_req(inline_path, map()->memory());
5804 result_phi->init_req(inline_path, result);
5805
5806 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5807 clear_upper_avx();
5808 }
5809 }
5810 }
5811
5812 if (call_stub_path != NULL) {
5813 set_control(call_stub_path);
5814
5815 Node* call = make_runtime_call(RC_LEAF,
5816 OptoRuntime::vectorizedMismatch_Type(),
5817 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5818 obja_adr, objb_adr, length, scale);
5819
5820 exit_block->init_req(stub_path, control());
5821 memory_phi->init_req(stub_path, map()->memory());
5822 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5823 }
5824
5825 exit_block = _gvn.transform(exit_block);
5826 memory_phi = _gvn.transform(memory_phi);
5827 result_phi = _gvn.transform(result_phi);
5828
5829 set_control(exit_block);
5830 set_all_memory(memory_phi);
5831 set_result(result_phi);
5832
5833 return true;
5834 }
5835
5836 /**
5837 * Calculate CRC32 for byte.
5838 * int java.util.zip.CRC32.update(int crc, int b)
5839 */
5840 bool LibraryCallKit::inline_updateCRC32() {
5841 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5842 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5843 // no receiver since it is static method
5844 Node* crc = argument(0); // type: int
5845 Node* b = argument(1); // type: int
5846
5847 /*
5848 * int c = ~ crc;
5849 * b = timesXtoThe32[(b ^ c) & 0xFF];
5850 * b = b ^ (c >>> 8);
5851 * crc = ~b;
5852 */
5853
5854 Node* M1 = intcon(-1);
5855 crc = _gvn.transform(new XorINode(crc, M1));
5856 Node* result = _gvn.transform(new XorINode(crc, b));
5857 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5858
5859 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5860 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5861 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5862 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5863
5864 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5865 result = _gvn.transform(new XorINode(crc, result));
5866 result = _gvn.transform(new XorINode(result, M1));
5867 set_result(result);
5868 return true;
5869 }
5870
5871 /**
5872 * Calculate CRC32 for byte[] array.
5873 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5874 */
5875 bool LibraryCallKit::inline_updateBytesCRC32() {
5876 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5877 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5878 // no receiver since it is static method
5879 Node* crc = argument(0); // type: int
5880 Node* src = argument(1); // type: oop
5881 Node* offset = argument(2); // type: int
5882 Node* length = argument(3); // type: int
5883
5884 const Type* src_type = src->Value(&_gvn);
5885 const TypeAryPtr* top_src = src_type->isa_aryptr();
5886 if (top_src == NULL || top_src->klass() == NULL) {
5887 // failed array check
5888 return false;
5889 }
5890
5891 // Figure out the size and type of the elements we will be copying.
5892 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5893 if (src_elem != T_BYTE) {
5894 return false;
5895 }
5896
5897 // 'src_start' points to src array + scaled offset
5898 src = must_be_not_null(src, true);
5899 Node* src_start = array_element_address(src, offset, src_elem);
5900
5901 // We assume that range check is done by caller.
5902 // TODO: generate range check (offset+length < src.length) in debug VM.
5903
5904 // Call the stub.
5905 address stubAddr = StubRoutines::updateBytesCRC32();
5906 const char *stubName = "updateBytesCRC32";
5907
5908 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5909 stubAddr, stubName, TypePtr::BOTTOM,
5910 crc, src_start, length);
5911 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5912 set_result(result);
5913 return true;
5914 }
5915
5916 /**
5917 * Calculate CRC32 for ByteBuffer.
5918 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5919 */
5920 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5921 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5922 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5923 // no receiver since it is static method
5924 Node* crc = argument(0); // type: int
5925 Node* src = argument(1); // type: long
5926 Node* offset = argument(3); // type: int
5927 Node* length = argument(4); // type: int
5928
5929 src = ConvL2X(src); // adjust Java long to machine word
5930 Node* base = _gvn.transform(new CastX2PNode(src));
5931 offset = ConvI2X(offset);
5932
5933 // 'src_start' points to src array + scaled offset
5934 Node* src_start = basic_plus_adr(top(), base, offset);
5935
5936 // Call the stub.
5937 address stubAddr = StubRoutines::updateBytesCRC32();
5938 const char *stubName = "updateBytesCRC32";
5939
5940 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5941 stubAddr, stubName, TypePtr::BOTTOM,
5942 crc, src_start, length);
5943 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5944 set_result(result);
5945 return true;
5946 }
5947
5948 //------------------------------get_table_from_crc32c_class-----------------------
5949 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5950 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
5951 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5952
5953 return table;
5954 }
5955
5956 //------------------------------inline_updateBytesCRC32C-----------------------
5957 //
5958 // Calculate CRC32C for byte[] array.
5959 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5960 //
5961 bool LibraryCallKit::inline_updateBytesCRC32C() {
5962 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5963 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5964 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5965 // no receiver since it is a static method
5966 Node* crc = argument(0); // type: int
5967 Node* src = argument(1); // type: oop
5968 Node* offset = argument(2); // type: int
5969 Node* end = argument(3); // type: int
5970
5971 Node* length = _gvn.transform(new SubINode(end, offset));
5972
5973 const Type* src_type = src->Value(&_gvn);
5974 const TypeAryPtr* top_src = src_type->isa_aryptr();
5975 if (top_src == NULL || top_src->klass() == NULL) {
5976 // failed array check
5977 return false;
5978 }
5979
5980 // Figure out the size and type of the elements we will be copying.
5981 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5982 if (src_elem != T_BYTE) {
5983 return false;
5984 }
5985
5986 // 'src_start' points to src array + scaled offset
5987 src = must_be_not_null(src, true);
5988 Node* src_start = array_element_address(src, offset, src_elem);
5989
5990 // static final int[] byteTable in class CRC32C
5991 Node* table = get_table_from_crc32c_class(callee()->holder());
5992 table = must_be_not_null(table, true);
5993 Node* table_start = array_element_address(table, intcon(0), T_INT);
5994
5995 // We assume that range check is done by caller.
5996 // TODO: generate range check (offset+length < src.length) in debug VM.
5997
5998 // Call the stub.
5999 address stubAddr = StubRoutines::updateBytesCRC32C();
6000 const char *stubName = "updateBytesCRC32C";
6001
6002 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6003 stubAddr, stubName, TypePtr::BOTTOM,
6004 crc, src_start, length, table_start);
6005 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6006 set_result(result);
6007 return true;
6008 }
6009
6010 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6011 //
6012 // Calculate CRC32C for DirectByteBuffer.
6013 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6014 //
6015 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6016 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6017 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6018 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6019 // no receiver since it is a static method
6020 Node* crc = argument(0); // type: int
6021 Node* src = argument(1); // type: long
6022 Node* offset = argument(3); // type: int
6023 Node* end = argument(4); // type: int
6024
6025 Node* length = _gvn.transform(new SubINode(end, offset));
6026
6027 src = ConvL2X(src); // adjust Java long to machine word
6028 Node* base = _gvn.transform(new CastX2PNode(src));
6029 offset = ConvI2X(offset);
6030
6031 // 'src_start' points to src array + scaled offset
6032 Node* src_start = basic_plus_adr(top(), base, offset);
6033
6034 // static final int[] byteTable in class CRC32C
6035 Node* table = get_table_from_crc32c_class(callee()->holder());
6036 table = must_be_not_null(table, true);
6037 Node* table_start = array_element_address(table, intcon(0), T_INT);
6038
6039 // Call the stub.
6040 address stubAddr = StubRoutines::updateBytesCRC32C();
6041 const char *stubName = "updateBytesCRC32C";
6042
6043 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6044 stubAddr, stubName, TypePtr::BOTTOM,
6045 crc, src_start, length, table_start);
6046 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6047 set_result(result);
6048 return true;
6049 }
6050
6051 //------------------------------inline_updateBytesAdler32----------------------
6052 //
6053 // Calculate Adler32 checksum for byte[] array.
6054 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6055 //
6056 bool LibraryCallKit::inline_updateBytesAdler32() {
6057 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6058 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6059 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6060 // no receiver since it is static method
6061 Node* crc = argument(0); // type: int
6062 Node* src = argument(1); // type: oop
6063 Node* offset = argument(2); // type: int
6064 Node* length = argument(3); // type: int
6065
6066 const Type* src_type = src->Value(&_gvn);
6067 const TypeAryPtr* top_src = src_type->isa_aryptr();
6068 if (top_src == NULL || top_src->klass() == NULL) {
6069 // failed array check
6070 return false;
6071 }
6072
6073 // Figure out the size and type of the elements we will be copying.
6074 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6075 if (src_elem != T_BYTE) {
6076 return false;
6077 }
6078
6079 // 'src_start' points to src array + scaled offset
6080 Node* src_start = array_element_address(src, offset, src_elem);
6081
6082 // We assume that range check is done by caller.
6083 // TODO: generate range check (offset+length < src.length) in debug VM.
6084
6085 // Call the stub.
6086 address stubAddr = StubRoutines::updateBytesAdler32();
6087 const char *stubName = "updateBytesAdler32";
6088
6089 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6090 stubAddr, stubName, TypePtr::BOTTOM,
6091 crc, src_start, length);
6092 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6093 set_result(result);
6094 return true;
6095 }
6096
6097 //------------------------------inline_updateByteBufferAdler32---------------
6098 //
6099 // Calculate Adler32 checksum for DirectByteBuffer.
6100 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6101 //
6102 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6103 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6104 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6105 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6106 // no receiver since it is static method
6107 Node* crc = argument(0); // type: int
6108 Node* src = argument(1); // type: long
6109 Node* offset = argument(3); // type: int
6110 Node* length = argument(4); // type: int
6111
6112 src = ConvL2X(src); // adjust Java long to machine word
6113 Node* base = _gvn.transform(new CastX2PNode(src));
6114 offset = ConvI2X(offset);
6115
6116 // 'src_start' points to src array + scaled offset
6117 Node* src_start = basic_plus_adr(top(), base, offset);
6118
6119 // Call the stub.
6120 address stubAddr = StubRoutines::updateBytesAdler32();
6121 const char *stubName = "updateBytesAdler32";
6122
6123 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6124 stubAddr, stubName, TypePtr::BOTTOM,
6125 crc, src_start, length);
6126
6127 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6128 set_result(result);
6129 return true;
6130 }
6131
6132 //----------------------------inline_reference_get----------------------------
6133 // public T java.lang.ref.Reference.get();
6134 bool LibraryCallKit::inline_reference_get() {
6135 const int referent_offset = java_lang_ref_Reference::referent_offset();
6136
6137 // Get the argument:
6138 Node* reference_obj = null_check_receiver();
6139 if (stopped()) return true;
6140
6141 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6142 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6143 decorators, /*is_static*/ false, NULL);
6144 if (result == NULL) return false;
6145
6146 // Add memory barrier to prevent commoning reads from this field
6147 // across safepoint since GC can change its value.
6148 insert_mem_bar(Op_MemBarCPUOrder);
6149
6150 set_result(result);
6151 return true;
6152 }
6153
6154 //----------------------------inline_reference_refersTo0----------------------------
6155 // bool java.lang.ref.Reference.refersTo0();
6156 // bool java.lang.ref.PhantomReference.refersTo0();
6157 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6158 // Get arguments:
6159 Node* reference_obj = null_check_receiver();
6160 Node* other_obj = argument(1);
6161 if (stopped()) return true;
6162
6163 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6164 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6165 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6166 decorators, /*is_static*/ false, NULL);
6167 if (referent == NULL) return false;
6168
6169 // Add memory barrier to prevent commoning reads from this field
6170 // across safepoint since GC can change its value.
6171 insert_mem_bar(Op_MemBarCPUOrder);
6172
6173 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6174 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6175 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6176
6177 RegionNode* region = new RegionNode(3);
6178 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6179
6180 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6181 region->init_req(1, if_true);
6182 phi->init_req(1, intcon(1));
6183
6184 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6185 region->init_req(2, if_false);
6186 phi->init_req(2, intcon(0));
6187
6188 set_control(_gvn.transform(region));
6189 record_for_igvn(region);
6190 set_result(_gvn.transform(phi));
6191 return true;
6192 }
6193
6194
6195 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6196 DecoratorSet decorators = IN_HEAP, bool is_static = false,
6197 ciInstanceKlass* fromKls = NULL) {
6198 if (fromKls == NULL) {
6199 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6200 assert(tinst != NULL, "obj is null");
6201 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6202 fromKls = tinst->klass()->as_instance_klass();
6203 } else {
6204 assert(is_static, "only for static field access");
6205 }
6206 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6207 ciSymbol::make(fieldTypeString),
6208 is_static);
6209
6210 assert (field != NULL, "undefined field");
6211 if (field == NULL) return (Node *) NULL;
6212
6213 if (is_static) {
6214 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6215 fromObj = makecon(tip);
6216 }
6217
6218 // Next code copied from Parse::do_get_xxx():
6219
6220 // Compute address and memory type.
6221 int offset = field->offset_in_bytes();
6222 bool is_vol = field->is_volatile();
6223 ciType* field_klass = field->type();
6224 assert(field_klass->is_loaded(), "should be loaded");
6225 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6226 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6227 BasicType bt = field->layout_type();
6228
6229 // Build the resultant type of the load
6230 const Type *type;
6231 if (bt == T_OBJECT) {
6232 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6233 } else {
6234 type = Type::get_const_basic_type(bt);
6235 }
6236
6237 if (is_vol) {
6238 decorators |= MO_SEQ_CST;
6239 }
6240
6241 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6242 }
6243
6244 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6245 bool is_exact = true, bool is_static = false,
6246 ciInstanceKlass * fromKls = NULL) {
6247 if (fromKls == NULL) {
6248 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6249 assert(tinst != NULL, "obj is null");
6250 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6251 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6252 fromKls = tinst->klass()->as_instance_klass();
6253 }
6254 else {
6255 assert(is_static, "only for static field access");
6256 }
6257 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6258 ciSymbol::make(fieldTypeString),
6259 is_static);
6260
6261 assert(field != NULL, "undefined field");
6262 assert(!field->is_volatile(), "not defined for volatile fields");
6263
6264 if (is_static) {
6265 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6266 fromObj = makecon(tip);
6267 }
6268
6269 // Next code copied from Parse::do_get_xxx():
6270
6271 // Compute address and memory type.
6272 int offset = field->offset_in_bytes();
6273 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6274
6275 return adr;
6276 }
6277
6278 //------------------------------inline_aescrypt_Block-----------------------
6279 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6280 address stubAddr = NULL;
6281 const char *stubName;
6282 assert(UseAES, "need AES instruction support");
6283
6284 switch(id) {
6285 case vmIntrinsics::_aescrypt_encryptBlock:
6286 stubAddr = StubRoutines::aescrypt_encryptBlock();
6287 stubName = "aescrypt_encryptBlock";
6288 break;
6289 case vmIntrinsics::_aescrypt_decryptBlock:
6290 stubAddr = StubRoutines::aescrypt_decryptBlock();
6291 stubName = "aescrypt_decryptBlock";
6292 break;
6293 default:
6294 break;
6295 }
6296 if (stubAddr == NULL) return false;
6297
6298 Node* aescrypt_object = argument(0);
6299 Node* src = argument(1);
6300 Node* src_offset = argument(2);
6301 Node* dest = argument(3);
6302 Node* dest_offset = argument(4);
6303
6304 src = must_be_not_null(src, true);
6305 dest = must_be_not_null(dest, true);
6306
6307 // (1) src and dest are arrays.
6308 const Type* src_type = src->Value(&_gvn);
6309 const Type* dest_type = dest->Value(&_gvn);
6310 const TypeAryPtr* top_src = src_type->isa_aryptr();
6311 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6312 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6313
6314 // for the quick and dirty code we will skip all the checks.
6315 // we are just trying to get the call to be generated.
6316 Node* src_start = src;
6317 Node* dest_start = dest;
6318 if (src_offset != NULL || dest_offset != NULL) {
6319 assert(src_offset != NULL && dest_offset != NULL, "");
6320 src_start = array_element_address(src, src_offset, T_BYTE);
6321 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6322 }
6323
6324 // now need to get the start of its expanded key array
6325 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6326 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6327 if (k_start == NULL) return false;
6328
6329 // Call the stub.
6330 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6331 stubAddr, stubName, TypePtr::BOTTOM,
6332 src_start, dest_start, k_start);
6333
6334 return true;
6335 }
6336
6337 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6338 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6339 address stubAddr = NULL;
6340 const char *stubName = NULL;
6341
6342 assert(UseAES, "need AES instruction support");
6343
6344 switch(id) {
6345 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6346 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6347 stubName = "cipherBlockChaining_encryptAESCrypt";
6348 break;
6349 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6350 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6351 stubName = "cipherBlockChaining_decryptAESCrypt";
6352 break;
6353 default:
6354 break;
6355 }
6356 if (stubAddr == NULL) return false;
6357
6358 Node* cipherBlockChaining_object = argument(0);
6359 Node* src = argument(1);
6360 Node* src_offset = argument(2);
6361 Node* len = argument(3);
6362 Node* dest = argument(4);
6363 Node* dest_offset = argument(5);
6364
6365 src = must_be_not_null(src, false);
6366 dest = must_be_not_null(dest, false);
6367
6368 // (1) src and dest are arrays.
6369 const Type* src_type = src->Value(&_gvn);
6370 const Type* dest_type = dest->Value(&_gvn);
6371 const TypeAryPtr* top_src = src_type->isa_aryptr();
6372 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6373 assert (top_src != NULL && top_src->klass() != NULL
6374 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6375
6376 // checks are the responsibility of the caller
6377 Node* src_start = src;
6378 Node* dest_start = dest;
6379 if (src_offset != NULL || dest_offset != NULL) {
6380 assert(src_offset != NULL && dest_offset != NULL, "");
6381 src_start = array_element_address(src, src_offset, T_BYTE);
6382 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6383 }
6384
6385 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6386 // (because of the predicated logic executed earlier).
6387 // so we cast it here safely.
6388 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6389
6390 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6391 if (embeddedCipherObj == NULL) return false;
6392
6393 // cast it to what we know it will be at runtime
6394 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6395 assert(tinst != NULL, "CBC obj is null");
6396 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6397 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6398 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6399
6400 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6401 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6402 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6403 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6404 aescrypt_object = _gvn.transform(aescrypt_object);
6405
6406 // we need to get the start of the aescrypt_object's expanded key array
6407 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6408 if (k_start == NULL) return false;
6409
6410 // similarly, get the start address of the r vector
6411 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6412 if (objRvec == NULL) return false;
6413 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6414
6415 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6416 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6417 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6418 stubAddr, stubName, TypePtr::BOTTOM,
6419 src_start, dest_start, k_start, r_start, len);
6420
6421 // return cipher length (int)
6422 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6423 set_result(retvalue);
6424 return true;
6425 }
6426
6427 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6428 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6429 address stubAddr = NULL;
6430 const char *stubName = NULL;
6431
6432 assert(UseAES, "need AES instruction support");
6433
6434 switch (id) {
6435 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6436 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6437 stubName = "electronicCodeBook_encryptAESCrypt";
6438 break;
6439 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6440 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6441 stubName = "electronicCodeBook_decryptAESCrypt";
6442 break;
6443 default:
6444 break;
6445 }
6446
6447 if (stubAddr == NULL) return false;
6448
6449 Node* electronicCodeBook_object = argument(0);
6450 Node* src = argument(1);
6451 Node* src_offset = argument(2);
6452 Node* len = argument(3);
6453 Node* dest = argument(4);
6454 Node* dest_offset = argument(5);
6455
6456 // (1) src and dest are arrays.
6457 const Type* src_type = src->Value(&_gvn);
6458 const Type* dest_type = dest->Value(&_gvn);
6459 const TypeAryPtr* top_src = src_type->isa_aryptr();
6460 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6461 assert(top_src != NULL && top_src->klass() != NULL
6462 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6463
6464 // checks are the responsibility of the caller
6465 Node* src_start = src;
6466 Node* dest_start = dest;
6467 if (src_offset != NULL || dest_offset != NULL) {
6468 assert(src_offset != NULL && dest_offset != NULL, "");
6469 src_start = array_element_address(src, src_offset, T_BYTE);
6470 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6471 }
6472
6473 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6474 // (because of the predicated logic executed earlier).
6475 // so we cast it here safely.
6476 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6477
6478 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6479 if (embeddedCipherObj == NULL) return false;
6480
6481 // cast it to what we know it will be at runtime
6482 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6483 assert(tinst != NULL, "ECB obj is null");
6484 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6485 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6486 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6487
6488 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6489 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6490 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6491 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6492 aescrypt_object = _gvn.transform(aescrypt_object);
6493
6494 // we need to get the start of the aescrypt_object's expanded key array
6495 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6496 if (k_start == NULL) return false;
6497
6498 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6499 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6500 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6501 stubAddr, stubName, TypePtr::BOTTOM,
6502 src_start, dest_start, k_start, len);
6503
6504 // return cipher length (int)
6505 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6506 set_result(retvalue);
6507 return true;
6508 }
6509
6510 //------------------------------inline_counterMode_AESCrypt-----------------------
6511 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6512 assert(UseAES, "need AES instruction support");
6513 if (!UseAESCTRIntrinsics) return false;
6514
6515 address stubAddr = NULL;
6516 const char *stubName = NULL;
6517 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6518 stubAddr = StubRoutines::counterMode_AESCrypt();
6519 stubName = "counterMode_AESCrypt";
6520 }
6521 if (stubAddr == NULL) return false;
6522
6523 Node* counterMode_object = argument(0);
6524 Node* src = argument(1);
6525 Node* src_offset = argument(2);
6526 Node* len = argument(3);
6527 Node* dest = argument(4);
6528 Node* dest_offset = argument(5);
6529
6530 // (1) src and dest are arrays.
6531 const Type* src_type = src->Value(&_gvn);
6532 const Type* dest_type = dest->Value(&_gvn);
6533 const TypeAryPtr* top_src = src_type->isa_aryptr();
6534 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6535 assert(top_src != NULL && top_src->klass() != NULL &&
6536 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6537
6538 // checks are the responsibility of the caller
6539 Node* src_start = src;
6540 Node* dest_start = dest;
6541 if (src_offset != NULL || dest_offset != NULL) {
6542 assert(src_offset != NULL && dest_offset != NULL, "");
6543 src_start = array_element_address(src, src_offset, T_BYTE);
6544 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6545 }
6546
6547 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6548 // (because of the predicated logic executed earlier).
6549 // so we cast it here safely.
6550 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6551 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6552 if (embeddedCipherObj == NULL) return false;
6553 // cast it to what we know it will be at runtime
6554 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6555 assert(tinst != NULL, "CTR obj is null");
6556 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6557 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6558 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6559 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6560 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6561 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6562 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6563 aescrypt_object = _gvn.transform(aescrypt_object);
6564 // we need to get the start of the aescrypt_object's expanded key array
6565 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6566 if (k_start == NULL) return false;
6567 // similarly, get the start address of the r vector
6568 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6569 if (obj_counter == NULL) return false;
6570 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6571
6572 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6573 if (saved_encCounter == NULL) return false;
6574 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6575 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6576
6577 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6578 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6579 OptoRuntime::counterMode_aescrypt_Type(),
6580 stubAddr, stubName, TypePtr::BOTTOM,
6581 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6582
6583 // return cipher length (int)
6584 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6585 set_result(retvalue);
6586 return true;
6587 }
6588
6589 //------------------------------get_key_start_from_aescrypt_object-----------------------
6590 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6591 #if defined(PPC64) || defined(S390)
6592 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6593 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6594 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6595 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6596 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6597 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6598 if (objSessionK == NULL) {
6599 return (Node *) NULL;
6600 }
6601 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6602 #else
6603 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6604 #endif // PPC64
6605 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6606 if (objAESCryptKey == NULL) return (Node *) NULL;
6607
6608 // now have the array, need to get the start address of the K array
6609 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6610 return k_start;
6611 }
6612
6613 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6614 // Return node representing slow path of predicate check.
6615 // the pseudo code we want to emulate with this predicate is:
6616 // for encryption:
6617 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6618 // for decryption:
6619 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6620 // note cipher==plain is more conservative than the original java code but that's OK
6621 //
6622 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6623 // The receiver was checked for NULL already.
6624 Node* objCBC = argument(0);
6625
6626 Node* src = argument(1);
6627 Node* dest = argument(4);
6628
6629 // Load embeddedCipher field of CipherBlockChaining object.
6630 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6631
6632 // get AESCrypt klass for instanceOf check
6633 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6634 // will have same classloader as CipherBlockChaining object
6635 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6636 assert(tinst != NULL, "CBCobj is null");
6637 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6638
6639 // we want to do an instanceof comparison against the AESCrypt class
6640 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6641 if (!klass_AESCrypt->is_loaded()) {
6642 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6643 Node* ctrl = control();
6644 set_control(top()); // no regular fast path
6645 return ctrl;
6646 }
6647
6648 src = must_be_not_null(src, true);
6649 dest = must_be_not_null(dest, true);
6650
6651 // Resolve oops to stable for CmpP below.
6652 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6653
6654 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6655 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6656 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6657
6658 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6659
6660 // for encryption, we are done
6661 if (!decrypting)
6662 return instof_false; // even if it is NULL
6663
6664 // for decryption, we need to add a further check to avoid
6665 // taking the intrinsic path when cipher and plain are the same
6666 // see the original java code for why.
6667 RegionNode* region = new RegionNode(3);
6668 region->init_req(1, instof_false);
6669
6670 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6671 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6672 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6673 region->init_req(2, src_dest_conjoint);
6674
6675 record_for_igvn(region);
6676 return _gvn.transform(region);
6677 }
6678
6679 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6680 // Return node representing slow path of predicate check.
6681 // the pseudo code we want to emulate with this predicate is:
6682 // for encryption:
6683 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6684 // for decryption:
6685 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6686 // note cipher==plain is more conservative than the original java code but that's OK
6687 //
6688 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6689 // The receiver was checked for NULL already.
6690 Node* objECB = argument(0);
6691
6692 // Load embeddedCipher field of ElectronicCodeBook object.
6693 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6694
6695 // get AESCrypt klass for instanceOf check
6696 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6697 // will have same classloader as ElectronicCodeBook object
6698 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6699 assert(tinst != NULL, "ECBobj is null");
6700 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6701
6702 // we want to do an instanceof comparison against the AESCrypt class
6703 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6704 if (!klass_AESCrypt->is_loaded()) {
6705 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6706 Node* ctrl = control();
6707 set_control(top()); // no regular fast path
6708 return ctrl;
6709 }
6710 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6711
6712 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6713 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6714 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6715
6716 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6717
6718 // for encryption, we are done
6719 if (!decrypting)
6720 return instof_false; // even if it is NULL
6721
6722 // for decryption, we need to add a further check to avoid
6723 // taking the intrinsic path when cipher and plain are the same
6724 // see the original java code for why.
6725 RegionNode* region = new RegionNode(3);
6726 region->init_req(1, instof_false);
6727 Node* src = argument(1);
6728 Node* dest = argument(4);
6729 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6730 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6731 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6732 region->init_req(2, src_dest_conjoint);
6733
6734 record_for_igvn(region);
6735 return _gvn.transform(region);
6736 }
6737
6738 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6739 // Return node representing slow path of predicate check.
6740 // the pseudo code we want to emulate with this predicate is:
6741 // for encryption:
6742 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6743 // for decryption:
6744 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6745 // note cipher==plain is more conservative than the original java code but that's OK
6746 //
6747
6748 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6749 // The receiver was checked for NULL already.
6750 Node* objCTR = argument(0);
6751
6752 // Load embeddedCipher field of CipherBlockChaining object.
6753 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6754
6755 // get AESCrypt klass for instanceOf check
6756 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6757 // will have same classloader as CipherBlockChaining object
6758 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6759 assert(tinst != NULL, "CTRobj is null");
6760 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6761
6762 // we want to do an instanceof comparison against the AESCrypt class
6763 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6764 if (!klass_AESCrypt->is_loaded()) {
6765 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6766 Node* ctrl = control();
6767 set_control(top()); // no regular fast path
6768 return ctrl;
6769 }
6770
6771 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6772 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6773 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6774 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6775 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6776
6777 return instof_false; // even if it is NULL
6778 }
6779
6780 //------------------------------inline_ghash_processBlocks
6781 bool LibraryCallKit::inline_ghash_processBlocks() {
6782 address stubAddr;
6783 const char *stubName;
6784 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6785
6786 stubAddr = StubRoutines::ghash_processBlocks();
6787 stubName = "ghash_processBlocks";
6788
6789 Node* data = argument(0);
6790 Node* offset = argument(1);
6791 Node* len = argument(2);
6792 Node* state = argument(3);
6793 Node* subkeyH = argument(4);
6794
6795 state = must_be_not_null(state, true);
6796 subkeyH = must_be_not_null(subkeyH, true);
6797 data = must_be_not_null(data, true);
6798
6799 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6800 assert(state_start, "state is NULL");
6801 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6802 assert(subkeyH_start, "subkeyH is NULL");
6803 Node* data_start = array_element_address(data, offset, T_BYTE);
6804 assert(data_start, "data is NULL");
6805
6806 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6807 OptoRuntime::ghash_processBlocks_Type(),
6808 stubAddr, stubName, TypePtr::BOTTOM,
6809 state_start, subkeyH_start, data_start, len);
6810 return true;
6811 }
6812
6813 bool LibraryCallKit::inline_base64_encodeBlock() {
6814 address stubAddr;
6815 const char *stubName;
6816 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6817 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6818 stubAddr = StubRoutines::base64_encodeBlock();
6819 stubName = "encodeBlock";
6820
6821 if (!stubAddr) return false;
6822 Node* base64obj = argument(0);
6823 Node* src = argument(1);
6824 Node* offset = argument(2);
6825 Node* len = argument(3);
6826 Node* dest = argument(4);
6827 Node* dp = argument(5);
6828 Node* isURL = argument(6);
6829
6830 src = must_be_not_null(src, true);
6831 dest = must_be_not_null(dest, true);
6832
6833 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6834 assert(src_start, "source array is NULL");
6835 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6836 assert(dest_start, "destination array is NULL");
6837
6838 Node* base64 = make_runtime_call(RC_LEAF,
6839 OptoRuntime::base64_encodeBlock_Type(),
6840 stubAddr, stubName, TypePtr::BOTTOM,
6841 src_start, offset, len, dest_start, dp, isURL);
6842 return true;
6843 }
6844
6845 bool LibraryCallKit::inline_base64_decodeBlock() {
6846 address stubAddr;
6847 const char *stubName;
6848 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6849 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6850 stubAddr = StubRoutines::base64_decodeBlock();
6851 stubName = "decodeBlock";
6852
6853 if (!stubAddr) return false;
6854 Node* base64obj = argument(0);
6855 Node* src = argument(1);
6856 Node* src_offset = argument(2);
6857 Node* len = argument(3);
6858 Node* dest = argument(4);
6859 Node* dest_offset = argument(5);
6860 Node* isURL = argument(6);
6861 Node* isMIME = argument(7);
6862
6863 src = must_be_not_null(src, true);
6864 dest = must_be_not_null(dest, true);
6865
6866 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6867 assert(src_start, "source array is NULL");
6868 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6869 assert(dest_start, "destination array is NULL");
6870
6871 Node* call = make_runtime_call(RC_LEAF,
6872 OptoRuntime::base64_decodeBlock_Type(),
6873 stubAddr, stubName, TypePtr::BOTTOM,
6874 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6875 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6876 set_result(result);
6877 return true;
6878 }
6879
6880 //------------------------------inline_digestBase_implCompress-----------------------
6881 //
6882 // Calculate MD5 for single-block byte[] array.
6883 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6884 //
6885 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6886 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6887 //
6888 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6889 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6890 //
6891 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6892 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6893 //
6894 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6895 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6896 //
6897 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6898 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6899
6900 Node* digestBase_obj = argument(0);
6901 Node* src = argument(1); // type oop
6902 Node* ofs = argument(2); // type int
6903
6904 const Type* src_type = src->Value(&_gvn);
6905 const TypeAryPtr* top_src = src_type->isa_aryptr();
6906 if (top_src == NULL || top_src->klass() == NULL) {
6907 // failed array check
6908 return false;
6909 }
6910 // Figure out the size and type of the elements we will be copying.
6911 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6912 if (src_elem != T_BYTE) {
6913 return false;
6914 }
6915 // 'src_start' points to src array + offset
6916 src = must_be_not_null(src, true);
6917 Node* src_start = array_element_address(src, ofs, src_elem);
6918 Node* state = NULL;
6919 Node* digest_length = NULL;
6920 address stubAddr;
6921 const char *stubName;
6922
6923 switch(id) {
6924 case vmIntrinsics::_md5_implCompress:
6925 assert(UseMD5Intrinsics, "need MD5 instruction support");
6926 state = get_state_from_digest_object(digestBase_obj, "[I");
6927 stubAddr = StubRoutines::md5_implCompress();
6928 stubName = "md5_implCompress";
6929 break;
6930 case vmIntrinsics::_sha_implCompress:
6931 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6932 state = get_state_from_digest_object(digestBase_obj, "[I");
6933 stubAddr = StubRoutines::sha1_implCompress();
6934 stubName = "sha1_implCompress";
6935 break;
6936 case vmIntrinsics::_sha2_implCompress:
6937 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6938 state = get_state_from_digest_object(digestBase_obj, "[I");
6939 stubAddr = StubRoutines::sha256_implCompress();
6940 stubName = "sha256_implCompress";
6941 break;
6942 case vmIntrinsics::_sha5_implCompress:
6943 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6944 state = get_state_from_digest_object(digestBase_obj, "[J");
6945 stubAddr = StubRoutines::sha512_implCompress();
6946 stubName = "sha512_implCompress";
6947 break;
6948 case vmIntrinsics::_sha3_implCompress:
6949 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6950 state = get_state_from_digest_object(digestBase_obj, "[B");
6951 stubAddr = StubRoutines::sha3_implCompress();
6952 stubName = "sha3_implCompress";
6953 digest_length = get_digest_length_from_digest_object(digestBase_obj);
6954 if (digest_length == NULL) return false;
6955 break;
6956 default:
6957 fatal_unexpected_iid(id);
6958 return false;
6959 }
6960 if (state == NULL) return false;
6961
6962 assert(stubAddr != NULL, "Stub is generated");
6963 if (stubAddr == NULL) return false;
6964
6965 // Call the stub.
6966 Node* call;
6967 if (digest_length == NULL) {
6968 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6969 stubAddr, stubName, TypePtr::BOTTOM,
6970 src_start, state);
6971 } else {
6972 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6973 stubAddr, stubName, TypePtr::BOTTOM,
6974 src_start, state, digest_length);
6975 }
6976
6977 return true;
6978 }
6979
6980 //------------------------------inline_digestBase_implCompressMB-----------------------
6981 //
6982 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
6983 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6984 //
6985 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6986 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6987 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6988 assert((uint)predicate < 5, "sanity");
6989 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6990
6991 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6992 Node* src = argument(1); // byte[] array
6993 Node* ofs = argument(2); // type int
6994 Node* limit = argument(3); // type int
6995
6996 const Type* src_type = src->Value(&_gvn);
6997 const TypeAryPtr* top_src = src_type->isa_aryptr();
6998 if (top_src == NULL || top_src->klass() == NULL) {
6999 // failed array check
7000 return false;
7001 }
7002 // Figure out the size and type of the elements we will be copying.
7003 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
7004 if (src_elem != T_BYTE) {
7005 return false;
7006 }
7007 // 'src_start' points to src array + offset
7008 src = must_be_not_null(src, false);
7009 Node* src_start = array_element_address(src, ofs, src_elem);
7010
7011 const char* klass_digestBase_name = NULL;
7012 const char* stub_name = NULL;
7013 address stub_addr = NULL;
7014 const char* state_type = "[I";
7015
7016 switch (predicate) {
7017 case 0:
7018 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7019 klass_digestBase_name = "sun/security/provider/MD5";
7020 stub_name = "md5_implCompressMB";
7021 stub_addr = StubRoutines::md5_implCompressMB();
7022 }
7023 break;
7024 case 1:
7025 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7026 klass_digestBase_name = "sun/security/provider/SHA";
7027 stub_name = "sha1_implCompressMB";
7028 stub_addr = StubRoutines::sha1_implCompressMB();
7029 }
7030 break;
7031 case 2:
7032 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7033 klass_digestBase_name = "sun/security/provider/SHA2";
7034 stub_name = "sha256_implCompressMB";
7035 stub_addr = StubRoutines::sha256_implCompressMB();
7036 }
7037 break;
7038 case 3:
7039 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7040 klass_digestBase_name = "sun/security/provider/SHA5";
7041 stub_name = "sha512_implCompressMB";
7042 stub_addr = StubRoutines::sha512_implCompressMB();
7043 state_type = "[J";
7044 }
7045 break;
7046 case 4:
7047 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7048 klass_digestBase_name = "sun/security/provider/SHA3";
7049 stub_name = "sha3_implCompressMB";
7050 stub_addr = StubRoutines::sha3_implCompressMB();
7051 state_type = "[B";
7052 }
7053 break;
7054 default:
7055 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7056 }
7057 if (klass_digestBase_name != NULL) {
7058 assert(stub_addr != NULL, "Stub is generated");
7059 if (stub_addr == NULL) return false;
7060
7061 // get DigestBase klass to lookup for SHA klass
7062 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7063 assert(tinst != NULL, "digestBase_obj is not instance???");
7064 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7065
7066 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7067 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7068 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7069 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, state_type, stub_addr, stub_name, src_start, ofs, limit);
7070 }
7071 return false;
7072 }
7073
7074 //------------------------------inline_digestBase_implCompressMB-----------------------
7075 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7076 const char* state_type, address stubAddr, const char *stubName,
7077 Node* src_start, Node* ofs, Node* limit) {
7078 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7079 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7080 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7081 digest_obj = _gvn.transform(digest_obj);
7082
7083 Node* state = get_state_from_digest_object(digest_obj, state_type);
7084 if (state == NULL) return false;
7085
7086 Node* digest_length = NULL;
7087 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7088 digest_length = get_digest_length_from_digest_object(digest_obj);
7089 if (digest_length == NULL) return false;
7090 }
7091
7092 // Call the stub.
7093 Node* call;
7094 if (digest_length == NULL) {
7095 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7096 OptoRuntime::digestBase_implCompressMB_Type(false),
7097 stubAddr, stubName, TypePtr::BOTTOM,
7098 src_start, state, ofs, limit);
7099 } else {
7100 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7101 OptoRuntime::digestBase_implCompressMB_Type(true),
7102 stubAddr, stubName, TypePtr::BOTTOM,
7103 src_start, state, digest_length, ofs, limit);
7104 }
7105
7106 // return ofs (int)
7107 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7108 set_result(result);
7109
7110 return true;
7111 }
7112
7113 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7114 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7115 assert(UseAES, "need AES instruction support");
7116 address stubAddr = NULL;
7117 const char *stubName = NULL;
7118 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7119 stubName = "galoisCounterMode_AESCrypt";
7120
7121 if (stubAddr == NULL) return false;
7122
7123 Node* in = argument(0);
7124 Node* inOfs = argument(1);
7125 Node* len = argument(2);
7126 Node* ct = argument(3);
7127 Node* ctOfs = argument(4);
7128 Node* out = argument(5);
7129 Node* outOfs = argument(6);
7130 Node* gctr_object = argument(7);
7131 Node* ghash_object = argument(8);
7132
7133 // (1) in, ct and out are arrays.
7134 const Type* in_type = in->Value(&_gvn);
7135 const Type* ct_type = ct->Value(&_gvn);
7136 const Type* out_type = out->Value(&_gvn);
7137 const TypeAryPtr* top_in = in_type->isa_aryptr();
7138 const TypeAryPtr* top_ct = ct_type->isa_aryptr();
7139 const TypeAryPtr* top_out = out_type->isa_aryptr();
7140 assert(top_in != NULL && top_in->klass() != NULL &&
7141 top_ct != NULL && top_ct->klass() != NULL &&
7142 top_out != NULL && top_out->klass() != NULL, "args are strange");
7143
7144 // checks are the responsibility of the caller
7145 Node* in_start = in;
7146 Node* ct_start = ct;
7147 Node* out_start = out;
7148 if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
7149 assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
7150 in_start = array_element_address(in, inOfs, T_BYTE);
7151 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7152 out_start = array_element_address(out, outOfs, T_BYTE);
7153 }
7154
7155 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7156 // (because of the predicated logic executed earlier).
7157 // so we cast it here safely.
7158 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7159 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7160 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7161 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7162 Node* state = load_field_from_object(ghash_object, "state", "[J");
7163
7164 if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
7165 return false;
7166 }
7167 // cast it to what we know it will be at runtime
7168 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7169 assert(tinst != NULL, "GCTR obj is null");
7170 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
7171 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7172 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7173 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7174 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7175 const TypeOopPtr* xtype = aklass->as_instance_type();
7176 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7177 aescrypt_object = _gvn.transform(aescrypt_object);
7178 // we need to get the start of the aescrypt_object's expanded key array
7179 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7180 if (k_start == NULL) return false;
7181
7182 // similarly, get the start address of the r vector
7183 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7184 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7185 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7186
7187 ciKlass* klass = ciTypeArrayKlass::make(T_LONG);
7188 Node* klass_node = makecon(TypeKlassPtr::make(klass));
7189
7190 // Does this target support this intrinsic?
7191 if (Matcher::htbl_entries == -1) return false;
7192
7193 Node* subkeyHtbl_48_entries_start;
7194 if (Matcher::htbl_entries != 0) {
7195 // new array to hold 48 computed htbl entries
7196 Node* subkeyHtbl_48_entries = new_array(klass_node, intcon(Matcher::htbl_entries), 0);
7197 if (subkeyHtbl_48_entries == NULL) return false;
7198 subkeyHtbl_48_entries_start = array_element_address(subkeyHtbl_48_entries, intcon(0), T_LONG);
7199 } else {
7200 // This target doesn't need the extra-large Htbl.
7201 subkeyHtbl_48_entries_start = ConvL2X(intcon(0));
7202 }
7203
7204 // Call the stub, passing params
7205 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7206 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7207 stubAddr, stubName, TypePtr::BOTTOM,
7208 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, subkeyHtbl_48_entries_start, cnt_start);
7209
7210 // return cipher length (int)
7211 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7212 set_result(retvalue);
7213 return true;
7214 }
7215
7216 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7217 // Return node representing slow path of predicate check.
7218 // the pseudo code we want to emulate with this predicate is:
7219 // for encryption:
7220 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7221 // for decryption:
7222 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7223 // note cipher==plain is more conservative than the original java code but that's OK
7224 //
7225
7226 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7227 // The receiver was checked for NULL already.
7228 Node* objGCTR = argument(7);
7229 // Load embeddedCipher field of GCTR object.
7230 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7231 assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
7232
7233 // get AESCrypt klass for instanceOf check
7234 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7235 // will have same classloader as CipherBlockChaining object
7236 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7237 assert(tinst != NULL, "GCTR obj is null");
7238 assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
7239
7240 // we want to do an instanceof comparison against the AESCrypt class
7241 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7242 if (!klass_AESCrypt->is_loaded()) {
7243 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7244 Node* ctrl = control();
7245 set_control(top()); // no regular fast path
7246 return ctrl;
7247 }
7248
7249 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7250 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7251 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7252 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7253 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7254
7255 return instof_false; // even if it is NULL
7256 }
7257
7258 //------------------------------get_state_from_digest_object-----------------------
7259 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, const char *state_type) {
7260 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7261 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7262 if (digest_state == NULL) return (Node *) NULL;
7263
7264 // now have the array, need to get the start address of the state array
7265 Node* state = array_element_address(digest_state, intcon(0), T_INT);
7266 return state;
7267 }
7268
7269 //------------------------------get_digest_length_from_sha3_object----------------------------------
7270 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
7271 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
7272 assert (digest_length != NULL, "sanity");
7273 return digest_length;
7274 }
7275
7276 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7277 // Return node representing slow path of predicate check.
7278 // the pseudo code we want to emulate with this predicate is:
7279 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7280 //
7281 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7282 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7283 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7284 assert((uint)predicate < 5, "sanity");
7285
7286 // The receiver was checked for NULL already.
7287 Node* digestBaseObj = argument(0);
7288
7289 // get DigestBase klass for instanceOf check
7290 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7291 assert(tinst != NULL, "digestBaseObj is null");
7292 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7293
7294 const char* klass_name = NULL;
7295 switch (predicate) {
7296 case 0:
7297 if (UseMD5Intrinsics) {
7298 // we want to do an instanceof comparison against the MD5 class
7299 klass_name = "sun/security/provider/MD5";
7300 }
7301 break;
7302 case 1:
7303 if (UseSHA1Intrinsics) {
7304 // we want to do an instanceof comparison against the SHA class
7305 klass_name = "sun/security/provider/SHA";
7306 }
7307 break;
7308 case 2:
7309 if (UseSHA256Intrinsics) {
7310 // we want to do an instanceof comparison against the SHA2 class
7311 klass_name = "sun/security/provider/SHA2";
7312 }
7313 break;
7314 case 3:
7315 if (UseSHA512Intrinsics) {
7316 // we want to do an instanceof comparison against the SHA5 class
7317 klass_name = "sun/security/provider/SHA5";
7318 }
7319 break;
7320 case 4:
7321 if (UseSHA3Intrinsics) {
7322 // we want to do an instanceof comparison against the SHA3 class
7323 klass_name = "sun/security/provider/SHA3";
7324 }
7325 break;
7326 default:
7327 fatal("unknown SHA intrinsic predicate: %d", predicate);
7328 }
7329
7330 ciKlass* klass = NULL;
7331 if (klass_name != NULL) {
7332 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
7333 }
7334 if ((klass == NULL) || !klass->is_loaded()) {
7335 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7336 Node* ctrl = control();
7337 set_control(top()); // no intrinsic path
7338 return ctrl;
7339 }
7340 ciInstanceKlass* instklass = klass->as_instance_klass();
7341
7342 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7343 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7344 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7345 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7346
7347 return instof_false; // even if it is NULL
7348 }
7349
7350 //-------------inline_fma-----------------------------------
7351 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7352 Node *a = NULL;
7353 Node *b = NULL;
7354 Node *c = NULL;
7355 Node* result = NULL;
7356 switch (id) {
7357 case vmIntrinsics::_fmaD:
7358 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7359 // no receiver since it is static method
7360 a = round_double_node(argument(0));
7361 b = round_double_node(argument(2));
7362 c = round_double_node(argument(4));
7363 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7364 break;
7365 case vmIntrinsics::_fmaF:
7366 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7367 a = argument(0);
7368 b = argument(1);
7369 c = argument(2);
7370 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7371 break;
7372 default:
7373 fatal_unexpected_iid(id); break;
7374 }
7375 set_result(result);
7376 return true;
7377 }
7378
7379 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7380 // argument(0) is receiver
7381 Node* codePoint = argument(1);
7382 Node* n = NULL;
7383
7384 switch (id) {
7385 case vmIntrinsics::_isDigit :
7386 n = new DigitNode(control(), codePoint);
7387 break;
7388 case vmIntrinsics::_isLowerCase :
7389 n = new LowerCaseNode(control(), codePoint);
7390 break;
7391 case vmIntrinsics::_isUpperCase :
7392 n = new UpperCaseNode(control(), codePoint);
7393 break;
7394 case vmIntrinsics::_isWhitespace :
7395 n = new WhitespaceNode(control(), codePoint);
7396 break;
7397 default:
7398 fatal_unexpected_iid(id);
7399 }
7400
7401 set_result(_gvn.transform(n));
7402 return true;
7403 }
7404
7405 //------------------------------inline_fp_min_max------------------------------
7406 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7407 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7408
7409 // The intrinsic should be used only when the API branches aren't predictable,
7410 // the last one performing the most important comparison. The following heuristic
7411 // uses the branch statistics to eventually bail out if necessary.
7412
7413 ciMethodData *md = callee()->method_data();
7414
7415 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7416 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7417
7418 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7419 // Bail out if the call-site didn't contribute enough to the statistics.
7420 return false;
7421 }
7422
7423 uint taken = 0, not_taken = 0;
7424
7425 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7426 if (p->is_BranchData()) {
7427 taken = ((ciBranchData*)p)->taken();
7428 not_taken = ((ciBranchData*)p)->not_taken();
7429 }
7430 }
7431
7432 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7433 balance = balance < 0 ? -balance : balance;
7434 if ( balance > 0.2 ) {
7435 // Bail out if the most important branch is predictable enough.
7436 return false;
7437 }
7438 }
7439 */
7440
7441 Node *a = NULL;
7442 Node *b = NULL;
7443 Node *n = NULL;
7444 switch (id) {
7445 case vmIntrinsics::_maxF:
7446 case vmIntrinsics::_minF:
7447 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7448 a = argument(0);
7449 b = argument(1);
7450 break;
7451 case vmIntrinsics::_maxD:
7452 case vmIntrinsics::_minD:
7453 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7454 a = round_double_node(argument(0));
7455 b = round_double_node(argument(2));
7456 break;
7457 default:
7458 fatal_unexpected_iid(id);
7459 break;
7460 }
7461 switch (id) {
7462 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
7463 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
7464 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
7465 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
7466 default: fatal_unexpected_iid(id); break;
7467 }
7468 set_result(_gvn.transform(n));
7469 return true;
7470 }
7471
7472 bool LibraryCallKit::inline_profileBoolean() {
7473 Node* counts = argument(1);
7474 const TypeAryPtr* ary = NULL;
7475 ciArray* aobj = NULL;
7476 if (counts->is_Con()
7477 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7478 && (aobj = ary->const_oop()->as_array()) != NULL
7479 && (aobj->length() == 2)) {
7480 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7481 jint false_cnt = aobj->element_value(0).as_int();
7482 jint true_cnt = aobj->element_value(1).as_int();
7483
7484 if (C->log() != NULL) {
7485 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7486 false_cnt, true_cnt);
7487 }
7488
7489 if (false_cnt + true_cnt == 0) {
7490 // According to profile, never executed.
7491 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7492 Deoptimization::Action_reinterpret);
7493 return true;
7494 }
7495
7496 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7497 // is a number of each value occurrences.
7498 Node* result = argument(0);
7499 if (false_cnt == 0 || true_cnt == 0) {
7500 // According to profile, one value has been never seen.
7501 int expected_val = (false_cnt == 0) ? 1 : 0;
7502
7503 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7504 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7505
7506 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7507 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7508 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7509
7510 { // Slow path: uncommon trap for never seen value and then reexecute
7511 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7512 // the value has been seen at least once.
7513 PreserveJVMState pjvms(this);
7514 PreserveReexecuteState preexecs(this);
7515 jvms()->set_should_reexecute(true);
7516
7517 set_control(slow_path);
7518 set_i_o(i_o());
7519
7520 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7521 Deoptimization::Action_reinterpret);
7522 }
7523 // The guard for never seen value enables sharpening of the result and
7524 // returning a constant. It allows to eliminate branches on the same value
7525 // later on.
7526 set_control(fast_path);
7527 result = intcon(expected_val);
7528 }
7529 // Stop profiling.
7530 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7531 // By replacing method body with profile data (represented as ProfileBooleanNode
7532 // on IR level) we effectively disable profiling.
7533 // It enables full speed execution once optimized code is generated.
7534 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7535 C->record_for_igvn(profile);
7536 set_result(profile);
7537 return true;
7538 } else {
7539 // Continue profiling.
7540 // Profile data isn't available at the moment. So, execute method's bytecode version.
7541 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7542 // is compiled and counters aren't available since corresponding MethodHandle
7543 // isn't a compile-time constant.
7544 return false;
7545 }
7546 }
7547
7548 bool LibraryCallKit::inline_isCompileConstant() {
7549 Node* n = argument(0);
7550 set_result(n->is_Con() ? intcon(1) : intcon(0));
7551 return true;
7552 }
7553
7554 //------------------------------- inline_getObjectSize --------------------------------------
7555 //
7556 // Calculate the runtime size of the object/array.
7557 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7558 //
7559 bool LibraryCallKit::inline_getObjectSize() {
7560 Node* obj = argument(3);
7561 Node* klass_node = load_object_klass(obj);
7562
7563 jint layout_con = Klass::_lh_neutral_value;
7564 Node* layout_val = get_layout_helper(klass_node, layout_con);
7565 int layout_is_con = (layout_val == NULL);
7566
7567 if (layout_is_con) {
7568 // Layout helper is constant, can figure out things at compile time.
7569
7570 if (Klass::layout_helper_is_instance(layout_con)) {
7571 // Instance case: layout_con contains the size itself.
7572 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7573 set_result(size);
7574 } else {
7575 // Array case: size is round(header + element_size*arraylength).
7576 // Since arraylength is different for every array instance, we have to
7577 // compute the whole thing at runtime.
7578
7579 Node* arr_length = load_array_length(obj);
7580
7581 int round_mask = MinObjAlignmentInBytes - 1;
7582 int hsize = Klass::layout_helper_header_size(layout_con);
7583 int eshift = Klass::layout_helper_log2_element_size(layout_con);
7584
7585 if ((round_mask & ~right_n_bits(eshift)) == 0) {
7586 round_mask = 0; // strength-reduce it if it goes away completely
7587 }
7588 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7589 Node* header_size = intcon(hsize + round_mask);
7590
7591 Node* lengthx = ConvI2X(arr_length);
7592 Node* headerx = ConvI2X(header_size);
7593
7594 Node* abody = lengthx;
7595 if (eshift != 0) {
7596 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7597 }
7598 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7599 if (round_mask != 0) {
7600 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7601 }
7602 size = ConvX2L(size);
7603 set_result(size);
7604 }
7605 } else {
7606 // Layout helper is not constant, need to test for array-ness at runtime.
7607
7608 enum { _instance_path = 1, _array_path, PATH_LIMIT };
7609 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7610 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7611 record_for_igvn(result_reg);
7612
7613 Node* array_ctl = generate_array_guard(klass_node, NULL);
7614 if (array_ctl != NULL) {
7615 // Array case: size is round(header + element_size*arraylength).
7616 // Since arraylength is different for every array instance, we have to
7617 // compute the whole thing at runtime.
7618
7619 PreserveJVMState pjvms(this);
7620 set_control(array_ctl);
7621 Node* arr_length = load_array_length(obj);
7622
7623 int round_mask = MinObjAlignmentInBytes - 1;
7624 Node* mask = intcon(round_mask);
7625
7626 Node* hss = intcon(Klass::_lh_header_size_shift);
7627 Node* hsm = intcon(Klass::_lh_header_size_mask);
7628 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7629 header_size = _gvn.transform(new AndINode(header_size, hsm));
7630 header_size = _gvn.transform(new AddINode(header_size, mask));
7631
7632 // There is no need to mask or shift this value.
7633 // The semantics of LShiftINode include an implicit mask to 0x1F.
7634 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7635 Node* elem_shift = layout_val;
7636
7637 Node* lengthx = ConvI2X(arr_length);
7638 Node* headerx = ConvI2X(header_size);
7639
7640 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7641 Node* size = _gvn.transform(new AddXNode(headerx, abody));
7642 if (round_mask != 0) {
7643 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7644 }
7645 size = ConvX2L(size);
7646
7647 result_reg->init_req(_array_path, control());
7648 result_val->init_req(_array_path, size);
7649 }
7650
7651 if (!stopped()) {
7652 // Instance case: the layout helper gives us instance size almost directly,
7653 // but we need to mask out the _lh_instance_slow_path_bit.
7654 Node* size = ConvI2X(layout_val);
7655 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7656 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7657 size = _gvn.transform(new AndXNode(size, mask));
7658 size = ConvX2L(size);
7659
7660 result_reg->init_req(_instance_path, control());
7661 result_val->init_req(_instance_path, size);
7662 }
7663
7664 set_result(result_reg, result_val);
7665 }
7666
7667 return true;
7668 }
7669
7670 //------------------------------- inline_blackhole --------------------------------------
7671 //
7672 // Make sure all arguments to this node are alive.
7673 // This matches methods that were requested to be blackholed through compile commands.
7674 //
7675 bool LibraryCallKit::inline_blackhole() {
7676 assert(callee()->is_static(), "Should have been checked before: only static methods here");
7677 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7678 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7679
7680 // Bind call arguments as blackhole arguments to keep them alive
7681 Node* bh = insert_mem_bar(Op_Blackhole);
7682 uint nargs = callee()->arg_size();
7683 for (uint i = 0; i < nargs; i++) {
7684 bh->add_req(argument(i));
7685 }
7686
7687 return true;
7688 }